If you’ve followed Arctic over the past few years, you probably think of oversized radiators, VRM fans, and thermal domination. The company's Liquid Freezer III Pro 360mm AIO is a favorite of ours, providing top-tier performance while retaining a relatively affordable price tag.

But not every system needs a thick 360mm radiator, and many smaller cases can’t fit radiators at all. And let’s be honest: While having the strongest performance is great if you’re pushing CPU limits with overclocking, more basic cooling options are sufficient for most consumers.

That’s where Arctic’s latest cooler, the Freezer 36-S, comes in. This is an entry-level air cooler available for less than $30, featuring only a single fan and tower connected by four direct-touch heatpipes.

Let's take a look at the specifications and features of the cooler, then we’ll go over thermal and noise benchmarks so you can decide if the Freezer 36-S will make our list of the best CPU coolers.

Cooler specifications

Features of the Arctic Freezer 36S

● Four direct-touch heatpipes

Four direct-touch copper heatpipes, slightly staggered to ensure even heat distribution and transfer heat from the CPU into the tower’s fins.

The backside (shown above) is mostly flat, whereas the fan-touching frontside has a “chainsaw” design to it.

● Anti-bending CPU contact frame for Intel systems

Arctic includes an anti-bending CPU contact frame for Intel systems, which can help ensure consistent thermal performance – it’s especially useful for those who run high-intensity workloads, which subject the chip to more frequent thermal cycles.

However, using one of these frames is not without risk during the installation process. Improper installation can lead to RAM instability and other problems. It's not the best option for novice builders.

● High-speed 3000 RPM P12 Pro fan

There’s more to a cooler than just the heatsink; the included fans have a direct impact on noise levels and cooling performance. Arctic includes its high-performance P12 Pro fan here, which reaches up to 3000RPM and features 77 CFM (cubic feet per minute) airflow rating, with an impressive maximum static pressure of 6.9 mmH₂O – on paper, that's two to three times stronger than most fans on the market.

● Premium MX-7 thermal paste

Arctic includes a small 0.8-gram tube of MX-7, its newest thermal paste. We haven’t tested this paste independently (yet), but Arctic’s MX-6 already performs very well in our thermal paste testing, with performance similar to the most effective competitors on the market.

● Packaging

The heatsink and fan are in a cardboard-only package, reducing waste by forgoing plastic bags.

Included below is a picture showing all of the parts included: mounting bars for AMD systems, a contact frame for Intel systems, screws, the heatsink tower, and a 120mm fan.

● AM5 Installation

Intel systems require the use of a contact frame, but if you’re installing on an AMD Ryzen system, the process is much simpler. You’ll need to first remove the default retention mechanism, and then replace it with the included mounting bars.

The next step is applying the MX-7 thermal paste. If you’re not sure how to do that, we have a thermal paste application guide that covers the different methods you can use.

Place the heatsink tower against the CPU and mounting bars, then use a screwdriver to secure it.

Lastly, attach the 120mm P12 Pro using the fan clips and connect it to your motherboard headers to complete the installation process.

Real-world testing configuration – AMD AM5 platform

We’ve tested coolers with both the Ryzen 9950X3D and its non-V-Cache sibling, the 9950X. There are some differences in how the 9950X and 9950X3D CPUs are impacted by thermal events. While the heat output of the CCDs of AMD’s 9950X3D are relatively balanced, the 9950X I’m using has one CCD that runs much hotter than the other, with a difference of over 10 degrees Celsius in some scenarios, shown below.

We’ve returned to using a 9950X3D for cooler testing, as it has a more balanced heat profile and is almost certainly a more widely adopted CPU.

The benchmark results shared in these reviews may differ from others because I place an emphasis on having results that are comparable to real-world use.

This means that I generally test CPU coolers inside of a closed desktop case, which increases cooling difficulty compared to other testing methods. Some people test coolers on an open test bench, with lowered ambient temperatures. This can make weak coolers appear stronger than they really are. Some publications have also used generic thermal plates to test cooling solutions. I reject both of these methods because they don’t accurately reflect real-world cooler conditions.

Our latest testing setup uses the FLOVA F50 computer case from Tryx.

This case features a unique “crossflow” fan that pulls air from the side, which the company claims is more effective than traditional intake fans. For air cooling tests, we’ve added a single Noctua NF-A12 G2 intake fan.

PBO Performance and maximum noise levels

We’re going to start this benchmark section by focusing on a traditional maximum performance test, with the CPU cooler’s fans allowed to reach their fastest speeds, for the best cooling possible.

Turning on PBO allows AMD’s Ryzen 9 9950X3D to stretch its legs to an extent, and all air coolers I have tested with PBO enabled using MSI’s X870E Carbon Wifi motherboard reach the maximum CPU temperature of 95 degrees C (203 F) and thermally throttle, to a small extent.

Most of the coolers in this comparison are dual-tower, with two fans. As such, it isn’t a surprise that the single-tower Freezer 36S is outperformed by most of the competition here. What is interesting is that despite being only a single-tower, single-fan air cooler, it still manages to perform very closely to the larger heatsinks – falling only 6W behind DeepCool’s AK620 G2.

To give a wider variety of comparison examples, I’ve included some data of the same tests from our last AMD Ryzen 9 9950X3D cooling test bench. But keep in mind these results aren’t 100% comparable, due to a different case and fans being used.

Arctic’s Freezer 36S performs rather well for a single-tower cooler, and part of the reason for that is the included fan – which features speeds of up to 3000RPM, about 50% faster than most fans included on consumer CPU coolers. These faster top speeds come at the cost of noise levels, though, with the 36S reaching 47.4 dBA.

200W thermal benchmarks

For the next thermal test, I leave the motherboard settings at their defaults, which results in a power limit of about 200W when running Cinebench R23.

With a “stock” power profile enabled, Arctic’s latest single-tower air cooler averages 54.3 ΔC (76.5C/169.7F) – less than 2 degrees C from the best result we have from dual-tower air coolers.

150W + GPU thermal results, noise levels

Our next test runs Cinebench on the CPU with a 150W power limit, while also running Furmark on MSI’s RTX 4070 Ti Super Ventus 3x OC. This causes the GPU to consume ~295W of power. This test is designed to emulate the thermals of games, which primarily stress the CPU and GPU.

All of the air coolers here perform similarly, with a range of just 2.7 degrees C between the best dual-tower air cooler and the worst-performing result. Let’s take a look at the noise levels now, as that will be more important to most users in this scenario.

I measured 42.4 dBA from Arctic’s Freezer 36S with a Pyle PSP25 soundmeter. This isn’t exactly what I would call “loud,” but it certainly ain’t quiet, either. If you’re particular about noise levels, you might want to implement a custom fan curve.

Noise-normalized testing

Most testing is performed with the cooler tied to the default fan curve of our MSI X870E Carbon motherboard, but some of y’all prefer to see tests when the noise levels of coolers are equalized. This is especially important to those of you who prefer silent computers.

This next test has the CPU cooler noise-normalized to 38.9 dBA, with PBO enabled for the Ryzen 9 9950X3D CPU. Our results have it near the bottom of the list, which makes sense considering that it’s only a single-tower, single-fan cooler.

With this review, we have another noise-normalized benchmark, measuring the amount of time the cooler is able to keep AMD’s Ryzen 9 9950X3D under its peak temperature (TJ Max) of 95 degrees C while running Cinebench R23 multi-core, all while the fan is noise-normalized to 38.9 dBA.

Karhu DDR5 RAM thermals testing!

Your CPU cooler does not operate in isolation. It has an impact on not just your CPU’s temperatures, but also the other components in your build, like your RAM and GPU. To that end, I’ve run the Karhu RAM stress test. This places a load of ~153W on the CPU and ensures system RAM (DDR5 in my case) is fully stable. In this type of scenario, most AIOs tend to produce worse results than air coolers.

The results here are extremely interesting. I had expected DDR5 temperatures to be worse than dual-tower air coolers, but both Arctic’s Freezer 36S and DeepCool’s AK400 G2 single-tower air coolers performed better than all traditional results – being outperformed only by Silverstone’s IceMyst Pro 360 when equipped with VRM fans.

Conclusion

Arctic’s Freezer 36S punches above its weight and price class, offering performance comparable to many dual-tower air coolers with only one fan paired to a single-tower heatsink – all with a price starting at just $28.49. It’s not the quietest cooler, and it requires installing a contact frame for Intel systems (which makes it a poor fit for novice builders), but if you aren’t trying to overclock and you don’t want to overspend on cooling to get something a bit quieter, it’s worth considering.

Arctic has just released a new mini-PC aimed at enthusiasts looking for a clean, minimalist, and silent setup, without compromising on performance. The "Senza AI 370" is the latest in the company's existing Senza lineup that has shipped with the Ryzen 5000 series so far, but it's now been upgraded to AMD's Strix Point. Specifically, it features the Ryzen AI 9 HX 370 flagship APU and a neat trick up its sleeve.

Or should we say "below its sleeve"? Because this mini-PC mounts under your desk, completely out of sight, cable management isn't an issue either. Arctic calls it an "Under Desk PC," and it's designed to clear your workspace. You could put it on top of your desk, but that would defeat its entire purpose. You really want to make sure you get the most out of it because it costs 1200 Euros, or almost $1,400.

For that money, you're getting a top-end chip, the Ryzen AI 370, which has 12 cores and 24 threads that can boost up to 5.1 GHz. It launched almost two years ago, but it was the Red Team's premier laptop offering until Gorgon Point did so a few months ago. It's based on AMD's Zen 5 architecture, featuring 4 standard Zen 5 cores and 8 Zen 5c efficiency cores — both are split off into their own CCX.

Testing has shown that this hybrid core config doesn't suit either performance or efficiency, since any tasks requiring more cores spill over to the Zen 5c CCX, and there's a large latency penalty for cross-CCX communication. But that's mostly a concern for mobile devices (since this is a mobile chip), a wall-powered mini-PC circumvents these issues and gives you solid performance regardless.

The Senza AI 370 looks like an AIO radiator for the most part, down to the two tubes poking out of it that terminate in a front panel hub. This detachable box mounts near the edge of your desk, where you can easily interact with it, as it includes a power button, a couple of USB ports, and a headphone jack. The actual PC is meant to go farther back, toward the rear of your table, where it almost disappears.

The port selection on the PC itself is very strong: you get 2.5 Gbps Ethernet, USB4, HDMI 2.1, and more, plus Wi-Fi 7 for wireless connectivity. Owing to its design, the entire body is essentially a giant heatsink, with the actual PCB only in the middle, occupying about 30% of the chassis. Eight heatpipes run through this heatsink, cooling the APU without any fans, which allows it to be completely silent.

Despite nonexistent noise levels, Arctic is claiming great thermals — 50 degrees Celsius under a gaming load, while a "competitor mini PC" reached 67 degrees Celsius. The company also says the Senza reaches 98 FPS in Counter-Strike 2 (at 1080p High), while a similarly equipped mini PC only reaches 86 FPS. This is possible due to the higher-spec RAM configuration on the Senza.

It features 32 GB of LPDDR5X-8000 memory, while many other devices use standard DDR5 at slower speeds. The Ryzen AI 370 also has a Radeon 890M iGPU, which is fairly performant for integrated graphics. It's no Strix Halo, but it will get the job done in lightweight games, and can produce decent results in AAA titles with upscaling. It's got 16 RDNA 3.5 Compute Units and can boost up to 2.9 GHz.

To round out the device, other specs include a 1 TB PCIe 4.0 SSD, support for up to 8K external monitors, pre-installed Windows 11, and, of course, Microsoft Copilot+ branding for AI workloads. You can buy the Senza AI 370 right away from Arctic's website, and while it does cost 1,200 Euros, we noticed that a 170 Euros payback is automatically applied, bringing the price down to 1,029 Euros ($1,192). That's still quite a lot for this performance tier.

Cooling a computer isn’t terribly complicated in and of itself. But if you want to do it right, selecting the best cooling fan can be a daunting task. It’s easy to throw money at Noctua, and we wouldn’t fault you for doing so, but that strategy gets very expensive very quickly. And not everyone digs the beige-and-brown theme – and the company flat-out refuses to do RGB.

Of course, you can opt to go the other easy route and just use what’s included with the case and cooler you bought. But that’s also rarely the optimal solution, and many premium cases come with no fans or just one for exhaust. Ideally, you want a fan that’s quiet, effective at moving air, attractive, and affordable. And since you’ve landed on this page, chances are you bought one of the cases that doesn’t come with enough fans, or you’re not happy with the included fans' performance or audio profile, and are looking for something better.

Recently, we’ve picked up fan testing again, and we’re working hard to test a bunch of different models to figure out the best options in this very crowded field. Over the past several months, we’ve worked in cooperation with industry experts to flesh out a benchmarking methodology, and we expect to update this page regularly with new models we’ve tested.

Our Testing methodology is built on real-world applications, and there are a handful of scenarios a fan needs to perform well in. Naturally, we’re testing airflow in an unrestricted environment as a ‘vanilla’ test, but this is often not indicative of performance inside a PC case.

What’s also not indicative of real-world performance is static pressure testing. Yes, it’s an industry standard, but it really only measures how much pressure a fan can generate when blowing into a closed tube, or up against a wall in a more real-world scenario. It does not account for the changes in acoustics when a fan is mounted on a radiator, heatsink, or behind a mesh intake.

Instead, we’ve decided to measure a fan’s acoustic performance not only when unrestricted and suspended mid-air, but also when strapped to a radiator and again when behind a mesh intake. The results from this paint a much more nuanced, real-world picture, often with interesting outcomes.

For more information about how we test PC fans, scroll to the bottom of this page.

Quick List

Best PC fans you can buy today

Best silence optimized fans

Noctua’s NF-A12x25 G2 is game-changing, offering the quietest, most refined fan we’ve ever seen. In our testing, it delivered whisper-quiet performance and managed an impressively huge amount of airflow despite running at very low RPMs.

Managing such big airflow despite not needing to spin fast to accomplish it means it’s also not a very turbulent fan, which translates to chart-topping performance when installed on a radiator or behind an intake mesh. In fact, it’s the only fan we’ve tested thus far that ran quieter on a radiator than when suspended mid-air without any load.

In addition to pulling off airflow miracles, the NF-A12x25 also offers a seriously advanced internal hub, featuring Noctua’s refined, tried-and-tested SSO2 bearing, a new 3-phase etaPERF motor, and a new PWM controller with SupraTorque functionality. This sounds like a bunch of marketing talk, but after testing the fan, we can tell you with confidence that it has zero audible mechanical noise.

Best budget silence optimized fans

The Arctic P12 Pro is an interesting fan, because it offers performance that is, as far as the charts go, nearly neck-and-neck with Noctua’s NF-A12x25. Except, the Arctic does something Noctua doesn’t: It only costs $8.50 for one fan.

The price you pay for the Arctic fan is, in part at least, in its level of refinement. It uses a plastic ring to strengthen the impeller. And, in order to achieve its airflow, it relies on spinning up to higher speeds. Consequently, it’s a more turbulent fan than the Noctua, so it takes a bit more of a hit when mounted close to a mesh intake. It also produces a few mechanical noises, but even so, it’s a highly capable fan.

If, however, you just want a fan to move lots of air quietly, and you don’t want it to cost much, the Arctic P12 Pro is in a league of its own.

Best RGB fans

It feels a bit weird to place essentially the same Arctic fan on the Best list twice, but it truly is that good. The P12 Pro RGB is, for all intents and purposes, the exact same fan as the non-RGB variant, except with a white translucent impeller so that the RGB lighting from the hub shines through.

It performs on par with the non-RGB variant in our testing, offering wicked performance at up to 3000 RPM and a fantastic noise-to-airflow ratio. The best part: it barely costs more than the non-RGB variant.

Best High-Performance fans

The Phanteks T30 (review here) is one of the fans that somewhat copied the NF-A12x25 – it featured a similar blade shape, and also used a LCP (liquid crystal polymer) plastic for the impeller. Except, it wasn’t quite a copy, and now, even half a decade later, it’s still king of the hill in the ultra-high-performance fan space.

Why is it so good? Phanteks cheated. I wish I could say they didn’t, but they did. The industry standard for a 120mm fan is 25mm thickness. So what did Phanteks do? They made a 30mm fan. In fact, Phanteks was so proud of this fan that when they sent it out for review, they put together a wind tunnel kit for us because they wanted everybody to know Phanteks meant business and had defeated Noctua at their own game. And they did – it performed notably better than Noctua’s finest when tested for noise-to-airflow.

Moreover, the good stuff doesn’t stop at just 5mm of extra thickness. The company also built in a mode selector at the back of the hub that lets you select quiet, regular, and advanced modes. If you ask me, though, advanced should be called Final Destination mode it lets spin all the way up to 3,000 RPM. With its added thickness, high speed, and LCP material, it’ll gobble air and your fingers if you’re not careful.

Choosing the best fan for you

Fans. They spin. They move air. They make noise. Sometimes they have lights, and more recently, some have LCD screens. It may seem like a silly thing to obsess over, but picking the right, or wrong, fan can make a major difference in a PC build.

Not everybody looks for the same thing in a fan, and whereas picking the Best CPU or Best GPU is often a matter of what’s best-performing for the dollar in your allocated budget, fans don’t evolve at nearly the same pace, so picking one is much more of a subjective exercise. With that said, depending on your situation, there can be a few key considerations that can make the choice easier.

• If you’re someone who comes home from work, whips on a headset, and plays games for an hour, noise levels or MTBF (durability) figures may not matter much to you.
• Alternatively, if you’re someone who works on your PC in silence all day long, it may be worth investing in more refined fans that offer less wind and mechanical noise, with a more durable motor.
• Personal sensitivity to certain noise types and frequencies also plays a role.
• Whether the PC is placed on the desk or on the floor is another factor to consider. Fan noise can be less noticeable from the floor.
• Some fans are designed to perform better on radiators and heatsinks, whereas others perform better as case intake or exhaust. Pick accordingly.
• Fans with daisy-chaining connectors can ease installation and provide tidier cable management.
• Cases with tinted windows require fans with brighter RGB for a visual spectacle.
• Mini-ITX PCs, though fantastically small, can benefit more from a fan upgrade to quiet them down than big full-tower E-ATX machines.

More about how we test PC fans

Our PC fan testing consists of multiple phases. First, we test a fan’s vanilla acoustics by suspending it mid-air with four pieces of string so that no frames or holders interfere with the airflow. We start with a basic measurement, noting whether the fan breaks through the noise floor at its minimum speed, then we record the RPM for every whole dB between 34 and 44 dB(A). Finally, we record the maximum RPM and its associated noise level.

We then repeat these tests with a radiator mounted behind the fan, and again with a mesh intake in front of the fan. We do this because radiators and intake mesh fundamentally change a fan’s acoustics and the RPM they must spin at to achieve the desired noise targets.

After the acoustic data, we mount the fan in our wind tunnel to collect airflow data.

The windspeed is measured with an anemometer. However, although the fan’s impellers may push air, no air comes from the space behind the motor hub, and fans don’t generally push air in only one direction. In fact, the space immediately behind a motor hub usually becomes a low-pressure pocket, which is why we need a wind tunnel – to form the airstream into one cohesive, non-turbulent column of moving air.

Inside the wind tunnel (at the halfway point and near the end) are two air-straighteners, which together with the length of the tunnel at 77cm (just over 30 inches) neatly combs the airstream into something we can measure, without creating tons of backpressure that would affect the readings.

Using the RPM datapoints from the acoustic testing phase, we now set the fan to each RPM point and record the associated wind speed. We repeat this with a radiator behind the fan, and again without the radiator but with the mesh intake in front of the fan. Of course, the wind speed isn’t actually a useful measure (yet) of how much air a fan can displace, until we convert it to meters cubed per hour (m³/h).This testing yields up to 84 datapoints per fan (assuming the fan can achieve each point), which tells us enough about a fan's capabilities, quirks, and acoustics to judge whether it's one of the best PC fans you can buy.

Not long ago, we tested two popular case fans, pitting Noctua’s NF-A12x25 G2, an industry legend, against the also well-liked Arctic P12 Pro. We had some surprising results, as the Arctic fan almost kept up with the much pricier Noctua. However, Arctic also sent us some P12 Pro’s of the RGB variety, so we thought it would be fitting to put Artic’s colorful spinner up against some competitors in its own price category.

In this round of testing we’ve rounded up five 120mm RGB fans that don’t cost a whole lot, so here, you won’t be seeing Lian Li’s fancy LCD-infused, wireless, infinity-mirror fans. Instead, we’re pinning the P12 Pro up against Be Quiet’s Light Wings LX, Fractal Design’s Aspect 12 RGB, Phanteks’ M25 G2 RGB, and finally, Levelplay’s Combat Fan 120 A-RGB. This last one was sent to us as a surprise, and I thought it was fitting to include in the roundup.

Let’s dive straight into the testing.

Testing Methodology

For evaluating PC fans, there are two main terms you should be aware of: Airflow and Static Pressure.

Airflow is easy – how much air does a fan move (through a wind tunnel in our tests) when not burdened by a radiator, mesh, or any other restriction.

Static pressure can be measured by placing the fan on a wind tunnel, closing the end, and measuring the pressure generated with a sensor, which spits out a value in millimeters of water. The higher the number, the better a fan is likely to be at overcoming restrictions.

However, I find that a fan’s measure of static pressure isn’t actually all that useful – all it tells you, really, is how well a fan can push air up against a wall, which is obviously not how you want to position the back of your PC. Instead, I prefer to test for noise and airflow through a radiator and intake mesh, as well as the standard suite of unrestricted tests for a more nuanced, real-world picture.

Testing Results of 5 Budget RGB Fans

(Not) Breaking the Noise Floor

On the day of this testing, the noise floor in my office sat between 32.4 and 32.7 dBA, and I am happy to be able to write that all five fans, when run at their minimum speed, were completely inaudible and did not break the noise floor, even with the dB meter placed at just 20 cm from their hubs.

This used to be something I only saw in premium fans, where manufacturers have designed quiet hubs, so I was delighted to see this kind of performance from budget-friendly RGB fans.

Airflow is not recorded for these tests, as none offer enough air speed for the anemometer to register.

dBA VS. Airflow: Unrestricted

The first series of tests I subject the fans to is measuring their acoustics in an unrestricted environment. I target every whole dB between 34 and 44 dB(A), and record the associated RPM. Then, I put that same fan on our wind tunnel, and record the windspeed at said RPM figures. This is then calculated over to cubic meters per hour for airflow.

In this test, Arctic’s P12 Pro leads the pack by a significant margin, followed by Be Quiet’s Light Wings LX. Phanteks’ M25 G2 RGB only performs a smidge worse than Be quiet’s fan.

Note the dip in the curve of Fractal Design’s Aspect 12 RGB at 37 dB(A). This fan creates quite an annoying resonance at certain speeds, where it gets louder at a certain RPM, and then quiets down again once past this RPM. I suspect that if I were to test at even smaller increments, the graph would become quite wobbly.

Most fans exhibit this kind of behavior to some extent, where they create unwanted noises at certain RPMs – even Arctic’s fan does this, but the level to which the Aspect 12 RGB made this noise was really quite bothersome.

Similarly, for the Phanteks’ fan (while its curve looks alright), I was unable to test at 35 dBA, which is why its line starts at 36. No matter how much I fidgeted with its speed, I simply could not get the dB meter to hit 35 – it was either over, or under, but not on.

The Levelplay Combat fan had the same issue, but at 41 dB(A), where I just couldn’t hit the mark.

All of this is not to say that Arctic’s fan is without imperfections either, though. This sample exhibited a light high pitch at certain RPMs which didn’t affect the readings, but I did notice, and at speed it started wobbling – a sign that it is unbalanced.

The predicament: Whether to test with more fans

I’ve wondered throughout testing these fans whether I should test with more fans and take the average results, but I have two problems with this: firstly, testing this thoroughly already takes quite some time, and more importantly, averaging out the results would mask defects.

In a premium fan from Noctua, we expect near-perfection. However, these are RGB fans that cost no more than $15 each. There will be some variance in issues between samples, but not averaging out the results means you get to see the fan’s performance for what it really is with one specific fan. Some of these behaviors will be down to their design and consistent across samples, while other issues may differ between units.

I do request three samples for each fan to review, however. If I do run into any issue like those mentioned above, I grab the other samples and quickly run them from lowest to max RPM and judge the results. If they exhibit similar behavior, I roll with the data, but if they to suddenly sound much cleaner, then I know there’s a manufacturing defect at play.

dBA VS. Airflow: Radiator Restricted

Pin these fans on a radiator, and something very interesting happens. Four of the fans perform nearly the same as each other, with the Arctic P12 Pro standing out by a huge margin.

Note that for this test, the acoustics are measured separately from the acoustics without the radiator, and the fans follow similar noise-to-RPM patterns.

dBA VS. Airflow: Mesh + Filter Intake Restricted

When placing the fans behind a mesh intake, things again get interesting. This is the most challenging test of the three, as placing intake mesh in front of a fan fundamentally changes the way they sound – intakes punish airflow more than radiators, not because they’re more restrictive, but because placing something in front of a fan has a much more profound effect on the noise than something behind it, and this is where a fan’s blade design really starts to shine.

The Arctic P12 Pro continues to lead the race here, with Be Qquiet’s Light Wings LX following behind. The Combat fan, despite being from a lesser-known brand, does OK for itself.

However, Fractal Design’s Aspect 12 RGB and Phanteks’ M25 G2 put down appalling performances on this test, offering only 60% of the P12 Pro’s airflow at the same noise levels.

Maximum performance

Finally, the fans get a test at full blast.

This isn’t a useful test to compare noise-to-performance efficiency of the fans, but if you’re looking for a fan with low maximum speed to keep noise levels down without having to make a custom fan curve, Levelplay’s Combat Fan is your best pick. On the other hand, if you’re looking for maximum peak performance, Arctic’s P12 Pro RGB is a beast in a league of its own that can spin at well above 3000 RPM.

The dBA figure doesn’t say everything

Just because a fan performs well on the charts doesn’t necessarily mean it may be pleasant to listen to. One example of this is the Combat Fan from Levelplay. It’s not a terrible performer, but at certain speeds, something starts humming at an annoying mid-frequency. It was one of those frequencies that echoes through the room, and if I weren’t testing it but had it in my PC, it would drive me mad, wondering what the source of the noise in the room is.

The Fractal Design Aspect 12 RGB also suffers from a painfully dirty noise profile that did show up on the charts, as it limited the fan’s ability to reach the speeds it otherwise would have had it not made these noises. It sounds a bit raspy, almost like there’s grit in the motor, and it had similar resonant noises as the Combat Fan.

The M25 G2 didn’t have weird noises – it was just inefficient in its noise-to-airflow ratio.

Judging the RGB, feature set and pricing

Fractal Design Aspect 12 RGB

Fractal Design’s Aspect 12 RGB fans have the worst RGB of the lot. The blades are almost opaque, which makes the RGB on anything but the hub rather dim compared to the competition. It also does not have a zero-RPM mode – when fed a 0% PWM duty, it will still spin, and with a minimum speed of about 500 RPM; I consider this a problem. Yes, it’s quiet at its minimum speed, so that’s not an issue, but high minimum speeds combined with no Zero-RPM implementation can lead to excessive dust buildup, as the constantly whirring blades accomplish nothing but to pull small particles towards your filter and / or fan.

This fan is available for $51.47 as a 3-pack, which brings the per-unit cost to $17.16.

Combat Fan 120 A-RGB

Levelplay’s Combat Fan 120 A-RGB comes with a ring around its blades, which makes for a cool look. Despite the RGB coming from the hub, it still lights up the ring moderately. It also responds correctly to a Zero-RPM signal.

Pricing for the Levelplay Combat fan sits at $12.23 per unit.

Be Quiet! Light Wings LX

Be Quiet’s Light Wings LX fans feature good RGB and a slick blade design, but no Zero-RPM feature. That being said, the minimum RPM is extremely low, tested at just 213 RPM, so I can forgive the lack of a zero-RPM mode.

Pricing sits at $47.90 for a 3-pack, bringing the per-unit price to $15.97.

Phanteks M25 G2 RGB

Phanteks M25 G2 RGB, although not being the quietest fan, offers a properly complete feature set that makes a compelling case for forgiving its performance shortcomings. The RGB is gorgeous and it comes with an infinity-mirror hub. The fans can also clip on to each other, daisy-chaining in a not-quite-wireless fashion but certainly better than the competition. The fan even comes with plastic covers for the screw holes for a tidy appearance. It also responds correctly to a 0% PWM duty signal.

Pricing sits at $36.99 for a 3-pack, which is a surprisingly good price if you consider the amount of effort that went into their design. It may not be the quietest fan, but it sure is the prettiest and comes with most bits and pieces to complete the set.

Arctic P12 Pro RGB

Arctic’s P12 Pro lights up bright and uniform, offering a fantastic visual spectacle. It also comes with a proper Zero-RPM implementation, but that’s where its feature set ends.

This fan costs just $28.49 in a 3-pack, which is a mighty good deal that’s hard to pass up at $9.50 per fan.

Picking a winner

It would come as no surprise if I said that Arctic’s P12 Pro was the winner of this test, but I’m not sure about that. The P12 Pro RGB is clearly the best performer, but it was a tad unbalanced and had some light high-pithed whining. It also clicked a few times upon startup, but most of the fans did that. Mechanical noises.

If you’re looking for the best value offering, the Arctic P12 Pro RGB is hard to pass up with the 3-pack costing just $9.50 per fan.

The Light Wings LX, although being the second-place fan in testing for noise levels, is my favorite of the bunch. Why? It’s clean! It makes a crisp, clean sound ---just a whoosh, with no extraneous noises on top. It’s a slightly louder whoosh, but a clean whoosh. Of course, that’s not to say it may be your favorite. Personally, I’m sensitive to these kinds of noises, and I’m willing to pay a little extra for a clean-sounding PC, but that little bit extra is quite a bit if you tally it up over multiple spinners.

Costing $15.97, this puts the be quiet! Light Wings LX right on the edge between budget and mid-range RGB fans.

If, on the other hand, you’re someone who doesn’t mind noise too much, and wants to build something that you’re truly proud to show off, Phanteks’ M25 G2 fans do away with the focus on silence, and prioritize looks with bright RGB, an infinity mirror, seamless daisy-chaining, and even a mechanism that clips the spinners into one unit and covers for the screw holes.

I’m someone that works many hours a day in silence, with my open air PC on my desk near me, so I’ll prioritize silence any day of the week. But if you’re a gamer, keep the PC on the ground, or wear a headset all the time anyway, the quality of the RGB may be far more important.

If picking a winner is so difficult, should we pick a loser?

If there’s one clear conclusion I can draw from this batch of testing, it’s that Fractal Design’s Aspect 12 RGB is on my do-not-buy list. It offers no real redeeming features to make up for its shortcomings – it produces a dirty, inconsistent noise, performs disappointingly compared to the competition, has no Zero-RPM mode, and even the RGB doesn’t look all that great. It doesn’t even come with vibration-damping rubber pieces on the corners.

I honestly don’t understand what it’s doing in Fractal Design’s product portfolio at all, other than making the company’s other fans look good. And the wildest thing is, it’s also the most expensive fan tested here, at $17.16 a piece.

In this respect, fans are much like tires for cars: they come in all shapes and sizes, each with different purposes, from different manufacturers, and they all seem to do the same thing. And much like tires, in the world of fans there are a few great ones, many good ones, and some to avoid entirely.

Arctic is a name well-known by many PC enthusiasts for often providing high-value, quality cooling solutions. Arctic’s MX-6 is one of the best thermal pastes on the market, and the Liquid Freezer III Pro 360 showed some of the best liquid cooling results we’ve seen when tested on AMD’s Ryzen 9 9950X3D and Intel’s i7-14700K.

Since the release of the Liquid Freezer III Pro, many loyal readers have been requesting that we test the 420mm version of this powerful cooler. And so we’ve tested it with AMD’s Ryzen 9 9950X.

Not to get ahead of our testing, but this cooler will certainly be a strong contender for a spot on our Best AIO Coolers list. It offers industry-leading cooling performance at a very reasonable price. But before we get to testing, let’s look at the cooler’s specs and what’s included in the box.

Cooler specifications

Packing and included contents

Arctic’s Liquid Freezer III Pro 420 arrives in a triangle-shaped box, a very unusual design decision, which certainly makes it stand out. It also certainly makes it harder to ship, but that shouldn’t be an issue for potential buyers.

Included with the product are the following:

• 420mm AIO with pre-installed fans
• MX6 thermal paste
• Mounting for AMD and Intel platforms
• All-in-one or separated PWM cords
• VRM fan

Features of Arctic’s Liquid Freezer III Pro 420

▶️ 38mm thick radiator

This AIO’s radiator is thick. At 38mm for the radiator and 27mm for the fans, that’s a combined thickness of 65mm. This will present compatibility issues with many cases, so make sure to do some measuring before buying. You probably aren’t considering a compact case with a 420mm cooler. Even with our MSI MAG Pano 100R PZ test case, we had to mount the radiator above the top frame, as it doesn’t technically support coolers this large. More on that below in our testing section.

▶️ Anti-bending frame for Intel systems

Intel LGA 1700/1800 CPUs, especially Raptor Lake, have been known to warp or bend in certain conditions. And as a result, the CPU can become harder to cool. This issue can be avoided altogether with the use of a contact frame, which Arctic includes for users of Intel systems. It’s a nice inclusion, but it also complicates the installation process and is a requirement for mounting Artic’s current-gen AIOs on Intel-based systems.

▶️ Low price

Officially, the LF3 Pro 420 carries a MSRP of $148.99, which is a reasonable price in and of itself. However, this AIO is frequently on sale, making it an even better value. At the time of this writing, it is available for only $103 on Amazon.

▶️ MX-6 thermal paste

Arctic includes a small tube of its MX-6 thermal paste. We’ve tested it on both AMD and Intel systems, and it’s one of the best on the market! Check out our thermal paste benchmarks for more detailed information.

▶️ Improved CPU mounting

Different CPUs currently on the market can have different hotspots, which can impact the effectiveness of a cooler. This is why some coolers seem to do better – or worse – with AMD or Intel. Arctic refined its mounting for AMD AM4/AM5 and LGA 1851 systems, resulting in claimed performance increases of up to 9 degrees Celsius with a 250W power limit. We’ll, of course, see how that shakes out shortly, in our testing.

▶️ New P14 Pro Fans – and a VRM fan too!

There’s more to a cooler than just the heatsink or radiator. The bundled fans have a significant impact on cooling and noise levels, as well as how the cooler looks in your case.

The new P12 Pro fans include seven blades instead of the five-bladed fans included with previous Liquid Freezer CPUs, and are pre-installed for convenience. Arctic claims they perform at the same level as its MAX fans, but at a lower noise level. That’s not to say these fans run quietlym though. As you’ll see in our benchmarks below, they can get pretty loud.

A quick connect system for the PWM and ARGB connections is utilized to minimize cable clutter.

AM5 installation

The installation of this cooler is simple, but a bit of a pain for Intel users, due to the required use of a custom contact frame for the CPU. This guide will go over the installation on AMD systems. Intel users can check out our review of the smaller Arctic Liquid Freezer III Pro for installation help.

1. Remove the default AM4/AM5 mounting mechanism.

2. Place the rubber standoffs around the screw holes surrounding the CPU.

3. Take the included mounting bars and place them on top of the standoffs, and then secrure them with a screwdriver and the included screws.

4. Apply the included Arctic MX-6 thermal paste to your CPU. If you aren’t sure how to do this properly, please refer to our handy guide on how to apply thermal paste.

5. Press the pump block on top of the CPU, and secure the built-in screws using a screwdriver.

6. Once complete, connect the cabling to the CPU block’s head. For the purposes of testing, I chose the option with individual PWM control for the fans, pump, and VRM fan. However, some may prefer the simplicity of the all-in-one cable. Connect the cables, place the VRM fan on top of the CPU block, and installation is complete.

Real world testing configuration –AMD AM5 platform

The 9950X3D setup that we started testing coolers with last year has suffered an unfortunate death. As such, we’ve begun to test coolers on a system with AMD’s 9950X.

There are some differences in how the 9950X and 9950X3D CPUs are impacted by thermal events: while the heat output of the CCDs of AMD’s 9950X3D are relatively balanced, the 9950X I’m using has one CCD that runs much hotter than the other, with a difference of over 10 degrees C in some scenarios, shown below.

The benchmark results shared in these reviews may differ from others because I emphasize results that are comparable to real-world use.

This means that I generally test CPU coolers inside of a closed desktop case, which increases cooling difficulty compared to other testing methods. Open benches have lowered ambient temperatures, which in turn makes weak coolers appear stronger than they really are. Some have used generic thermal plates to test cooling solutions. I reject both of these methods, because they don’t accurately reflect the real-world conditions where a CPU cooler is used.

That said, this review bends this rule a little: The MSI Pano 100R PZ case that we use for cooler testing doesn’t technically support 420mm radiators. So I had to place it on top of the case to make it fit. This likely affects thermals and acoustics at least somewhat, but short of re-testing all our coolers in a different case, we didn’t have a better option. We tested Be Quiet’s competing Silent Loop 3 420 in this same way for comparison, while the smaller 360mm coolers were tested in a typical fashion, with the radiator and fans inside the case. If there is heavy interest in large 420mm cooler testing, we will revisit these results in a larger case later this year.

Noise-normalized testing

Most testing is performed with the cooler tied to the default fan curve of the MSI x870E Carbon motherboard, but many of y’all prefer to see tests when the noise levels of coolers are equalized. This is especially important to those of you who prefer silent computers.

We’ll start with the “easiest” of these tests, which runs Cinebench R23 with a stock power limit. On this motherboard, that means the CPU will consume roughly 200W. The Liquid Freezer III Pro 420 achieved the best result we’ve seen yet on this system, with a temperature of only 73.6C – outperforming all of the 360mm AIOs we’ve tested here, and even beating BeQuiet’s Silent Loop 3 420 by 1.4 degrees C.

The next test is a bit harder, with PBO enabled to allow the CPU to use as much power as can be handled, and the cooler handled a whopping 262.3W, on average.

For our last test, we’ve added the heat of a GPU to the mix, which contributes 295W of heat into the PC case. The result of 249.3W is very impressive, outperforming the best 360mm result, with an increased thermal capacity of 16.5W.

Noise levels

We’ve measured noise levels at two points to give you an idea of how the cooler operates when tied to a motherboard’s default fan curve. Our first measurement is when the fans are allowed to run at full speed, their loudest noise levels possible.

In terms of maximum volume, the Liquid Freezer III Pro 420 is quite noisy, at 52.7 dBA. This is the second loudest cooler we’ve tested in the past few years.

Our next acoustic measurement is the cooler’s volume when the CPU consumes 150W. This might be similar to your “worst case” gaming scenarios, especially if PBO is enabled. At 42.4 dBA, it isn’t too loud – but it ain’t quiet by PC standards, either!

PBO Performance – full fan speeds

While some enthusiasts prefer to run coolers at low noise levels, I know a lot of you just want maximum performance. This benchmark lets the fans run at their full speeds to test the full potential of the AIO's cooling potential. Turning on PBO allows AMD’s Ryzen 9 9950X to stretch its legs and guzzle power, pushing the limits of any CPU cooler on the market.

Arctic’s Liquid Freezer III Pro 420 stayed true to its name, cooling over 268W here.

200W thermal benchmarks and noise levels

For the next thermal test, I leave the motherboard settings at their defaults, which results in a power limit of 200W when running Cinebench R23. The performance gap between 420mm and 360mm AIOs shrinks at this power level. While the Liquid Freezer III Pro 420’s result of 73.6 degrees C is the best we have seen thus far, it is only 0.9 degrees lower than the best 360mm AIO we’ve tested.

150W thermal benchmarks and noise levels

For the next thermal test, I’ve set the power limit to 150W.

The story for our 150W results is about the same as our 200W results. The LF3 Pro 420 takes the lead with a relatively cool temperature of only 58.2 degrees C, maintaining a 0.9 degree lead over the best 360mm result we have in this chart.

100W thermal results

Our next test is the “easiest” benchmark featured here, with a workload consuming only 100W. This is a simple test, one that even most mainstream coolers should have no issue dealing with. Results, as expected, are excellent once again, with the average temperature of the CPU measuring 44.6 degrees C, an improvement of 3 degrees CC compared to the best 360mm AIOs we’ve tested.

Conclusion

The 360mm version of Arctic’s Liquid Freezer III Pro is among the best AIOs on the market, and the newer 420mm variant only pushes the bar for cooling performance further, delivering the best temperatures we’ve seen from any AIO when paired with AMD’s Ryzen 9 9950X CPU. If you have the need for this level of cooling and your case has the space for it, the Artic Liquid Freezer III 420 is an excellent cooler and an excellent value – especially if you can find it on sale significantly below its $149 MSRP.

Arctic has a well-earned reputation for its thermal pastes, with its MX-6 being one of the best thermal pastes on the market. The company has introduced the MX-7 to stay ahead of the competition. It claims the MX-7 is the company's coolest thermal paste to date, with prices ranging from $6-10.

Having launched three years ago, the MX-6 still feels quite new, but in a competitive market where new products come out every year, staying relevant is key. The MX-7 features a new formula that Arctic understandably keeps under wraps. Still, the company shares that the MX-7 boasts low adhesion for even distribution, high viscosity for flow resistance, and high cohesion for more durable bonding.

• Arctic MX-7 2 grams: $6.89
• Arctic MX-7 4 grams: $7.39
• Arctic MX-7 4 grams with six MX-Cleaner wipes: $9.59
• Arctic MX-7 8 grams: $8.59

Arctic takes a different approach from other thermal paste brands by not emphasizing thermal conductivity values for its products. As detailed in the specification sheet, the MX-7 is up to 22% less viscous than the MX-6 and approximately 12% denser than its predecessor. In Arctic's own testing, using a Core Ultra 9 285K at 4.5 GHz and drawing around 284W in an ambient temperature of 25 degrees Celsius, the MX-7 managed to deliver temperatures about 3% cooler than the MX-6 and up to 6% cooler than the MX-4.

Arctic MX-7 Specifications

Rest assured, the thermal paste is completely safe to use with direct-die cooling solutions because it's not electrically conductive or capacitive. This means you don't need to worry about any risk of short circuits or discharges that could damage your devices.

While the MX-7 is less thick than the MX-6, it still has a nice, viscous consistency. Arctic suggests avoiding spreading the paste with a spatula. Instead, applying it in a simple cross pattern is best, and let your CPU cooler do the magic. The pressure from the cooler will evenly spread the MX-7, filling in any gaps and giving you complete coverage. This helpful tip also works well for coolers with direct-touch heat pipes.

Since the MX-7 is non-capacitive and non-conductive, it's simple to wipe away excess or old applications with ease. Arctic also offers a separate MX Cleaner wipe that's perfect for cleaning thermal paste. Plus, these wipes can be purchased together as an MX-7 bundle for added convenience.

You can easily find the MX-7 for sale on Amazon. The latest thermal paste isn't much pricier than the previous version. To give you an idea, the 2-gram and 8-gram syringes cost just about $1.40 more, while the other MX-7 options are less than $0.50 more expensive than their older versions.

With so many Arctic knockoffs out there, it's a good idea to buy your MX-7 from trusted retailers. Arctic has also improved its MX Authenticity Check system, allowing you to easily verify your product’s authenticity using a unique QR code for each item. This helps ensure you get the real deal and have peace of mind with your purchase.

German PC cooling solution maker Arctic is showing off its products at Computex 2025, and the Tom’s Hardware team scoped out their booth to see its latest offerings. Of course, it has its flagship Liquid Freezer III Pro 360 A-RGB and Freezer 36 A-RGB on display, but we’ve also seen a few new items.

These include the Freezer 36S A-RGB, the Freezer 8 series, and the Pro version of the P12 and P14 case fans. Arctic is also getting back into PC cases with the Xtender and Xtender VG, plus it launched a solution for PC gamers who get too hot while playing.

The Freezer 36S A-RGB is basically the same as Arctic’s flagship air cooler, but with just a single fan. Despite that, the air cooler uses the new P12 Pro A-RGB fan, which offers a higher airflow than the previously used P12 PWM A-RGB fan. So, even though it has a more compact form factor that makes it fit more easily in smaller cases, we still expect it to deliver efficient and effective cooling.

Arctic also introduced a new tower air cooler for those on a budget. The Freezer 8 family of air coolers is designed for small-form-factor (SFF) builds and those who aren’t satisfied with the stock air coolers that come with affordable CPUs. It comes with a single P10 PWM fan that can hit 200 to 2300 RPM and is already pre-applied with MX-6 thermal paste, making installation a breeze.

The company launched two versions of this cooler — the Freezer 8A is created for AMD AM5 and AM4 sockets, while the Freezer 8i works with Intel LGA 1851 and LGA 1700 sockets. And if you plan to use your system for professional or industrial uses, the Freezer 8A and 8i also come with a Continuous Operation (CO) version, which uses a fan that has two ball bearings instead of the usual one.

Its cooling fans also received an upgrade with the Pro versions of the P12 and P14. These models offer higher airflow and a lower noise profile than the standard version, delivering better cooling for your entire PC case without making it sound like there’s an aircraft taking off from under your desk.

German-Swiss cooling specialist Arctic Cooling has emailed Tom's Hardware with updated details on its MX Authenticity Check system. The QR code-based system assures customers that they can enjoy "safety with genuine Arctic quality." The company confirmed that the program is in response to counterfeit pastes.

Arctic reiterates the common wisdom that "thermal paste plays a central role in any PC build that relies on effective cooling, as it ensures optimal heat transfer from the processor to the heat sink." We have extensive articles testing individual new pastes, compounds, pads, and our bumper round up with over 50 pastes tested and ranked (updated earlier this year). Arctic MX pastes rub shoulders with the best, and might be the nearest we have to an industry standard, as all Tom's Hardware CPU testing is performed with a blob of MX-4 betwixt IHS and cooler block.

Our email from Arctic mentions that counterfeits are becoming "increasingly prevalent," which seems to be why the new product verification system has been introduced.

So, how does it work? Well, Arctic has a live MX Authenticity Check site already, so you can go and see for yourself. In brief, when you receive your new thermal paste from Arctic or one of the many retailers that carry these compounds, you first check the seal, scratch it to reveal a QR code, scan it with your mobile, and then verify it online. According to Arctic, the whole process takes "just a few seconds." After you are done, you will know if your paste is genuine Arctic gloop, or a dastardly doppelgänger dupe. While the program has been around for a while, the new 4-step graphic (seen above) is a new feature to make the verification process more readily apparent to customers.

Is the Arctic paste tube in your spares drawer genuine?

Many PC enthusiasts will already have a tube or two of thermal paste in their toolboxes, tool bags, or drawers. We even reported on tests showing 20-year-old paste – an ancient syringe of Arctic Cooling Silicone Paste – is still useful, despite being left in the rearview mirror by the latest formulations. But without the new QR-code system in place, there doesn't seem to be a way to confirm that your Arctic paste is genuine unless you bought it directly from the company.

A syringe of ancient thermal paste has been tested by Igor’s Lab. With some surprise, the hardware testing stalwart reports that Arctic Cooling’s Silicone Paste worked just as well as a factory fresh industrial alternative with a similar formulation. However, Wallossek’s testing clarified that modern PC thermal pastes aimed at PC enthusiasts perform much better.

A PC hardware dealer in Germany discovered some 20-year-old thermal paste at the back of a drawer and decided to send it to Igor’s Lab. Wallossek looked at the Intel Pentium 4-era paste from Arctic Cooling and thought it was like finding an old can of soup – but one that “can’t spoil.” Without much hesitation, testing began.

The initial finding was that, since the syringe had a good seal, the paste squeezed out as it should, with an acceptable consistency.

The ingredients also piqued Wallossek’s interest. According to the packaging, Arctic Cooling's old Silicone Paste featured 50% silicone, 20% carbon, and 30% metal oxide. This required verification, decided the veteran hardware tester and ex-Tom’s Hardware writer. So, he employed a laser-induced breakdown spectroscopy device to analyze the paste’s ingredients. Wallossek reckons that the ingredients on the label were bunk, as the analysis tool reported the paste was made from 50% silicone and 50% zinc oxide. He concluded the mention of carbon on the packaging was solely for marketing purposes…

Moving onto the comparison testing, Igor’s Lab noted that a TCTG-4.0 from MCT, an industrial “durability-focused reference paste,” performs very similarly to the antique Arctic Cooling paste. This indicates that little, if any, degradation occurred during the old paste’s slumber at the back of the PC hardware dealer’s drawer.

The Pentium 4 era paste significantly underperformed compared to a modern thermal paste, though. Wallossek tested an Intel Core i9-13900K with both the old Arctic paste and the “modern, good” Alphacool Apex thermal paste. The graph of the comparative test run (below) shows a delta of six degrees Celsius, representing 20 years of progress in thermal paste formulation. It's not astonishing, but a very welcome improvement.

From the above tale, we might conclude that thermal paste packaging that features ‘use by’ or ‘best before’ advice from manufacturers might be wise to ignore.  This may only apply if the paste is kept well, like in a sealed tube in a dark, cool drawer.  I just checked my drawer and found three partly-used thermal paste syringes, and none of them seemed to have any ‘use by’ date indicated on the packaging.

For more thermal paste comparisons, look at our colossal roundup in our Best Thermal Paste for CPUs 2023: 90 Pastes Tested and Ranked feature.

Silent PC enthusiast and redditor revocases has been showing off a very interesting AMD Radeon RX 6400 modification (h/t FanlessTech). The modification work meant chucking away the stock active cooler, precisely machining a 3 mm copper cold plate and mount (covering both GPU and memory), and then fixing on an Arctic Accelero S1 passive cooler (discontinued). In testing, the project was a resounding success, with the GPU running at just a smidgeon over 50 degrees Celsius under load – but we have to ask, is this passive solution overkill?

Passive PC systems have a strong following, as silence is golden. However, to cool higher performance PC components a passive cooler might sometimes start to dominate your build. We have seen this with some recent implementations of passive coolers and cases. To be fair, we wouldn’t put revocases' RX 6400 mod into this category, though.

The Radeon RX 6400 is a 54W graphics card, so doesn’t present a strenuous challenge for passive cooling technology. Instead we could say the low TGP makes the card a comfortable target for a modest passive cooling adaptation. In revocases’ shared images you can see that the passively cooled (ex)low-profile Radeon ran at a more than comfortable average GPU temperature of 54 degrees Celsius (hot spot 60) when under full load. Sadly, off-the-peg choices for passive GPU coolers are incredibly limited or we would suggest using a smaller one.

In the images, the RX 6400 PCB is absolutely dwarfed by the Arctic Accelero S1, and might look somewhat silly outside of the context of  a PC. However, in November last year we reported on the Intel Arc A380 (>75W) being cooled by the same Arctic Accelero S1. In the visual context of a complete passive build that card looked reasonable – it was installed next to a Noctua NH-P1 cooled CPU.

Considering things from another angle, a commercially produced passive card like the Palit GeForce GTX 1650 KalmX looks well proportioned at 138 x 178 x 38 mm, and the GPU will peek at around 80 degrees Celsius under load. The Palit isn’t a great amount smaller than the Arctic Accelero S1 cooler at 138 x 215 x 33 mm (but you have to add the card which will add further bulk).

We hope that the great efficiency gains touted for the latest AMD and Nvidia graphics architectures means that some more compact passive commercially available graphics cards hit the market. Graphics card makers could be attracted by the passive premium that silent PC enthusiasts might pay for something more performant than the likes of the RX 6400, Arc A380, or GTX 1650.

Quiet PCs maker and enthusiast ‘Fully Silent PCs’ has revealed a new open-air test build which sports a passively cooled Intel Arc A380 graphics card. Apparently the setup was a cinch, as the 15 year old Arctic Accelero S1 passive cooler mated with the Intel Arc PCB without modifications or the use of special brackets.

Intel’s 75W TDP Arc A380 graphics cards are just about the limit for no-fuss passive cooling, but we have yet to see any AIBs market such a product. The fastest currently available passive graphics card that we are aware of is the Palit GeForce GTX 1650 KalmX. The Arc A380 isn’t too far ahead of the GTX 1650 in gaming, but it has some attractive modern media encoding advantages for content creators.

Silent PCs revealed that the graphics card was fitted with an old Arctic Accelero S1 heatsink with “no modifications necessary.” The silent PC builder went on to reveal that the 53mm spaced mounting holes on the Intel graphics card PCB aligned perfectly with the Arctic passive cooler, so no adjusting brackets were required.

It would be great to see some thermal testing and benchmarks with the passively cooled Arc A380, and Fanless Tech reveals that Matt, the owner of Fully Silent PCs, is working on producing this informative data right now. The site says that the finished build will be of the open air type, and updates with temperature readings will be provided in due course.

If we peek at the pictures shared by both Fully Silent PCs and Fanless Tech, the passive GPU cooler from Arctic is dwarfed by the passive CPU cooling block. We don’t have any reviews of this exact Arctic Accelero S1 product in the Tom’s Hardware archives, but we have featured passive PC builds with the Arctic Accelero Xtreme III, and more recently (2017) modification of a GTX 1050 Ti, dialled back to 52W for unthrottled performance, and fitted with an Arctic Accelero III.

In the pictures we can see that the Arc A380 being used is hooked up to a 6-pin power connector. Some models are designed to take over 75W, so the extra connector is necessary for peak performance, but there are models without 6-pin connectors on the market.

Hopefully, Fully Silent PCs will enjoy some success with the passively cooled Arc A380, and it will be interesting to find out any tweaks that are implemented to achieve acceptable unthrottled performance. Some actively cooled Arc A380 comparisons, in the same system, would be welcome too.

Custom printed board designs (PCB) and custom liquid cooling systems are pricey. Inno3D's GeForce RTX 4090 iChill Black and GeForce RTX 4090 iChill Frostbite boards have those features in an attempt to tame Nvidia's latest entry in our list of the best graphics cards, so expect them to be more expensive than typical AD102-based products. Yet the Frostbite version will likely be cheaper than buying an FE or a reference GeForce RTX 4090 board and an aftermarket waterblock from companies like EKWB.

Nvidia's AD102 graphics processing unit has been specially architected to hit very high clocks, but it needs decent cooling to push its way towards 3 GHz. To help the GPU to unleash its beastly nature, Inno3D teamed up with Arctic to build one of the industry's first GeForce RTX 4090 graphics card with a pre-installed all-in-one liquid cooling system. Also, Inno3D has a GeForce RTX 4090 with a pre-installed waterblock for custom liquid cooling systems.  

Actually, Inno3D is the only graphics cards supplier that currently has two liquid cooling options in its GeForce RTX 4090 lineup. The Inno3D GeForce RTX 4090 iChill Black comes with a custom closed-loop liquid cooling system from Arctic (via TechPowerUp), whereas the Inno3D GeForce RTX 4090 iChill Frostbite is equipped with a specially designed waterblock for custom-built liquid cooling systems. 

Both boards naturally carry Nvidia's AD102 GPU (with 16384 CUDA cores clocked at 2580 MHz boost frequency out-of-box) coupled with 24GB of 21 GT/s GDDR6X memory and features 12VHPWR (12+4-pin) auxiliary PCIe power connector designed to feed the board with up to 450W of power.

While the out-of-box GPU boost clocks of both cards do not seem to be too high (2580 MHz vs. 2520 MHz recommended by Nvidia is hardly a tangible increase), the highly efficient hybrid all-in-one liquid cooling system or a waterblock for a custom LCS promise to provide a very significant headroom both for higher boost clocks and for overclocking. Keeping in mind what typically can be done with GeForce RTX 4090, we are probably looking at 2.70 – 2.80 GHz boost clocks here without much hassle. With some additional luck and tweaking, some of these graphics cards may approach a 3 GHz GPU frequency (keep in mind that overclocking is a lottery). 

Another advantage of Inno3D's GeForce RTX 4090 iChill boards is their dimensions. Unlike Nvidia's GeForce RTX 4090 Founders Edition as well as AD102-based add-in-board with air cooling, the Inno3D GeForce RTX 4090 iChill Black and GeForce RTX 4090 iChill Frostbite are dual-wide graphics cards that can fit into any cases that can accommodate hybrid AIO or custom LCS. They are also shorter than air cooled boards, which will affect internal airflows within a chassis and will make it easier to install additional components after the boards are plugged in.

Update 09/22/2022 3:00 pm PT

Amazon Germany has listed the Arctic MX-6 for a whopping price of €28.75 ($23.76 without VAT) per 4 grams.

Original Article

Arctic's MX-series thermal compounds are among the most popular, among PC builders, and the company's MX-5 is widely considered the best thermal grease that it produces. But earlier this week MX-5 status was changed to end-of-life at the company's website (as discovered by @momomo_us) as apparently it is readying all-new MX-6 thermal paste.  

The biggest surprise is that Arctic discontinues its MX-5 just about 1.5 years after its launch in early 2021 and weeks before AMD and Intel are set to start selling their Ryzen 7000-series 'Raphael' and 13th Generation Core 'Raptor Lake' processors with increased maximum processor power. Yet, two images of the Arctic MX-6 packages posted by @momomo_us, a high-tech leaker with a good reputation, indicate that the new thermal grease is incoming. 

Arctic's MX-5 thermal grease is comprised of aluminum oxide, aluminum powder, aluminum nitride, polydimethylsiloxane, magnesium oxide and boron nitride. The paste consists microparticles which fill cavities on the surface of CPU heat spreaders as well as cooler bases. Thermal conductivity of the MX-5 is 8.5 W/m-K and it has a temperature range between -40 and 180 degrees Celsius, which essentially means that it's not designed for extreme overclocking. It has a viscosity of 550 pascal seconds, which is below that of MX-4, but it is still easy to apply. 

Arctic introduced its MX-5 in 2021, but its thermal conductivity did not change from MX-4 launched in 2019. While 8.5 W/m-K is not bad compared to cheap and mainstream thermal greases, it is considerably below thermal conductivity of market leaders like Thermal Grizzly's Kryonaut and Kryonaut Extreme that are designed with extreme overclockers in mind. In general, the MX-5 shoed very good performance under normal conditions, according to our rating of the best thermal pastes for CPUs around, but it looks like Arctic has something better, which is why it replaces the MX-5 while keeping the MX-4 afloat. 

Arctic's MX-5 did not replace the MX-4 and the two pastes continued to co-exist up until recently, even though the new one often performed better than its predecessor. By contrast, the MX-6 actually replaces the MX-5. 

We don't yet know exactly what to expect from  MX-6, but our hunch tell us that this will be a yet another all-around performer from Arctic, but perhaps designed to better suit the upcoming CPUs from AMD and Intel. In any case, the new thermal grease should be available shortly, so we are going to find out soon enough.

Arctic's MX-4 thermal compound has proven itself well in the DIY space, but it's been around for quite some time and seemingly, like everything in tech, in need of an update. So in comes, low-and-behold, the MX-5. The MX-5 thermal paste hasn't been officially announced yet nor is it listed on Arctic's website, but hardware detective @momomo_us found the compound listed on Amazon UK paired with an informative description.

Although full specifications aren't available at this time, what stands out most about the description is the paste's durability: it's alleged to last for up to eight years once applied.

Of course, enthusiasts generally change CPU multiple times in eight years, but MX-5's alleged durability can be extremely helpful for use in laptops, GPUs, or that PC you're building for a relative. Its not clear what the MX-5's thermal conductivity is, but if Arctic's historic products are anything to go by, it should be pretty good. 

MX-5 is metal-free and non-conductive, so it's also safe for use on GPUs without needing to worry about over-spilling on the area around the GPU.

A single 4g tube of paste (enough for well over a dozen builds) is listed on Amazon UK's site for £13.59, which translates to about $18.50. That's probably pre-order pricing though, and it will likely drop in due time. According to Amazon the MX-5 thermal grease will be released on March 15 2021.

Arctic's Liquid Freezer II line has been out for a while now, and we've already reviewed the 280 mm variant. It came out great, but what if you want even more cooling power? A 360mm variant is longer but uses smaller 120 mm fans, offering nearly identical performance. For the ultimate enthusiasts, Arctic is adding a new version to its product line: Liquid Freezer II - 420. 

The Liquid Freezer II 420 may potentially be the biggest all-in-one liquid cooler on the market, if not the biggest ever made. Most all-in-one liquid coolers come in 280 mm (2x 140mm) or 360 mm (3x 120mm) as their biggest variants, so this 3x 140mm flavor definitely stands out from the crowd.

The Liquid Freezer II 420's CPU block comes with its own little 40 mm fan to cool the area around it and add a little airflow to the VRM circuitry, and this fan can spin between 1000 and 3000 RPM using PWM control. The pump itself spins from 800 to 2000 RPM, and the three 140 mm fans offer a lavish range from just 200 to 1700 RPM. That being said, we can't imagine a scenario where an AIO needs all three 140 mm fans to spin this fast, so you're mostly buying this cooler for the silence benefits offered by using multiple large fans.

But the one thing to keep in mind with an AIO this big is case support. Generally, the largest size supported by most cases is 360 mm, so you'll have to carefully double-check your case's specs to see if this will fit. In total, the radiator is 458mm long, so do make sure you have the clearance for it. Most standard sockets are supported, including Intel's LGA1200 and AMD's AM4. 

No word on US pricing, but pricing in the EU is set at just €120, which is generously affordable for a unique product such as this one.

ID Cooling has introduced its new twin-tower 'mega cooler' for mainstream processors that is rated for a TDP of up to 280 watts. The SE-70 cooling system is designed for upcoming enthusiast-class CPUs from AMD and Intel that need a lot of cooling when overclocked. Only time will tell if it's one of the best CPU coolers.

The intensified competition on the market of enthusiast-grade mainstream processors for desktops have led to increase of their TDP in the recent years. Formally, modern processors dissipate from 95 watts to 125 watts and can be cooled down using relatively modest coolers. But if someone wants them to hit their maximum dynamic turbo clocks, they have to use considerably more powerful cooling systems. For many, the obvious choice is a closed-loop liquid cooling system, but others prefer good-old air coolers that are sometimes struggling to keep up with modern performance requirements.   

The ID Cooling SE-70 is a rather classic mega cooler that features a twin-tower design with dozens of aluminum fins, a copper base, seven 6-mm heat pipes, and two 120 mm fans. The unit weighs rather 1.3 kilograms and measures 122×141×156 mm, too large for a mainstream PC chassis, but not too huge for an enthusiast-grade PC case. Meanwhile, those who plan to use the cooling system will have to ensure that its large dimensions do not block space for memory modules with larger heatsinks. 

Keeping in mind that AMD has recommended to use 360-mm closed-loops liquid coolers with is Ryzen 5000-series processors and Intel's upcoming Rocket Lake-S CPUs might likely be rather hot too, mainstream platforms for enthusiasts are going to need advanced air coolers like ID Cooling's SE-70 that is rated for up to 280 W.

The fans with hydraulic bearing spin at a 700 RPM ~ 1800 RPM speed to produce an up to 76.16 CFM air pressure while generating an up to 35.2 dBA noise level.  

The SE-70 is designed for AMD's processors in AM4 packaging as well as Intel's CPUs in LGA1200/115x/20xx form-factors. Given design of its base and some other factors, the manufacturer decided not to equip its SE-70 with a mounting mechanism for AMD's sTR4 platform.  

So far, ID Cooling's SE-70 has been announced only in China and Japan, so it is unclear whether and when it will be available elsewhere. But since demand for high-performance air coolers exists in general, either ID Cooling or its rivals will offer mega coolers rated for 280W or higher in Europe and the U.S.

Arctic has unveiled a version of its legendary Freezer 50 cooler for mainstream AMD and Intel processors. This massive cooler weighs nearly 2.6 pounds (1.2  kilograms), has two fans, and promises to stay quiet even at reasonably high loads. The Freezer 50 launches just in time for AMD’s enthusiast-grade Ryzen 5000-series CPUs that will be available this November.  

The original Arctic Freezer 50 TR cooler was designed for AMD’s Ryzen Threadripper processors which can dissipate up to 280W at stock and a lot more when overclocked. Most owners of Threadripper systems use liquid coolers as they are concerned about properly cooling their power-hungry CPUs running at high clocks. This means that there are only a few air coolers for sTR4 processors on the market. Given the typical use cases of AMD’s Ryzen Threadripper, all of these air coolers are designed with overclocking in mind, so they can dissipate significantly more than 280 W. Applying this design to ‘mega coolers’ aimed at more mainstream CPUs will naturally give them a lot of headroom for overclocking. In fact, keeping in mind that AMD wants its Ryzen 5000-series CPUs with a TDP higher than 65 W to be cooled using a rather potent 280-mm all-in-one liquid cooling system or an ‘equivalent’ air cooling system, Arctic launches its Freezer 50 at the very right time.  

Indeed, the Arctic Freezer 50 promises to offer performance comparable to that of AIO liquid coolers. The unit comes with a gigantic dual-tower heatsink featuring 104 0.4-mm aluminum fins as well as six U-shaped 6-mm heat pipes that form a direct-touch base which covers the whole surface of a CPU’s IHS. 

The cooler uses two fluid dynamic bearing fans of different sizes — a 120-mm one and a 140-mm one — to maximize airflow and therefore increase performance. Arctic says that depending on the load, the 120-mm fan rotates at 200 ~ 1800 RPM, whereas the 140-mm fan rotates at 200 ~ 1700 RPM, so the cooler should be utterly quietly under light and even medium loads. Arctic claims that noise level produced by the cooler at max speeds should not exceed 0.4 Sone (23.5~25 dBA). 

Arctic’s Freezer 50 cooler is compatible with all modern mainstream platforms, including AMD’s AM4 as well as Intel’s LGA1200, LGA2066, LGA115x, and LGA2011-3 sockets. As far as dimensions are concerned, the Arctic’s Freezer 50 measures 5.8 x 5.9 x 6.5 inches (148×149.5×166 mm), just like its counterpart for Threadripper processors, so it will be compatible with the vast majority of spacious cases designed for enthusiasts. It also leaves a memory module clearance of 37.5 mm to maximize compatibility with high-end DRAM sticks.  

Since modern enthusiast-grade hardware should not only offer high performance, but also look flashy, the Freezer 50 comes with 13 A-RGB LEDs that can be controlled using software from leading makers of motherboards. 

The Arctic Freezer 50 is already available directly from the company in Europe for €59.99, including VAT. The cooler will probably hit major retailers in the coming weeks. 

AMD Ryzen 4000 mobile APUs already arrived this year, but according to a tweet from hardware detective @_rogame, it's possible we'll soon see AMD Ryzen 4000 APUs make their way to the desktop with AMD's Zen 2 microarchitecture. 

AMD has pretty much flushed out its processor product stack with 7nm parts. Only the desktop APU hasn't received the Zen 2 treatment. AMD's current desktop APUs, (codenamed Picasso), are still on the Zen+ microarchitecture and GlobalFoundries' 12nm manufacturing process. The transition to Zen 2 and the 7nm process node would bring some significant improvements to the segment.

The unidentified processor's core and thread count is currently unknown, but it reportedly runs within the 35W envelope. Given the TDP rating, it could be an Athlon or one of those Ryzen 3 or Ryzen 5 GE-series parts that are optimized for efficiency. For context, existing Ryzen desktop APUs span up to four cores, while Athlon chips are stuck at two cores. The hardware sleuth seemed to think that the processor could be the Ryzen 3 4200GE or Ryzen 3 4100GE.

According to the tweet, the CPU features a 3 GHz base clock with performance somewhere between a Ryzen 5 4600U and Ryzen 7 4700U. The hardware detective didn't share the exact design of the iGPU; however, it was seemingly operating with a 1.2 GHz clock speed. Graphics performance was allegedly slower than a Renoir chip with six Vega Compute Units (CUs).

The desktop AMD Ryzen 4000 APU was reportedly tested with a system using an AMD motherboard that appeared under the "Artic" codename, which could point to an unreleased chipset. The processor was paired with DDR4-3200 RAM, so there's no doubt that it's a desktop processor. In any case, the APU should be backwards-compatible with AMD's 300-and 400-series motherboards with a mere firmware update.

In Leipzig, Germany, Arctic presented a prototype for its upcoming Freezer 50 TR heatsink designed for Threadripper CPUs. Being billed as a semi-passive, low-noise cooler with addressable RGB lighting, the all-aluminum construction is absolutely massive and wields a total of eight heatpipes and two radiators.

Arctic hasn't published figures for the coolers' exact weight, dimensions, or fans, but it's likely this is among the largest and heaviest CPU coolers ever made. Arctic also has not disclosed which Threadripper CPUs this cooler can actually cool. However if its two fans are allowed to ramp up enough, it could likely cool even the Threadripper 2990WX. That would, of course, negate the purpose of having a silent workstation, though it would probably still be quieter than most coolers.

Being a prototype, it could be some months before this cooler comes to market. As such, Arctic did not disclose pricing, potential launch dates, or whether or not it will support upcoming Threadripper CPUs (which is likely as long as Threadripper doesn't get a new socket).

AnandTech speculates Arctic's cooler may launch around Computex, which is a likely date for any upcoming tech product, especially AMD's upcoming Ryzen (and perhaps Threadripper) 3000 series CPUs which are to be available this summer.

Want to comment on this story? Let us know what you think in the Tom's Hardware Forums.

Those of you planning on upgrading to one of AMD’s upcoming Threadripper processors will be happy to know that Arctic Cooling has updated its line of all-in-one coolers to be compatible with the TR4 socket.

The Liquid Freezer line of coolers all share the same the 12v water pump equipped with a micro skived fin copper cold plate and are available in 360mm, 240mm, and 120mm sizes. Although the radiators all share the same basic high fin count aluminum construction, each varies in size and thickness. The Liquid Freezer 360 radiator measures 120mm wide and 390mm long, but it’s relatively thin at just 27mm. Conversely, both the 120 and 240 models are 38mm thick and measure 120 x 272mm and 120 x 155mm (W x L), respectively.

All three Liquid Freezer models employ Arctic Cooling's proprietary F12 120mm cooling fans. They’re capable of 74CFM airflow each and ship in pairs (that is, 2x fans for the 120, 4x for the 240 and 6x for the 360). This allows you to install each cooler in a push-pull configuration if so desired.

According to Arctic Cooling, the Liquid Freezer 360 and 240 are both capable of handling heat loads up to 350W (recommended for TDP up to 300W). The Liquid Freezer 120 is rated at 300W (recommended for TDP up to 250W).

The Liquid Freezer line of coolers is compatible with  AMD's TR4, AM4, FM2+, FM2, FM1, AM3+, AM3, AM2+ and AM2 sockets as well as Intel's LGA2011, LGA115X, LGA775/1366 sockets.

Currently, the Liquid Freezer 360 retails for $140, the 240 for $100, and the 120 for $86. All three coolers are available from the company website.

More Radiator, More Fans, Less Money?

Arctic has gained a reputation among CPU overclockers as a value brand, often producing slightly less cooling capacity than top “Big Air” models at a substantially lower price. GPU gurus, on the other hand, will recognize Arctic as the brand behind many of the dual-fan and triple-fan graphics cooler designs that are too thick to fit into two slot spaces. Bridging the gap between value priced CPU coolers and oversized GPU coolers is the monstrously thick Liquid Freezer 240 closed-loop liquid CPU cooler.

The Liquid Freezer 240 starts off with a 1.5” thick radiator, which is somewhat of a standard for enthusiast-grade liquid cooling, and sandwiches it between four fans, which is the standard for making under-performing coolers perform. Based on those specifics alone we’d expect the Liquid Freezer 240 to be one of the top performers in the closed-loop class, though at the cost of some fitment issues. Many mainstream gaming cases are now designed with around 2.5” of radiator clearance, and the second layer of fans pushes the Liquid Freezer 240 a little past the 3.5” mark.

The Liquid Freezer 240 uses a mounting ring similar to the classic Asetek design, but adds locking barbs rather than hooked ends on the metal ring’s tabs. The ring is secured to four standoffs, and Arctic includes different standoffs for AMD, Intel LGA 115x (1156/1155/1150/1151), and Intel LGA 2011x (v1 and v3) motherboards.

The base itself appears to be unpolished, lathe turned copper.

The Intel mounting ring is drilled for both LGA 115x and 2011x installations. Builders who are willing to reuse their old LGA 1366 threaded socket support plates may be able to combine it with Arctic’s 115x standoffs by using the mounting ring’s 2011x holes, but package thickness (the thickness of the CPU, LGA, and heat spreader) could still be an issue for that type of unsanctioned installation. Anyone willing to perform an experimental installation should feel free to post their findings in this article’s response thread.

LGA 2011x standoffs are threaded differently on one end, to fit the motherboard’s integrated support mechanism. LGA 115x installations use an Arctic-supplied support plate, which must be mounted behind the board to receive the M3 x 0.50mm standoff ends. AMD installations are similar to those of LGA 2011x, except that they require builders to first unscrew the standard hook brackets from the top of their boards. AMD motherboards with pinned-on brackets aren’t compatible, unless the builder replaces the motherboard’s rear support plate with a nutted or threaded version.

Arctic recommends an intake orientation for its fans, even when the unit is mounted on the top panel. The firm also lists exhaust orientation as a secondary option. Putting aside any ideas about convection, we should note that our case only has dust filters on its front and bottom panels. A “standardized” test configuration doesn’t allow us to move the front fans, and that means top panel intake fans will create positive pressure that resists additional airflow. The easy solution for us is to test the Liquid Freezer 240 in both intake and exhaust fan configurations, though actual builders may instead choose to front-mount their Liquid Freezer 240 in intake orientation.

MORE: Best CPU Cooling
MORE: How To Choose A CPU CoolerMORE: All Cooling Content
MORE: In Pictures: 20 Clever Liquid-Cooled PC Setups

Test Method, Results And Analysis

We retain the hardware configuration from previous big cooler reviews to maintain result consistency.

Some of Arctic’s toughest closed-loop competition comes from Chinese brands with an American presence, both on the pricing and performance scale. ID-Cooling’s Frostflow 240L for example has proven itself formidable even against units priced up to 50% higher, and the Gamer Storm Captain 240 is no slouch either. Rather than create a false value narrative by comparing under-performing expensive models, we’re throwing the Liquid Freezer 240 right up against those two best-value models.

Liquid CPU Cooler Specifications

Arctic’s two-year warranty resembles those of ID-Cooling and Deepcool (Gamer Storm), but only at first blush. According to Arctic’s website: "ARCTIC does not offer a manufacturer warranty." Warranty returns are at the discretion of the seller, and Arctic is responsible to the seller (its direct customer) when a product is found defective. Buyers can expect no assistance when the seller refuses to act as an intermediary (or goes out of business).

[Update: Arctic contacted us to let us know that it has changed its warranty policy. Its new RMA instructions can be found at https://www.arctic.ac/de_en/rma/.]

Test Results

Thanks to its thick radiator and airflow-boosting fan-sandwich configuration, the Liquid Freezer 240 beats even the unbelievably cool Frostflow 240L in our thermal test. It even wins in its worst-performing configuration.

You may have also noticed that the Liquid Freezer 240 performs better when set as exhaust. Once again we’re going to ignore the miniscule impact of rising heat and instead credit the lower pressure difference, since our front fans are always tested as intakes.

While it appears that the fan speed actually goes down when the Liquid Freezer 240 is mounted as exhaust, the measurements on the charts are only coming from the fastest layer of fans. Our test notes show that while the fastest fan slowed from 1560 to 1420 RPM, the slowest fan sped up from 1340 to 1420 RPM.

The Liquid Freezer 240 is always quieter than either the Frostflow 240L or the Captain 240. Higher noise in its top-intake configuration is likely due to windage, as fan blades that are close to a case’s vent holes create a whistling noise as they pass each hole. When mounted as exhaust, fan frames act as spacers to prevent that noise.

Acoustic Efficiency, or the comparison of temperature to noise, is the true measure of a cooler’s overall performance. Arctic’s Liquid Freezer 240 destroys its rivals.

While we didn't include the Liquid Freezer 240's oversized rivals in the charts of this review, it outperforms them all, even making its higher price (by about $10) worth it. And, remembering that the coolers we compared directly in this review are some of the cheapest high-performance parts available, the Liquid Freezer 240 is still cheaper than most of its other rivals.

Astounding performance superiority easily opens the Liquid Freezer 240 to our elite “Editor’s Choice” award, but would this editor actually choose it? Unfortunately, its 3.6” thickness doesn’t fit any of my cases! I even had to offset it within the case of our own test machine, using vent holes rather than mounting holes to secure the radiator and fan sandwich. We have, however, tested cases with fairly severe default radiator offset, and those cases allow the cooler to hang well past the top of the motherboard and even below the tops of some memory modules.

Normally a fitment issue like that might exclude a product from making our “Recommended” list, yet the relatively broad availability of cases with significant radiator offset makes the case against this award less clear. Two things are clear in that buyers get superior cooling for under $100, and that warranty replacement is at the discretion of the seller. Thus, I’m recommending that new system builders keep this cooler in mind when choosing a case, and restricting its recommendation to cases that support it when purchased from trustworthy sellers.

MORE: Best CPU Cooling
MORE: How To Choose A CPU CoolerMORE: All Cooling Content
MORE: In Pictures: 20 Clever Liquid-Cooled PC Setups

Follow us on Facebook, Google+, RSS, Twitter and YouTube.

39 Thermal Compounds Get Benchmarked And Tested

If you missed part one of this series, take a quick second and check out Thermal Paste Comparison, Part One: Applying Grease And More.

We Are Curious About Two Special Compound Categories

After discussing the theory of conventional thermal pastes in part one, I want to go into a little more detail about liquid metal compounds, and also shed some light on thermal adhesives, both in paste and pad form. Thermal pads are used to mount small heat sinks on RAM chips or MOSFETs.

But let’s start with those liquid metal pastes. While their theoretical thermal conductivity sounds impressive, that one attribute is no guarantee of success. During the course of our testing, we'll be experimenting with a long burn-in phase and comparing the liquid metal compound to Gelid's Extreme paste.

Several years ago, Coollaboratory broke new ground in thermal management by introducing the Liquid Pro liquid metal compound. But while the look (and subsequent application) of Liquid Pro reminded us of a mercury thermometer that didn’t survive a fall, its successor, Liquid Ultra, is better, avoiding some of the first effort's weaknesses. Because Liquid Pro could be so problematic, I'm only presenting Liquid Ultra in the tutorial section. Both compounds show up in our performance charts, though.

Box Contents

In the Liquid Ultra box you get the paste itself, a cleaning swab, two brushes, a sponge, and an instruction manual. That's an impressive-sounding list, but it's really just the bare minimum to use the product, sufficient for a single application. If you want to apply Liquid Ultra a second time, you'll find yourself without the alcohol swab. At least a second brush is included. While we appreciate the inclusion of the coarse sponge, it's barely adequate for removing the compound. Of course, there's a separate cleaning kit available for a hefty price.

Enthusiasts may be torn about this product, and yes, there may even be those who are more courageous than I and use it on graphics cards. However, my personal opinion is that a majority of the people who read this tutorial are new to thermal compounds, and I wouldn’t want to encourage them to mess around with this stuff due to the skill it requires. I also want to point out that you'll probably void your CPU's warranty if you do use Liquid Ultra. After removing the last remnants of it, we discovered that all of the markings etched into the head spreader were gone. If you still want to try this product after reading our tutorial, you can probably expect very good cooling performance, assuming nothing goes wrong.

Surface Cleaning and Roughing

The most important prerequisites for using a liquid metal compound are clean heat sink and spreader surfaces. You can buy the aforementioned kit with the three cleaning liquids, or simply snag some isopropyl alcohol at the drug store. However, stay away from acetone and cleaning naphtha. Even denatured alcohol may contain additives that are detrimental to achieving a clean, degreased surface. Finally, make sure you wait until any remnants of the liquid have evaporated!

If the surface is too smooth, the older Liquid Pro and newer Liquid Ultra may only form loose droplets on your CPU. Thus, in contrast to what you would do for regular thermal pastes, you may consider roughing the heat sink and spreader a little bit. Just remember that you only get enough liquid metal for two tries.

Don't overdo this. If you scrub too hard, the innocent-looking sponge can cause deep grooves. Move in small, graceful circles.

Coollaboratory Liquid Ultra: Application

Applying the compound is tricky, which is why we don't really recommend Liquid Pro. Liquid Ultra is a little more like the pastes you're used to, though still a lot "wetter". Its composition is still a point of concern because it corrodes light metals like aluminum and certain alloys. This could have an adverse effect on thermal conductivity, even causing an insulating layer to form. Copper- and nickel-plated surfaces are alright though, as are heat spreaders.

Once you've roughened up your CPU, you can drip Liquid Ultra on it, spreading it around with an enclosed brush. The amount you'll use varies by CPU size, but plan to use about half of the compound on AMD processors and about a third on Intel's. The video shows that you can even add liquid metal while spreading it out. Depending on the mounting pressure of your cooler, you may want to spread the compound more than the video shows. Just be careful; you don't want this corrosive, conductive material to get squeezed out the sides.

The next video illustrates an alternate method that requires less material. What we want to stress, however, is that it is better to put the liquid metal on a CPU away from its processor interface, thus protecting the motherboard from spills. We're putting the chip on our cleaning sponge to make sure none of its pins get bent.

On one hand, the liquid is spread more evenly and less of it is needed. But on the other, more metal balls form. Since they could cause a short circuit, they need to be collected. This method only works when the surface has been roughened, and it yields (marginally) better results.

Coollaboratory Liquid Ultra: Application Problems And Clean-Up

The pictures below are staged. But their purpose is to warn you what might happen if you're not careful, or if you don't pull the CPU from its interface before applying the liquid. Even if you're being meticulous, small droplets will separate and run away. So, keep a thin, clean brush handy to deal with them. Don’t spread them or wipe them away; instead, lift them up!

If you have never experienced the sinking feeling of spotting a liquid metal droplet between CPU pins, look at the bottom picture. However, the tiny droplets at the PCB’s edge may be even worse because they're barely visible to the naked eye.

Cleaning with Chemical Agents and a Wire Brush

It's one thing to apply liquid metal. Removing it is an entirely different ballgame. Even popping off the heat sink can be a challenge. It's very easy to pull an AMD processor right out of its locked socket if you're not extra careful. Make life a little easier by starting the process while the processor is still hot, and detach the heat sink by gently turning it left and right. If you roughed both the heat sink and spreader, you're going to be met with resistance. You might be best off mounting a smooth heat sink on a roughed spreader. This combination seems to come apart most easily.

The cleaning agent from Coollaboratory's cleaning set works, but it won't get everything off. Thus, you still have to use the wire brush. Scrape, apply more solvent, wipe, rinse, and repeat. As you might guess, this takes some time.

In the end, you'll probably wind up with two scratched-up surfaces. Luckily, I bought two closed-loop liquid coolers, allowing him to continue testing conventional pastes with a smooth sink.

Coollaboratory Liquid Ultra: Performance Measurements

If you want to know if this solution is worth its high price, the best we can do is answer with a non-committal maybe. Enthusiasts requiring an expensive, hard-to-apply thermal compound to lower Tcase by 2 or 3 degrees Kelvin (which could mean as much as 5 degrees Tcore) might want to consider their build goals. After running our own tests, we can state with some confidence that the supposed improvements of 5 to 7 Kelvin Tcase are overoptimistic. If you see that big of a gain, it's because you weren't doing something right with the paste you were using before. In order to create a more fair measurement, I compared one of the best conventional pastes, the GC-Extreme from Gelid Solutions, against Liquid Ultra. One-eighth of an ounce of GC-Extreme may be good for five or more applications; Coollaboratory's Liquid Ultra allows three, at most. Even without the expensive add-on cleaning kit, you're still looking at the difference between less than $2 per use and more than $4. You're also looking at five seconds of application time compared to three minutes or so. And we still haven’t taken increased risk into account.

Measurement Results

While Liquid Ultra isn’t compatible with aluminum coolers, we're lucky in that the Core 2 Quad Q6600's old boxed cooler sports a copper slug.

It is amazing how well conventional thermal paste fares compared to liquid metal. A more familiar paste sports a lower difference between burned in versus not burned in, but at a mere 1°C ΔT, the Liquid Ultra certainly proves itself in the same league. Whether we run the interval test or the 24-hour stress test, only Gelid's paste manages to get slightly better as time passes.

Asked whether the 1.5 °C ΔT, which the Liquid Ultra holds over Gelid's GC-Extreme with a closed-loop cooler, or the 2.7 °C  ΔT, which it does in the boxed cooler test, are worth extra effort and cost, my personal answer is no. But of course, some enthusiasts swear by the liquid metal. Most folks don't need the stuff; it's really for the overclockers functioning at the very top of their game.

Thermal Adhesives: Arctic Silver And Akasa

Thermal adhesives binds surfaces together, acting as a thermal compound in the process. They're available both as pastes and in pad form, and may be necessary for mounting small heat sinks on memory packages or voltage regulation circuitry. Obviously, conventional pastes don't work in those cases, since the weight of a heat sink causes them to slide right off.

Adhesive pastes have better thermal properties than pads, but they are accompanied by a significant drawback: once thermal adhesive paste cures, you cannot remove the heat sink from the component without damaging something. We've seen too many pictures from folks who've torn off RAM packages or cracked MOSFETs in half. You can try a trick to avoid this: mix thermal adhesive with regular thermal paste. We'll discuss this shortly.

Paste: Arctic Silver Alumina Thermal Adhesive

Arctic Silver's Alumina is a classic among thermal adhesive pastes, but heed the above warning: once applied your heat sink won't come off again. But a similar non-adhesive version is also available, and you can mix about two parts of the adhesive with one part regular paste to achieve a compromise. Your sink shouldn't be affected by gravity or vibrations, but it should still come off if you need it to. After popping it off, remnants of the adhesive will remain on both surfaces, which you'll want to clean with a sponge.

Adhesive Pad: Akasa AK-TT12-80 Thermal Adhesive Tape

It is easier to use an adhesive pad, if you don't mind the compromise in thermal conductivity. But for making sure that RAM packages and MOSFETs don’t get too hot, pads are typically good enough. If you ever plan to change or remove the heat sinks you're attaching, use an adhesive pad rather than paste.

Pastes: Akasa And Alpenföhn

Akasa 455

Model 455 is basically the entry-level product from Akasa. However, it is quite thick and thus not suitable for beginners. A low price cannot make up for the paste’s difficult application.

Akasa Pro-Grade 460

This paste barely differs from the Akasa 455, described above. It is not any easier to use, and the results we measured are only slightly better. Since it's also more expensive than the 455 paste, its purchase is questionable.

Akasa Pro-Grade+ 5022

This is another product from Akasa that costs more than the one we just looked at. While the performance measurements we generated are again better than the previous two compounds, and it is somewhat softer, this is still a thick paste. Warm it up a bit, though, and it can be applied more easily, yielding very satisfactory results.

EKL Alpenföhn Schneekanone

Schneekanone translates to Snow Cannon, which sounds like a whimsical name promising icy-cold CPUs. While your processor certainly won't hit temperatures that low, this paste's thermal performance is still pretty good. However, its price is too high and it's not very easy to use. While it is more expensive than the classic Arctic MX-2, it isn’t any better. Fortunately, enthusiasts in the U.S. won't need to worry about this, since EKL's Alpenföhn Schneekanone isn't available here. 

Pastes: Arctic And Arctic Silver

Arctic MX-2

This classic compound started the premium paste frenzy. It is still a good entry-level choice, too. Inexpensive and easy to use, its thermal performance is in the upper mid-range. And for what you pay, MX-2 seems like it's meant to keep competitors out of the lucrative thermal paste market.

Arctic MX-4

While MX-4 is more expensive than the MX-2, it surprises us by offering only barely better results. This paste may have its strengths, but it's not your best option for high-power applications like GPUs. It's also tougher than Arctic's MX-2, making it more difficult to use, and less attractive to novices. The thermal performance we recorded puts MX-4 into the upper mid-range, but it also doesn’t stand out from the crowd. Fortunately, at $1.75 per gram, it is one of the least-expensive pastes.

Arctic Silver 5

Buyer beware: The brand Arctic Silver is not the same as Arctic. This is one of the few pastes on the market that still sports silver as an ingredient. However, it shows its age and does not keep up with the best pastes out there today. Based on its performance, it seems overpriced.

Pastes: be quiet!, Corsair, And Dimastech

be quiet! DC 1

As with a number of other manufacturers, be quiet! merely puts its label on an OEM product. That doesn't necessarily indicate poor quality, so long as the OEM maintains certain standards. And as it turns out, the DC 1 is a solid paste for beginners, ending up in the upper mid-range. Its handling is similar to MX-2, while its performance is a notch better. While we like the product’s quality, we cannot say the same about pricing. Amazon currently offers 3 g for close to $10.

be quiet! (Paste Bundled with be quiet! Heat Sinks)

We're curious about the pastes that heat sink vendors bundle with their coolers. Since cost is always a factor when it comes to value-adds, we want to know whether it makes sense to toss the included stuff and buy aftermarket compound instead.

In this case, we are positively surprised by the freebie. It should be good enough to complement be quiet!'s coolers, which employ high mounting pressure. Under those conditions, the paste comes in a mere 0.6 °C higher than the DC 1, mentioned above.

Corsair (Paste Enclosed With Corsair Coolers)

This time we're looking at thermal paste bundled with Corsair's coolers. The be quiet! and Corsair compounds achieve similar results, though Corsair's paste is perhaps slightly thicker. It's still pretty easy to use though, and we like what we see from our thermal performance results. In this case, it probably wouldn't make much sense to toss the tube and spend good money on something else.

Dimastech HTX-EE

This compound is one of the most difficult to use, and you should only apply it after warming to 110 °F or so. Its performance results are barely better than what you get for free with some heat sinks, though, and so we're left a little disappointed. We're not saying that this is a bad paste, but for the money you can find better options. This stuff is so thick that we were afraid our heat sink's screws would strip their threads as we tightened them.

Pastes: Coollaboratory

Coollaboratory Liquid Pro

This is the grandfather of liquid metal compounds, and it still holds the record for thermal conductivity. Unfortunately, it also requires the greatest degree of skill to apply. Only experienced, gutsy professionals should use it, and even then, it's both expensive and difficult to remove. Liquid Pro cannot be used with aluminum heat sinks, but can be used with copper- and nickel-plated ones.

Coollaboratory Liquid Ultra

This latest offering from Coollaboratory is somewhat easier to use than its predecessor. But that only means it's a little easier to apply than Liquid Pro. Its thermal attributes aren't quite as good as a result of the different composition. However, Liquid Ultra is still better than any conventional paste. At the end of the day, you'll have to decide if one or two degrees of improved cooling performance is worth the effort and risk. Again, you cannot use this stuff with aluminum heat sinks, though it is compatible with copper- and nickel-plated ones.

Coollaboratory Liquid MetalPad

While these pads can be handled and cut to size easily, the devil is in the details. During burn-in, you have to ensure that the Tcase temperature reaches at least 140 °F, which equates to a Tcore of about 80 °C (176 °F). Some AMD CPUs throttle back at 140 °F, and even our FX-8350 had a hard time holding the required temperature long enough for a successful burn-in. Older AMD processors may be on the brink of meltdown at 140 °F. Without a successful burn-in, however, this product's thermal conductivity is worse than the cheapest paste. It took disconnecting all of my fans to burn the pads in. Naturally, doing this involves some risk.

Coollaboratory Liquid Cleaning Set

High prices and tedious clean-up make Coollaboratory's products very hard for the newbie to use. They're great for experts who want to leave zero overclocking headroom unexploited, though you'll probably find this cleaning kit necessary.

Pastes: Cooler Master

Cooler Master IC Value V1

This is the least-expensive paste from Cooler Master. It's really easy to use, but doesn’t work any better than the random no-name compounds we've tested. While the price is right, performance leaves a lot to be desired. We can only recommend this product for experiments like this one. Fortunately, it's pretty hard to find in the U.S., so there's a good chance you'll never encounter it.

Cooler Master IC Essential E1

This mid-range compound from Cooler Master is approximately twice as expensive as the entry-level product, but is it twice as effective? Our results put the two pastes fairly close together; only in the GPU-oriented test are the differences substantial. But even so, this paste is fairly average, the amount you get is paltry (a mere 1.5 g), and the cost per gram is substantial. IC Essential E1 isn't a price/performance winner. It is easy to use, though.

Cooler Master Extreme Fusion X1

This is Cooler Master's high-end offering, which sets out to outclass the company's older pastes. It achieves this goal, finishing a close second to Gelid's GC-Extreme. Just be sure to warm up the paste before you use it.

Cooler Master (Paste Bundled With Cooler Master Coolers)

Initially, I suspected that the paste bundled with Cooler Master's heat sinks was the same as be quiet!'s, given almost identical syringes. However, spreading the pastes, which are also the same color, revealed different viscosity. That either means we're dealing with distinct products or the same product from widely varying production lots. Variance does happen, particularly when it comes to inexpensive products.

Be that as it may, Cooler Master's paste sports a higher viscosity and winds up trailing be quiet!'s in the test results. It was still better than some pricey aftermarket compounds, though.

Pastes: Gelid Solutions, Noctua, And Phanteks

Gelid Solutions GC-2

We're going to start with Gelid's most budget-oriented solution. Its thermal results are respectable, landing it in the vicinity of Arctic's MX-2, MX-4, and Arctic Silver 5. But it's also noticeably thicker, making it harder to manipulate at room temperature. Nevertheless, this is a solid product that sells for less than $1/g, making it the price/performance leader.

Gelid Solutions GC-Extreme

GC-Extreme is the best thermal paste in our round-up, maintaining a slim lead over the runner-up as long as you're using a heat sink with high mounting pressure. If this compound was a little softer, it'd be even more ideal and wouldn't require warming before application. Heating it up a bit does pay off though, and we recommend it for overclockers and enthusiasts. Due to its high viscosity, it's perhaps less ideal for beginners.

Noctua NT-H1

This paste is almost a classic by now, and it still gets widely recommended. Rightfully so? Our testing reveals that its price/performance ratio is almost as good as Gelid Solution's GC-2, and its thermal parameters are even better. Noctua's solution ends up in the upper mid-range of our contenders, and a low price makes it hard to pass up unless you're a hardcore overclocker.

Phanteks TH-NDC (Paste Bundled With Phanteks Coolers)

The paste that Phanteks includes with its heat sinks contains nano particles of diamond, approaches the thermal performance of Gelid's GC-Extreme, and can be applied as easily, if not more so. It is somewhat similar to Cooler Master's Extreme Fusion X1, but can be applied even by novices. A big shout-out to Phanteks for bundling TH-NDC with its coolers.

Pastes: Prolimatech

Prolimatech PK-1

If we were judging based on price, we certainly wouldn't call this an entry-level thermal compound. But how does it perform? Nano aluminum sounds pretty compelling, but the benchmark results are what matter to us. The PK-1 paste performs well enough, but doesn't blow us out of chairs. Really, it's the price that kills it. There are better alternatives out there, even if Prolimatech's entry-level solution is generally quite user-friendly.

Prolimatech PK-2

Yes, Prolimatech offers more than one thermal paste. The PK-2 compound in the green box sports better performance than PK-1. While it doesn't appear among the front runners, this paste is still in the upper mid-range.

Prolimatech PK-3

Prolimatech’s highest-end paste is more expensive than its other two models. Perhaps that yellow packaging is intended to suggest gold? This is a good thermal compound, but it's not the best one out there. It's also thick like the GC-Extreme, a product that the PK-3 can approach, but not quite reach. Warm this paste up before using it.

Pastes: Revoltec, Scythe, SilverStone, And Thermalright

Revoltec Thermal Grease Nano

No, this isn’t nail polish for female hardware enthusiasts; it's a liquid paste that's applied with a brush attached to the lid. Unfortunately, Revoltec's Thermal Grease Nano doesn't test particularly well. While it's easy to apply and remove, we measured lackluster thermal properties. Enthusiasts will want to stay away. At least it is inexpensive.

Scythe (Paste Bundled with Scythe Coolers)

Our tests with bundled thermal compounds continue. While Scythe's OEM manufacturer is unknown, at least the plastic bag has Scythe's name on it. The benchmark results come back average. They're not particularly good or bad.

SilverStone (Paste Bundled with SilverStone Coolers)

This is what you'll find when you pop open a heat sink from SilverStone. Although the syringe resembles the bundled pastes from be quiet! and Corsair, we come away with slightly different thermal results. The benchmark numbers aren't amazing. Rather, they fall into a range we'd consider pretty average. I wouldn't feel compelling to spend extra on aftermarket thermal compound, though doing so would yield better performance.

Thermalright Chill Factor III

For some reason, I only received a two-gram tube of this paste; the standard tube includes four grams. We generated some fairly average results with Chill Factor III as well. Consider this compound suitable for average users and beginners.

Pastes: Xigmatek

Xigmatek PTI-G3606

Xigmatek’s entry-level paste is called PTI-G3606, and it sells at a mid-range price. Measured thermal performance is acceptable and its ease of use is even commendable. At the end of the day, this is a solid product at a fair price for mainstream hardware enthusiasts. It's simply not what we'd recommend for more hardcore overclockers.

Xigmatek PTI-G4512

Xigmatek's premium thermal compound doesn't even cost more than its entry-level offering. But the trade-off for better thermal performance is higher viscosity. Before using it, warm this paste up. Although it's not really for beginners, we do get a good price/performance ratio out of the PTI-G4512 from Xigmatek.

Xigmatek (Paste Bundled with Xigmatek Coolers)

This is another paste bundled with a cooler we picked up off the shelf. In our tests, the freebie fares slightly worse than Xigmatek's low-end aftermarket stuff, but not by much. It's an average compound that you shouldn't feel compelled to shy away from.

Pastes: Xilence And Zalman

Xilence X5

According to our team in Germany, it appears that Xilence went out of business. Not that it matters much to our U.S. audience; the company's ZUB-XPTP.X5 thermal grease wasn't readily available here anyway. At least we have benchmark data for the compound in our charts for anyone overseas with it. In short, Xilence's X5 is suitable for entry-level systems. Just don't use it on an overclocked processor or a PC in Dubai.

Zalman ZM-STG1

Unfortunately, this one turned out to be a dud in our testing. Zalman's ZM-STG1 is easy enough to apply with a brush, but its effectiveness is sub-par. We'd consider it a notch above Revoltec’s competing compound, but not much higher. Overall, our measured price/performance ratio is a disaster.

Zalman ZM-STG2

The ZM-STG2 paste is priced comparably, but much better than the company's ZM-STG1. That doesn't make this offering good, per se. Surprisingly, both pastes share similar ingredient lists. The ZM-STG2 simply isn't watered down with too much silicone, and it remains tougher, even after burn-in.

Pastes: Just For Fun

Toothpaste

I made the mistake of buying a premium toothpaste with blue stripes through it. A cheap white paste would have probably fared better. Yes, you can use this stuff as a thermal compound, though its long-term stability is questionable and we wouldn't recommend it for overclocking. If all of the shops around you are closed, though, and you want to play a little Battlefield 4 on a new machine, you could probably get by for a couple of days.

Denture Adhesive

If you don't brush your teeth but do live with your grandma, denture adhesive could stand in for toothpaste during those same couple of days. Again, this is not a long-term solution. But then again, neither were grandma’s teeth (or yours, if you don't grab some toothpaste). Be careful with this adhesive. After burn-in, the cooler may stick to the heat spreader and prove hard to remove.

Results: Closed-Loop Liquid Cooler; High Mounting Pressure

I already described our test setup in Thermal Paste Comparison, Part One: Applying Grease And More. What follows is a performance chart on the liquid-cooled system employing high mounting pressure.

Results: Air Cooler; High Mounting Pressure

Likewise, we covered this configuration in Thermal Paste Comparison, Part One: Applying Grease And More. The following chart represents thermal performance of an air-cooled system employing high mounting pressure.

Results: CPU Air Cooler; Low Mounting Pressure

One last time, we covered this configuration in Thermal Paste Comparison, Part One: Applying Grease And More. This chart reflects our air-cooled machine with lower mounting pressure.

Results: Air-Cooled GPU; Medium Mounting Pressure

Our last bit of testing comes from applying these thermal compounds to GPUs, and the results are significantly different from what we saw cooling our CPUs.

The failure of some pastes may be explained by rapid hardening, and a second test with two additional cycles didn't change the outcome. What do we learn from those failures? While value-added pastes may work well with the CPU coolers they were bundled with, they may not behave well under the higher temperatures generated by powerful GPUs, and shouldn't be used with them.

It's Not As Easy As Picking A Favorite...

In the end, it doesn't necessarily matter how much you paid for your paste, or what thermal conductivity its manufacturer claims on the package. More important is the how: your application method makes or breaks this process.

Some of the pastes we tested remain classics, and are even relatively inexpensive. It only follows, then, that we'd caution you against spending big on a brand new paste that'll only make a temperature difference of 1 to 2 Kelvin. If that sort of delta plays a significant role in your overclock, you're courting disaster. Sufficient airflow through the enclosure should ensure that you're never dependent on a particularly high-performance thermal compound.

More hardcore overclockers tuning their systems competitively, or quiet computing enthusiasts who spin their fans down as low as possible for better acoustics might want to give more thought to their paste of choice. Even a majority of power users, however, should focus on a paste’s ease of use, leaning toward the softer compounds. A flawed application typically impacts thermal performance much more than the difference between pricey and value-oriented pastes ever could.

I don’t want to lift any one thermal compound up onto a pedestal, and I'm no fan of award inflation. Sure, I could drop a couple of favorites, but the assessment would be highly subjective because some folks weigh ease-of-use and performance differently. Moreover, a lot of these pastes perform a lot alike. We can still draw several conclusions from our work, though.

• Novices should gravitate to softer pastes and use the "lentil ball" application method
• Experts can warm up thicker pastes, making them easier to apply
• Softer pastes should be used with coolers that attach via low mounting pressure
• Pastes bundled with most CPU heat sinks are good enough
• Different rules apply for GPUs; you need to pick the compound you use on your graphics card very carefully
• While the legends spoken of liquid metals have a basis in fact, they are exaggerated

That’s it from my side. But let me also point out that our Thermal Compound Charts will be maintained with results from any new products that surface. What will I do with all the partially-used syringes and paste packs? Honestly, I don’t know. There's enough of this stuff around to last a lifetime...

Inno3D has released a semi-custom version of the GTX Titan, this one sporting Arctic's Accelero Hybrid cooler. It is called the GeForce GTX Titan iChill Black Series. While Nvidia doesn't allow manufacturers to make any changes to the reference PCB design, apparently, manufacturers can still choose to use a different cooler or change the clock speeds.

Arctic's Accelero Hybrid is a cooler that shares the idea from Asus' ARES II card, which has a liquid cooler for the GPU and a fan on the shroud to cool the other components, such as the RAM and the VRM; hence, it's called a hybrid cooler. The unit is also closed-loop and thus maintenance free.

Beyond the custom cooler, the GPU's core speed has been changed. What used to be 837 MHz base and 876 MHz boost is now 837 MHz base and 980 MHz boost. The memory clock speed has remained the same, at 6 GHz, as well as the 2688 CUDA cores, 224 TMUs, 48 ROPs, and the 384-bit memory bus. Since Inno3D can't make any changes to the PCB design, the card also still draws power from a combination of a 6-pin and 8-pin PCIe power connector. Display outputs are also still two DVI ports, HDMI and a DisplayPort.

_{Contact Us for News Tips, Corrections and Feedback}

AMD's Radeon HD 7970: Presenting A Unique Cooling Challenge

The factory-installed GPU coolers used on today’s graphics cards are generally pretty specific. You can't yank the thermal solution from a GeForce GTX 680 and drop it onto a GeForce GTX 660 Ti. In contrast, the companies that design aftermarket coolers try to make them a lot more flexible, accounting for differences in mounting holes, PCB dimensions, and the varying power and memory circuitry found on one card to the next.

Occasionally, AMD or Nvidia modify things even more dramatically, introducing a new design that sends ripples of change through the aftermarket. And that's exactly what happened when the Radeon HD 7970 and 7950 emerged, both of which come with reference coolers that abandon the typically-flat cooling block design. Instead, they use a large, thick shim around the Tahiti GPU, putting it below the surrounding components and relying on a raised contact area for heat transfer away from the graphics processor.

Of course, this cooler design renders AMD's other implementations unusable on the Radeon HD 7900s. It also torpedoes the multitude of aftermarket designs intended to work with as many cards as possible. Most of the companies that manufacture graphics card coolers are used to dealing with dramatic changes, however, and as we wrote this story up, there were already two vendors selling new products to replace the problematic thermal solution covering AMD's Radeon HD 7970: Arctic's Accelero Xtreme 7970 and Deepcool's Dracula 7970. We previewed the Dracula (sold under EKL's name) in Extreme Air Cooling: Our Five-Slot (Quiet) Radeon HD 7970.

In addition, one other manufacturer came up with a way to use a more generic heat sink and fan combination on the Radeon HD 7900-series cards. The EK-VGA Supreme HF HD7970 Cu Adapter from EKWB Cooling Solutions is basically a copper shim that fills the space between a more standard cooling block and the Radeon HD 7970's Tahiti GPU.

Available at frozencpu.com for $4, the adapter is perhaps the least-expensive solution for anyone who already has an aftermarket cooler and wants to re-use it. But how much efficiency do you lose? After all, the adapter is responsible for transferring thermal energy between two surfaces that'd normally be touching, and paste on each.

In order to illustrate the effect of this, we're testing Arctic's Accelero Xtreme III with the copper shim installed and comparing it to the performance of Arctic's Accelero Xtreme 7970. The only functional differences between them are the shape of their cooling blocks and the presence of the shim. 

Let's consider the other coolers we're testing, too:

AMD's reference cooler for the Radeon HD 7970 doesn't need heat pipes; it instead uses a copper vapor chamber with an array of fins attached. A single 75 mm centrifugal fan pulls air in from the chassis and exhausts out the I/O bracket. Technically, this is the approach we prefer to see taken because it doesn't recirculate hot air inside your case, affecting other components. Unfortunately, the resulting temperatures are merely acceptable, and the noise generated in getting there is pretty unbearable under load.

Do you see the raised GPU contact surface on the vapor chamber? This is what makes the Radeon HD 7900-series cooler unique.

Below is a picture of the naked reference card, showing the thick protective shim surrounding AMD's Tahiti GPU.

Arctic Cooling Accelero Xtreme 7970 And III

Aside from the shape of their cooling blocks, the Arctic Accelero Xtreme III and Accelero Xtreme 7970 are essentially identical.

In the shot above, the top cooler is Arctic's generic Accelero Extreme III, flanked by its bundled accessories. The bottom unit is the Accelero Xtreme 7970, surrounded by its complementary hardware. Clearly, the cooling blocks are different.

Both products share similar accessory packages, with a couple of notable changes. The Accelero Extreme III comes with more memory heat sinks to accommodate a wider range of cards, while, for some reason, the Accelero Xtreme 7970 includes a mixed thermal adhesive that requires a catalyst. The fan power cable is designed to plug directly into the graphics card, so the driver software can alter the RPM in response to the thermal load.

A closer look at the 7970-specific cooling block makes the functional difference between these two products even more apparent. Radeon HD 7900-series cards require that the GPU contact area be raised from the rest of the cooler to clear a thick protective shim. The contact area is also turned 45 degrees to match the graphics processor's orientation on the PCB. Both Accelero Xtreme models come with thermal paste pre-applied.

Mounted atop a reference Radeon HD 7970, the card’s dimensions become 12.5 inches long, five inches thick, and 2.25 inches deep. The board weighs in just over two pounds, too.

The assembled graphics card and cooler are just wide enough to chew up more than two expansion slots. AMD's reference cooler already makes the card 11.5-inches long from the factory. And while adding another inch doesn't seem like a big deal, you're going to find that smaller cases simply cannot accommodate the extra length. Enclosures with removable drive bays because particularly attractive here.

The Accelero Xtreme III employs three 92 mm fans and five heat pipes, which are enough to handle up to 300 W of heat dissipating, according to Arctic. Both heat sinks are built on copper cooling blocks and copper heat pipes that run through aluminum fin arrays. Although the fan shrouds are plastic, they're surprisingly sturdy.

Installing the Accelero Xtreme 7970 is easy, despite the perceived difficulty of swapping graphics coolers. First, mix the thermal adhesive and catalyst, apply it to the RAM and VRM heatsinks, and attach them to the card. Next, put the heat sink mounting hardware in the right place on the bracket, and attached that to the card as well.

The Accelero Xtreme III is a little more difficult to install because you have to use EKWB's copper adapter. We found that it was easiest to apply thermal paste to the adapter, place the shim on the GPU, and plop the cooler on top of that.

Arctic's Accelero Xtreme has the same circuit board clearance to the left and right of the GPU. In comparison, Deepcool's Dracula is higher on the right side, and less so on the left. In both cases, though, the vendors require that you glue heat sinks to the voltage regulation circuitry and RAM modules using adhesive, making it difficult to switch them out. As a result, we used Deepcool's low-profile sinks on the memory to the left of the GPU in the name of compatibility.

Here is how the cooler looks installed on a reference card and plugged in to one of our test beds. Because it's so long, the cooler sticks out pretty far beyond the right side of our ATX motherboard.

Deepcool Dracula 7970

The Deepcool Dracula 7970 ($75 MSRP) isn't yet available in North America, although the generic version is a lot easier to find.

The Dracula is atypical for a VGA cooler because Deepcool doesn't give you any fans for that $75 entry fee. Whichever coolers you do choose are mounted to an expansion bracket that sits next to the card, rather than attaching directly to the heat sink. Thermal paste for the GPU and thermal adhesive for the RAM and VRM sinks is included, fortunately.

A raised cooling surface differentiates the Dracula 7970 from Deepcool's generic Dracula. Whereas Arctic's Accelero sports a square-shaped pad for making contact with the GPU, Deepcool uses a circular area.

The Dracula features six heat pipes, or one more than Arctic's Accelero Xtreme. The cooler's copper base is nickel-plated and finished to a mirror shine. Copper heat pipes and the aluminum find are nickel-plated as well, resulting in a very shiny sink.

Deepcool's Dracula can accommodate two 140 mm fans or three 120 mm fans. However, we used the two 92 mm fans that came with the company's older V6000, which are rated for 66 CFM of airflow each. We don't anticipate any problems keeping the Radeon HD 7970's Tahiti GPU cool. Driving both fans with power supply leads ensures constant voltage, and, consequently, constant rotational speeds, regardless of load.

As with the Accelero, mounting Deepcool's Dracula 7970 is simple enough once you have AMD's reference cooler disassembled. Thermal adhesive is applied to the RAM and VRM heat sinks, and the cooler attaches directly to the circuit board using existing mounting holes.

The Dracula's heat pipes and cooling fins provide for plenty of clearance on the right side of the GPU. The left side is a lot less roomy due to a tighter heat pipe bend, necessitating lower-profile RAM sinks.

With Deepcool's Dracula mounted to AMD's reference Radeon HD 7970, the assembled card is 11.5 inches long, 4.75 inches wide, 1.75 inches deep, and it weighs just under two pounds. Before jumping to the conclusion that this solution is more compact and lighter than Arctic's competition, remember that the fans and mounting bracket are separate, and we still haven't added them.

Each configuration is going to be different since Deepcool makes you pick your own fans. In our case, the additional hardware measures 11" x 6.5" x 2.75", and adds another 13.8 ounces. With the heat sink and fans combined, the Dracula chews up considerable expansion space, likely killing the potential for CrossFire. On the other hand, the Dracula is about the same length as AMD's stock Radeon HD 7970 when it's installed, so it's more likely to fit horizontally.

Test System Setup And Benchmarks

Our testing includes idle and load temperature measurements, in addition to acoustic benchmarks. We're using Battlefield 3 as our graphics load,  with the game running under the Ultra quality preset at 1920x1080 for 10 minutes in the Operation Swordbreaker level. We log temperatures using GPU-Z, and measure noise two feet from the back of the case with a CM-130 SPL meter. Unfortunately, ambient noise is a challenge in our real-world test lab; we recorded 40.5 dB(A).

While the Arctic Accelero Xtreme fans are powered by headers on the graphics card and should vary rotational speed in response to thermal load, Deepcool's Dracula employs power supply or motherboard fan headers. We're using 12 V leads on the motherboard to drive constant RPMs.

We also plan to measure how these coolers handle a GPU overclock to 1100 MHz with a 1.2 V setting. This increase should generate significantly more heat than the stock settings.

Benchmark Results: Temperature And Noise

We start with thermal testing to compare AMD's reference Radeon HD 7970 to Arctic's Accelero Xtreme III with EKWB's copper shim, Arctic's Accelero Xtreme 7970, and Deepcool's Dracula 7970.

The Dracula 7970 and Accelero Xtreme 7970 perform almost identically, posting excellent sub-60-degree results under load. When you consider the reference cooler approaches 80 degrees, the improvement is quite impressive.

As we might have expected, adding a shim between the heat sink and GPU takes a toll on efficiency, yielding a mere eight-degree improvement over AMD's vapor chamber, and trailing the Accelero Xtreme 7970 by 14 degrees.

All of the aftermarket coolers generate comparable acoustic results. This is something we would expect from the Dracula, since its power source doesn't change (hence, there is no fan ramp). Nevertheless, the three models perform well enough to shine a spotlight on the reference design's weakness: noise. 

In contrast, we would have thought that Artic's cooler would have demonstrated more variable measurements. We are able to force a 100% duty cycle using AMD's driver, and we know it's possible to spin the Accelero's fans up because we can hear the difference. Thus, we can conclude that the fan profile on the Radeon HD 7970 doesn't impose significantly higher fan speeds until the GPU temperature rises above 68 degrees.

Benchmark Results: Overclocking

Next, we increase the clock rate of our card's core to 1100 MHz and increase its voltage to 1.2 V.

The charts are the same, except for the black bar, which illustrates the effect of overclocking on each solution's thermal performance. Arctic's Accelero Xtreme 7970 pulls away from Deepcool's Dracula here, but not by a wide margin. Also, remember that the Dracula can accept larger fans if you're willing to spend more money.

The Acceleo Xtreme III with its copper adapter gets even closer to the reference design, though it still performs slightly better at idle and under load.

AMD's reference cooler becomes even less attractive as the Tahiti core nears 80 degrees. More problematic, though, is how fast the centrifugal fan needs to spin in order to keep the GPU at 79 degrees.

We again see our aftermarket coolers generating low and constant noise levels. In contrast, the reference heat sink and fan combination jumps to an even more unacceptably loud volume under load.

High Performance Cooling Options For The Radeon HD 7970

A new vapor chamber design from AMD threw a curve ball at aftermarket cooler vendors. Fortunately, a handful of innovative companies made the changes necessary to counter some of what we think AMD screwed up with its reference Radeon HD 7970 heat sink and fan combination. Arctic, Deepcool, and EKWB are the first three to get hardware into our lab.

Arctic Cooling Accelero Xtreme 7970

Arctic hit a home run with its Accelero Xtreme, a cooler design that we first saw back in 2008. The Accelero Xtreme III is the newest iteration, though the company rebranded it the Accelero Xtreme 7970 with a cooling block modified to fit AMD's Radeon HD 7900-series cards.

The Accelero Xtreme 7970 operates both quietly and coolly. Its fan headers plug right into the graphics card for convenience and control, and the combination effectively brings down temperatures even in the face of aggressive overclocking. An $80 price tag is certainly appropriate in light of what the competition is asking. Our sole concern is that a fully built-up reference card with this cooler is 12.5" long, limiting its use in certain enclosures.

Aside from that one issue, Arctic's Accelero Xtreme 7970 is an excellent way to combat the reference Radeon HD 7970's loud centrifugal fan and vapor chamber-based heat sink. Perhaps more significant is the fact that it seems to be the only natively-compatible Radeon HD 7900-series aftermarket cooling option available for purchase in North America right now.

Deepcool Dracula 7970

Deepcool's generic Dracula is already available in North America. However, its Radeon HD 7970-compatible version is still en route, according to the company. If it were available today, the Dracula's $75 MSRP would be about five dollars less than the Accelero Xtreme 7970. Without fans, though, it quickly gets more expensive.

The Dracula 7970 is a powerful cooler for the Radeon HD 7970. Though it performed slightly below Arctic's Accelero Xtreme 7970 under load, remember that we armed it with a pair of 92 mm fans. The heat sink accommodates as many as two 140 mm or three 120 mm fans, so it's possible to improve this product's cooling potential if you're willing to spend the money. Deepcool's solution also measures 11.5" long, a full inch shorter than the Accelero Xtreme. If you have a space-constrained chassis, that inch could be critical.

On the other hand, the Dracula is very wide once you factor in its cooling fan bracket. It's conceivable to fit one of these inside an enthusiast-oriented enclosure, but two in CrossFire probably isn't going to happen. And with the price of fans added to its total, going the Deepcool route can get pretty expensive.

EK-VGA Supreme HF HD7970 Cu Adapter

EKWB's copper shim makes it possible to use more common coolers with flat contact surfaces and still maintain compatibility with AMD's Radeon HD 7970. Priced at $4 from frozencpu.com, this little adapter is an inexpensive way to help recycle a cooler you have on-hand without needing to drop $80 on something new.

Unfortunately, going that route imposes lower performance than a purpose-built heat sink for AMD's flagship graphics card. Transferring heat through an additional medium, plus two layers of thermal grease, results in GPU temperatures 10 degrees higher under load compared to Arctic's Accelero Xtreme 7970, which we'd consider a native solution.

Making an investment in a high-performance cooler like the Accelero should yield much better thermal and acoustic performance than a reference heat sink and fan combo. Otherwise, why buy it? The EKWB copper shim does the job it's supposed to. However, realizing a moderate gain required a high-end cooler like the Accelero III. A less-capable aftermarket heat sink wouldn't have fared as well, and we would have ended up even closer to what AMD's reference vapor chamber-based implementation was already able to do.

Ivy Bridge Overclocking: What Does It Entail?

Lower power consumption, purportedly less heat dissipation, a smaller die size, lower manufacturing costs for Intel...but does the 22 nm Ivy Bridge design also leave less room for mainstream overclocking? Our launch coverage (Intel Core i7-3770K Review: A Small Step Up For Ivy Bridge) revealed that overclocking the new processor design wasn't any more fruitful than the already-mature 32 nm Sandy Bridge-based Core i7-2700K flagship. Although stock temperatures were lower, they ramped up very quickly once we started applying the voltages thought necessary to approach 5 GHz on air.

Overclocking: What’s Required?

The length of time a transistor in a digital circuit delays an electronic signal depends on its size, fabrication technology, layout, temperature, and operating voltage. The highest achievable clock rate of a circuit depends on this delay and the number of logic levels that a signal has to traverse in a single clock period. The latter number is fixed (and dependent on the processor's architecture). So, for overclocking, we focus on how a transistor's latency is affected by its supply voltage. A higher supply voltage can shorten the delay, but will also raise the transistor’s power consumption. Cranking up the clock frequency also increases the dynamic power draw per time unit, and thus further raise the circuit’s power consumption, leading to a hotter chip.

Both effects, taken together, explain why overclocking at an elevated CPU voltage increases the power draw and heat output, and why cooling an overclocked CPU can quickly become challenging. As in sports or any engineering discipline, trying to eke out that last couple of percentage points is most difficult.

CPU manufacturers have put in some safeguards against reckless overocking by inexperienced users (and unscrupulous system builders); starting a few years ago, both AMD and Intel started shipping most of their models with locked multipliers, releasing more advanced models opened up with overclocking in mind. Of course, enthusiasts know that they can either tweak the multiplier through their BIOS or through a Windows-based utility provided by many vendors for easier access to those settings.

In case of Intel's unlocked Ivy Bridge-based K-series SKUs, the highest CPU multiplier was increased to 63x (from Sandy Bridge's 57x ceiling), translating to a theoretical 6.3 GHz limit if you don't touch the 100 MHz BCLK. Going higher requires changing the base clock, which is rather difficult. Above a 110 MHz threshold, very few systems are stable. Be that as it may, it's going to take more than conventional cooling to hit those clock rates. In reality, you'll only see the limits of these architectures pushed in overclocking contests and YouTube videos.

Overclocking: Expectations

In the past, shrinking gate lengths have been seen to increase overclocking headroom. Smaller transistors require less voltage and consume less power, generally leading to better overclocking margins. Intel's Sandy Bridge-based K-series models easily achieved 4.3 to 4.6 GHz using air coolers, sometimes scaling even higher. Thus, our expectation for Ivy Bridge (along with many other enthusiasts, we'd say), was closer to 5 GHz.

However, we failed to achieve that goal, despite multiple tests in multiple countries using multiple Ivy Bridge-based samples. But we also received reports that Intel's 22 nm chips can break through speed records if you overcome their rapid heat-ramp using extreme measures like liquid nitrogen.

Knowing that LN₂ is impractical in a production environment, we set out to achieve the highest overclock possible using conventional air cooling, discussing the causes of Ivy Bridge's limitations along the way.

Overclocking Ivy Bridge: Treating This Hot-Head Gingerly

Our Core i7-3770K and Core i7-2600K samples ship at very similar clock rates. The Ivy Bridge-based CPU runs at a nominal frequency of 3.5 GHz and can accelerate up to 3.9 GHz with a single core active by virtue of Turbo Boost, providing ample thermal headroom. These clock rates match Intel's Core i7-2700K, flagship of the mainstream Sandy Bridge family. Unfortunately, our only -2700K is in the U.S. So, our German team used a Core i7-2600K it had on hand, which is clocked at 3.4 GHz and capable of reaching 3.8 GHz thanks to Turbo Boost.

Initial Overclocking Successes

This hasn't been talked about a lot, but every overclocker has to appreciate the fact that Intel allows Ivy Bridge-based K-series parts to change their multipliers during operation. It's no longer necessary to reboot between modifications. We've been doing this with our AMD chips for a while now, so kudos to Intel for catching up. Using Intel’s Extreme Tuning Utility, a well-designed utility, we're able to tune our Core i7-3770K from within Windows.

We easily pushed the -3770K above 4 GHz without breaking a sweat. In fact, we hit 4.6 GHz quite easily. There, we encountered our first instabilities, which we tried to counter by raising the core voltage.

Frustration at High Clock Rates

As we proceeded in our overclocking efforts, regardless of whether we used a higher core voltage or not, we observed something frustrating: even below 4.5 GHz, our Ivy Bridge-based Core i7-3770K began thermal throttling. That is to say it reduced its clock rate in order to bring its temperature down. In other words, our overclocked -3770K was already running too hot, even at its default voltage setting.

Core Temp 1.0 RC3 reports that our Core i7-3770K reaches 90-100°C (194-212°F) internally when it's overclocked to 4.5 GHz. No wonder the chip's thermal monitor tripped, throttling the CPU. This phenomenon dropped the effective clock rate of our chip to approximately 3.5 GHz, corresponding to the CPU’s nominal frequency.

For comparison purposes, let’s look at the Core Temp readouts of Sandy Bridge- and Sandy Bridge-E-based processors:

Our 32 nm Sandy Bridge-based Core i7-2600K also got quite hot on the same test rig. However, each core stayed well below 90°C (194°F). As a result, the system maintained its overclocked frequency of almost 4.8 GHz. That's 300 MHz more than the new Ivy Bridge-based CPU!

Even our six-core Core i7-3960X (Sandy Bridge-E, the one with more than 2.2 billion transistors) posted better temperature readings. Each and every one of the six cores stayed at or below 81°C (176°F) at a very impressive clock rate of 4.7 GHz.

We want to share several observations that will help explain what’s going on.

More Voltage, More Heat

Our evidence shows that, at similar clock rates, 32 nm Sandy Bridge-based processors don't get as hot as the new 22 nm Ivy Bridge-based chips. As a result, our Core i7-3770K dropped its overclocked frequency sooner in order to protect the CPU's integrity, nullifying the effects of our overclocking efforts entirely. It also discourages setting higher core voltages, since that'd just exacerbate the issue.

Thus, throttling can only be avoided when the target overclock is achieved using the lowest core voltage possible. However, it still has to be high enough to allow the transistors to operate reliably. Old mantras like “a hefty voltage increase helps facilitate a hefty frequency bump” do not work on Ivy Bridge-based chips. Once throttling kicks in, average clock rates drop to around 3.6 GHz. That's a significant reduction compared to the 4.6 GHz we were targeting.

We think that you're going to have the most success overclocking Ivy Bridge if you can stay as close as possible to the CPU's default voltage setting or work really hard to overcome the heat resulting from more aggressive voltage settings. At 4.6 GHz, the voltage increase necessary to get our machine stable quickly resulted in throttling, forcing us to stay at 4.5 GHz. At that speed, we had to increase the core voltage by 70 mV in order for our benchmark suite to run reliably.

As in our launch coverage, 4.5 GHz turned out to be as high as we could go without running into stability issues. It wasn't difficult to get up to 4.7 GHz (or, using one of our samples, even 4.9 GHz). But certain tests consistently cause those frequencies to derail, pushing us back down.

Digging Into Ivy Bridge's Overclocking Issues

Why do Ivy Bridge processors behave this way? How come 32 nm Sandy Bridge-based processors stay cooler, even at higher clock rates and voltage settings? Is this Intel’s Ford Edsel moment? We don't necessarily think so, but let's take a look at the factors that cause us to be concerned.

More Heat Output Per Mm²

The Ivy Bridge-based Core i7-3770K sports 1.4 billion transistors on a die area of 160 mm². The Sandy Bridge die measures 216 mm² and contains 995 million transistors. While we're comparing, we should also keep in mind that the HD Graphics 4000 engine occupies about one-third of the largest Ivy Bridge die. On Sandy Bridge, HD Graphics 3000 is closer to one-fourth of the package. In total, the CPU's area shrank by roughly 40%.

Now, let's approximate the surface area consumed by everything except graphics resources:

• Ivy Bridge: ~105 mm²
• Sandy Bridge: ~160 mm²

The Sandy Bridge-based Core i7-2600K has a TDP of 95 W, whereas the Ivy Bridge-based Core i7-3770K’s TDP is 77 W. Suddenly, it becomes a clearer that overclocking causes the smaller Ivy Bridge die to emit as much heat per square millimeter as Sandy Bridge. This naturally affects how much higher than Sandy Bridge we should have expected Ivy Bridge to scale.

New Transistor Technology

Another variable we've seen presented are the new tri-gate transistors. Intel claims that, in a typical use case, they use up to 50% less energy compared to a traditional transistor, as their three-dimensional structure of one horizontal gate together with two vertical ones effectively triples the field effect of the gate, which should reduce leakage current drastically.

We want to emphasize the phrase “typical use case” here because the more you overclock, the further away from typical you get. While we can confirm the claimed power reduction for typical use cases (just wait until we get to the benchmarks), at this point, it is not entirely clear how these transistors perform at substantially increased clock rates. Maybe they're just not optimized for the speeds an overclocked chip pushes yet. It might be necessary to wait for Ivy Bridge's successor, Haswell, to see if the limits of 22 nm manufacturing can be pushed harder.

Thermal Engineering

Increased density and new transistor technology can certainly give rise to more heat. But we've seen these things before, and each time, cooling and packaging technology manage to cope with the resulting output. What could be holding Ivy Bridge back? It could be the fact that Intel decided to use thermal paste instead of the usual fluxless solder (you'll need to have a solid grasp of Japanese to understand the original article, or use Google Translate) between the CPU die and heat spreader.

Using a box knife (we do not recommend doing this at home), the author of the linked story pried the heat spreader away from the chip and was able to replace the cheap paste Intel used, trying out both OCZ Freeze Extreme and Coollaboratory Liquid Pro thermal pastes. The OCZ offering allowed for 1.55 V at 4.9 GHz, while the Coollaboratory material ramped up to 5.0 GHz, operating stably. This was accomplished even with air cooling, although the author did not use a stock heat sink, opting for a Thermalright Silver Arrow SB-E instead (Ed.: this information was pulled from the original Impress PC Watch site with Google's rather shaky translation of Japanese). If there is a smoking gun in this equation, we think this is it, especially considering that the researchers at Impress PC Watch managed 20% more efficient cooling.

Practical Advice: Sandy Or Ivy Bridge?

We've now illustrated that Ivy Bridge dissipates its heat in a much smaller die than Sandy Bridge, and then uses a less effective mechanism for transferring it away from the die and out to a heat spreader. Once the overclocked processor's four cores are saturated, the temperature increase happens so fast that the CPU's thermal monitor triggers throttling faster than we could take and save a screen shot of Core Temp. The jump was phenomenal, taking less than a second from idle to throttling temperature. In the end, we had to use a script to take the shot.

Getting Rid Of Excess Heat

Our experience highlights one of the obstacles that prevents higher clock rates on Ivy Bridge-based CPUs: the cooling subsystem must be able to operate effectively and without any delay. On air, the throttling mechanism triggers before a cooling fan can spin up. We didn't have the luxury of risking the destruction of our test chips by prying their heat spreaders off, and we don't recommend that drastic step to anyone, really. So, we're recommending a closed-loop liquid cooling setup, at least. More extreme enthusiasts can pick a more serious cooling technology, of course.

Recommendations

We state without any hesitation: air-cooled Ivy Bridge-based processors cannot be overclocked as much as Sandy Bridge-based processors. Overclockers hunting for the latest and greatest overclockable processor, yearning for high-frequency overclocks, should keep that in mind. Maybe a Sandy Bridge-based chip is still the best choice, even in a world where Ivy Bridge exists.

If scalability isn't as big of a worry for you, Ivy Bridge is the more natural choice. Its performance per clock is a few percent higher, so long as you're looking at the same frequency from both architectures. After all, a 4.5 GHz Ivy Bridge-based CPU wins benchmarks against a Sandy Bridge processor at slightly higher clock rates. When you limit your overclock of a chip like the Core i7-3770K to 4.2 or 4.3 GHz, you’re completely on the safe side. There is no temperature issue, and performance remains impressive. Then again, such a system won’t be significantly faster than a machine running at its stock clocks.

Test System And Benchmarks

For our test, we tuned each system to the maximum clock frequency it'd support without sacrificing stability. We used a Freezer 13 air cooler from Arctic Cooling. For the sake of a fair comparison, we used DDR3-1600 memory for all systems. While faster RAM could have resulted in a percentage point gain here and there, it does not change the relative performance of overclocked CPUs that much.

Benchmark Results: Professional Applications

3ds Max utilizes as many cores as it can. And although Ivy Bridge gives us slight IPC improvements over Sandy Bridge, the older architecture's more willing overclock ends up delivering better performance. Meanwhile, even though Sandy Bridge-E is running a little slower, its two extra cores enable a nice performance advantage.

The same story applies to FineReader, our OCR-based workload. Ivy Bridge's 4.5 GHz just can't keep up with a Core i7-2600K at 4.8 GHz. The Core i7-3960X trumps all, though.

Same thing in Blender. Ivy Bridge ends up in last place compared to the two other overclocked platforms.

Benchmark Results: Adobe CS 5.5

Everything we saw on the preceding page applies here. Sandy Bridge-E universally takes first place, and is trailed by Sandy Bridge by varying degrees, depending on how well-threaded the application is. In none of the Adobe applications does quad-core Ivy Bridge at 4.5 GHz outperform the 4.8 GHz Sandy Bridge-based processor.

Benchmark Results: Audio/Video

Converting audio files with iTunes, a single-threaded test, benefits from Ivy Bridge’s architectural improvements. Finally, the 4.5 GHz Core i7-3770K processor improves upon the 4.8 GHz Core i7-2600K's score.

Lame is also single-threaded, but it doesn't beat Sandy Bridge running 300 MHz faster. The difference is small enough that we'd call it a draw across the board, and if you already own a fast Core i7, an Ivy Bridge-based one certainly isn't going to help.

Our overclocked Core i7-3770K achieves an almost unnoticeable three-second win over the Core i7-2600K. But neither quad-core CPU even comes close to the Core i7-3960X.

In MainConcept, the benchmark shows another near-tie between Sandy Bridge at 4.8 GHz and Ivy Bridge at 4.5.

Benchmark Results: Matlab

In Matlab, the differences between overclocked CPUs are hardly noticeable.

Benchmark Results: File Compression And Power Consumption

7-Zip makes good use of additional cores and improvements to IPC. Thus, Ivy Bridge at 4.5 GHz scrapes out a win against the 4.8 GHz Sandy Bridge-based chip.

WinRAR benefits from high clock rates and extra cores, so Ivy Bridge finishes last, despite its better performance per clock.

Power Draw

Even overclocked, the Core i7-3770K is one of the most energy-friendly enthusiast-oriented CPUs we've seen. Despite its high performance, the whole system’s 52 W idle power draw, which includes a Radeon HD 6850 graphics card, is very impressive.

At maximum processor load, Intel's Ivy Bridge architecture struggles for victories in our performance metrics because it's forced to run 300 MHz slower than last generation's Sandy Bridge design. For the most part, though, improvements to performance per clock keep the two CPUs fairly even.

As we can see, though, Ivy Bridge's power draw is a good deal lower. Naturally, that bodes well for Ivy Bridge's outcome in our efficiency measurement. The Core i7-3960X clearly demonstrates superior performance, but it does so at the cost of 68% higher power draw. It's not efficient, clearly. That's the price for maximum performance, though.

Single- And Multi-Threaded Efficiency

Single-Threaded Applications

The total run time for all single-threaded applications doesn’t differ by a lot. In the end, though, Sandy Bridge's higher clock rate secures the victory. The power draw tests, however, favor Ivy Bridge.

Multi-Threaded Applications

Thread-optimized applications show us a similar picture. However, the Core i7-3960X and its six-core architecture exhibits dominant performance. Factoring power into the equation, though, gives Ivy Bridge another clear win.

Overall Efficiency

As the diagram shows, Intel's 4.5 GHz Core i7-3770K beats the two competing platforms in our efficiency test, which consists of almost all the applications used in this article, by using the fewest Wh (watt-hours) of energy.

The Core i7-3770K's elapsed run time is the longest by just a hair, since the Sandy Bridge chip runs at higher clock rates and the Sandy Bridge-E processor boasts more cores. But Ivy Bridge’s clear win in the efficiency department outweighs its small performance disadvantage.

The verdict is in: under load, Ivy Bridge leads in performance per Wh. An overclocked Core i7-3770K is the new enthusiast-oriented efficiency champ.

The efficiency diagram documents our findings in visual form.

Ivy Bridge Takes A Bronze In Overclocking; Gold In Efficiency

We're not going to mince words here. The new Ivy Bridge-based processors manufactured at 22 nm serve up high performance and moderate power consumption. They can be overclocked to respectable levels. With that said, though, Intel's preceding generations helped facilitate impressive overclocking, and our expectations for this generation were higher. The first retail CPUs employing the Ivy Bridge architecture cannot be overclocked as well as their predecessors on conventional air cooling. However, the difference in attainable clock rate translates to similar performance when you compare to the fastest Sandy Bridge-based chips.

In Practice There Are Clear Frequency Limits

The small die size of Ivy Bridge turns into a mixed blessing. Packing four cores, a more complex GPU, and 8 MB L3 cache into a mere 160 mm², the new chip occupies 26% less area than a comparable Sandy Bridge CPU. Disregarding the GPU, which is bigger now than, the CPU shrinks by approximately 40%.

Intel counters the smaller die by dropping Ivy Bridge's thermal ceiling. Overclocked, though, Ivy Bridge must dissipate close to the same heat as a Sandy Bridge-based CPU, but across less surface area. It is pretty clear that Intel's decision to use thermal paste instead of thermally-conductive solder puts a vice on the frequencies this CPU can stably achieve. The end result is a more pronounced and rapid thermal ramp-up that can only be effectively addressed using more aggressive cooling solutions. In a nutshell, for an average system builder, there are clear limits to overclocking Ivy Bridge. We will have to wait and see if future products from Intel employ better material between the processor die and heat spreader.

Clear Real-World Advantages

With all things considered, despite its limited overclocking potential, Core i7-3770K is only slightly slower than the 32 nm Core i7-2600K with both CPUs pushed as far as they can go. In practice, the difference isn't noticeably, though.

At idle and under load, though, the Ivy Bridge-based chip draws significantly less power. Power users satisfied with frequencies under 4.5 GHz can get great performance with low power use from a Core i7-3770K. But the recommendation we made in our launch story stands: this isn't worth an upgrade if you followed our advice last year and bought a fast Sandy Bridge-based chip. Really, it's only the right move for folks stuck on platforms more than a generation old.

LGA 2011: No Boxed Solution, Pick Your Own Instead

It seems strange to us that Intel no longer includes a cooler with its fastest retail CPUs. This is, after all, the same company that seems to insist that we include our original boxed cooler any time we send in one of our purchased processors in for a warranty exchange. Instead, it now suggests that enthusiasts spend extra money on something that resembles Antec’s Kühler H20 620, but actually costs as much as the larger Kühler H20 920: Intel's own BXRTS2011LQ sealed liquid CPU cooling system.

The rationale behind this move does make sense in that the Core i7-3960X and Core i7-3930K are multiplier-unlocked products that specifically target overclocking enthusiasts. Enthusiasts typically won’t settle for anything as underpowered as its RTS2011AC downdraft cooler, and Intel doesn’t think hanging two pounds of copper and aluminum up to six inches from the surface of a motherboard is a great idea. And while that kind of leverage tugging on your expensive motherboard is probably ill-advised under the harsh conditions of shipping a machine cross-country, systems that are handled gently and don't get moved often tolerate the weight of big, heavy coolers without a problem.

We invited every major CPU cooler manufacturer (that's right, all of them) to show off their LGA 2011-compatible heat sinks, and fourteen responded with products ranging from $30 to $99. That takes either a great deal of confidence or a great leap of faith from the budget cooler companies, since everyone participating in this piece knew that we'd be basing our evaluation on overclocking. The wide price range allows us to really test how much we need to spend on cooling, and how much we can benefit from spending more money than necessary.

Following up on our earlier picture story, here’s a brief overview of each cooler’s basic features, grouped by price first and arranged alphabetically. Please pay special attention to each cooler’s base height, which is measured with the intake fan at its highest position, along with thickness and offset. If the fan overhangs your DIMM slots, its base height (plus around 0.3" for the CPU/LGA package) represents your maximum supported DIMM height.

.

Since higher fan speeds typically increase cooling capacity at low cost, some of the more expensive models attempt to provide the best balance of cooling and noise. Today’s tests includes both heat and noise measurements, and we’ll even compare heat to noise, heat to price, and heat/noise to price!

Installing Akasa's Venom Voodoo

With glory shots covered in our previous picture story, today’s article focuses on installation, testing, and results. We begin with Akasa’s sharp-looking dual-fan Venom Voodoo.

Akasa’s installation kit covers everything from AMD’s Socket AM2 to Intel’s latest LGA 2011, with screw-in standoffs separating today’s installation from the also-possible LGA 1366.

All the base plates go back in the box thanks to LGA 2011’s integrated cooler support, into which we threaded Akasa’s custom cooler standoffs.

The Intel-style top bracket is still required, and attached to the top of the Venom Voodoo’s base using four screws from the bottom.

We like simple installations, and were pleased to find that the cooler drops straight on at this point. The heat sink’s concave side should face the cooler’s intake fan, while the exhaust fan clips to the other side.

Yellow and black signify danger, which is why it’s frequently used by both bees and construction crews. In this case, the danger might be to your competitors as you show off the power of your overclocked Sandy Bridge-E-based PC!.

Installing Arctic Cooling's Freezer i30

Arctic Cooling’s extremely simplified installation kit eliminates any LGA 1366 support plate in favor of LGA 2011-specific screws.

This could have even been considered a Sandy Bridge-only package if not for the fact that the included LGA 1155 hardware also fits LGA 1156. Even still, we're ignoring both of those interfaces in today’s installation.

Metal spacers fit over the LGA 2011’s integrated screw posts (foreground) before attaching each of the Freezer i30’s two adapter brackets (background).

With its fan removed, the Freezer i30’s integrated cross bracket screws onto its adapter brackets.

Plugging in the fan’s motherboard lead and clipping it to the heat sink completes Arctic Cooling’s Freezer i30 installation.

Installing Cooler Master's Hyper 212 Evo

Cooler Master didn’t have a retail package available when it sent this Hyper 212 Evo sample, but instead included just the cooler, an LGA 2011 installation bracket, and a small tube of thermal paste.

The bracket folds to allow easy insertion between the heat sink’s base and fins.

The unfolded bracket simply screws into Intel’s integrated LGA 2011 support plate. A steel pin in the base’s aluminum top plate engages groves on the bracket to keep the cooler from twisting.

Plugging in the fan and clipping it to the cooler completes the installation of Cooler Master’s Hyper 212 Evo.

Installing Coolink's Corator DS

The Corator DS we received is an LGA 2011-specific model, containing nothing more than a set of nuts and standoffs to connect its universal-pattern Intel LGA brackets.

Standoffs screw into the LGA 2011 interface's included threads (foreground), and the cross brackets are then screwed onto them (background).

The Corator DS’ factory-mounted bracket then screws onto the cross brackets from the kit. What appear to be spring-loaded screws are actually threaded tubes, with a hexagonal cap and a #2 Phillips socket atop each.

The Corator DS’ single fan clips between its twin radiators, and plugging it into the motherboard power header completes the installation process.

Installing Corsair's Air Series A70

Corsair’s installation kit resembles Akasa’s, but with two key differences. First, a clip-style bracket is used for AMD interface compatibility.

The second difference is far more pertinent to our installation: there are no special standoffs for use with Intel’s LGA 2011 integrated mounting plate. Instead, Corsair relies on a little-discussed feature of the LGA 2011 design to support this product:

LGA 2011 motherboards still have LGA 1366 mounting holes. These holes are covered by the factory-installed gasket, but we can see the gasket peeking through the holes in the photo above. These holes are likely intended to prevent damage from overly-long mounting screws when using the integrated support plate.

Corsair told us to simply pop the bracket’s hold-down screws through the gasket, but these screws weren’t sharp enough to go through easily. We instead poked slits in the LGA 2011 gasket using a pocket knife, and inserted the mounting screws through those slits.

The screws now stand atop the LGA 2011 interface the same way they did on our LGA 1366 motherboards.

As with Akasa, Corsair’s mounting bracket bolts atop the A70’s base using screws from its underside.

Nuts apply tension between the cooler’s bracket and support plate using those protruding screws.

Plugging the fans into their motherboard headers and clipping them onto the heat sink completes the A70's installation.

Installing Deepcool's (Gamerstorm) Assassin

Deepcool's Gamerstorm brand brings us the Assassin, which is also distributed by Logisys. Although you can find this cooler by searching any of those names, we found the deepcool-us.com server more reliable than gamerstorm.cn.

The Assassin installation kit includes a support plate for Intel’s LGA 1155, 1156, and 1366, plus AMD’s Sockets AM2 and AM3. LGA 2011 users can throw the big plate back into the box, though, as separate screws and spacers allow the upper brackets to be screwed directly to their motherboard’s integrated support plate.

The split plastic spacers are too thick to fit properly, but a little wiggling achieves the approximate position (foreground). Each cross-bracket is then screwed onto the spacers (background).

The Assassin heat sink’s hold-down is then attached to the cross brackets using spring-loaded screws. This process is slightly more cumbersome than competing solutions because the hold-down bracket isn’t attached to the cooler.

Plugging the fans into motherboard headers and clipping them onto the sink completes the Gamerstorm Assassin installation.

Installing Enermax's ETS-T40

Enermax's ETS-T40 fits Intel’s four-bolt LGA interfaces and replaces AMD’s four-bolt brackets. Four special standoffs add LGA 2011 to the support list.

Either set of cross braces (background) can be fit atop the LGA 2011 standoffs (foreground). Threaded studs on both sets of brackets engage clamping bolts for the top bracket.

Two pins in the top bracket position it precisely atop the ETS-T40’s base. We found that the two nuts that tighten this bracket cannot be completely tightened, though the space between it and the cross brackets is minimal.

Plugging the fan’s PWM connector into the motherboard’s header and clipping it onto the cooler’s face completes the installation of Enermax’s ETS-T40.

Installing Gelid's GX-7

Gelid had not yet added the LGA 2011 installation kit to its GX-7 at the time we asked for review samples, but instead sent its $5 accessory kit for our installation.

Two brackets attach to the GX-7’s base using included screws.

Spring-loaded screws then apply even tension to the brackets, using LGA 2011’s integrated cooler support plate.

Connecting the fan to a motherboard and clipping it to the heatsink completes the Gelid GX-7 installation.

Installing Noctua's NH-D14

The SE2011 (Special Edition 2011) version of Noctua’s NH-D14 contains only the parts need to support Intel’s latest processor interface.

Universal-pattern Intel brackets mate to LGA 2011-exlusive studs using included nuts.

Spring-loaded nuts assure even pressure as the NH-D14’s integrated clamping bracket screws onto each bracket’s studs.

Connecting the fans to our motherboard’s PWM headers and clipping them to the sink completes the installation of Noctua’s NH-D14 SE2011.

Installing Phantek's PH-TC14PE

Phanteks adds a diverse installation kit for Intel’s older LGAs and AMD’s four-bolt sockets, but LGA 2011 installation is made far easier through the use of its integrated support plate.

Four standoffs (foreground) space two cross brackets (background) above the socket, with threaded studs on each cross bracket made to engage the cooler’s hold-down strap.

Bolting the PH-TC14PE’s hold-down strap to the base keeps it from flopping around when installing the heat sink. Spring-loaded nuts at the end of that strap provide even pressure across the base as they're screwed onto cross brackets.

Plugging the fan leads into motherboard headers and clipping them onto the radiators completes the installation process for Phantek’s PH-TC14PE.

Note that both fans are 140 mm (5.5 inches) and can be clipped at several heights. DIMM clearance required us to raise the fan around ½” from center. Depending on the height of your DIMMs, this option may extend the fan beyond the closing point of your case’s side panel.

Also note that the fan clips touch metal components on the back of our video card. We tilted the card slightly to eliminate potential short circuits, but you might not have this option in a chassis.

Installing SilenX's EFZ-120HA5

SilenX's installation kit includes a top bracket that fits all popular desktop CPU sockets (LGA 775 to 2011, AMD Socket 939 to AM3+), a base plate that supports most popular performance-oriented processor interfaces (except for LGA 2011), and an extra set of mounting screws for LGA 2011’s integrated cooler support plate.

SilenX's universal top bracket screws to the sides of the cooler with its ears pointing towards the bottom, as shown.

The EFZ-120HA5’s top bracket is then attached directly to the LGA 2011 cooler support holes using spring-loaded screws. This is the point where everyone who doesn’t have LGA 2011 can start crying about all of the extra steps they're forced to endure.

The most difficult-to-install part of the EVA-120HA5 kit is its fan. First, four T-shaped rubber tacks must be pulled through the fan's mounting holes (on its back side). The top of the T must then be slid into groves on the heat sink.

Unfortunately, the fan’s hole spacing is slightly wider than the heat sink’s height, leaving each T dangling precariously, as only half of its length is supported by fins. The Ts are soft, the fins are rough, so the lightweight fan isn’t likely to shift. But getting it centered takes a couple of minutes.

With its fan’s power lead plugged into our motherboard’s header, SilenX's EFZ-120HA5 installation is complete.

Installing Thermalright's Archon SB-E

The Archon SB-E includes the cooler and AMD/Intel universal installation kit from Thermalright’s Archon Rev.A, plus an LGA 2011-specific standoff set.

While other platforms get to this point using long nuts and a separate support plate, LGA 2011 standoffs screw directly to the motherboard’s integrated cooler mount.

The Archon SB-E’s top bracket is installed onto standoffs using small screws.

Self-adhesive vibration dampening pads are most easily stuck to the cooler before attaching it to the motherboard.

The cooler is then clamped down to the hold-down plate using an included strap.

Note that the strap only centers the CPU cooler front-to-back, leaving around ¼” of side-to-side adjustability.

The fan’s lead is then plugged into the motherboard, and the fan itself clipped to the heatsink. Note that the fan cannot be rotated 90° because its hole pattern is rectangular rather than square. No matter how far we slide the base, the Archon SB-E always hangs past our top video card slot. Thus, we were forced to test the cooler with our graphics card in slot position three, which means that it wasn’t tested in a standard configuration. We’ll make sure to mention this as we discuss our thermal findings.

Installing Xigmatek's Venus XP-SD1266

Xigmatek uses the same bracket set as Enermax, though these are packed a little differently. Bushings for the aluminum pressure plate, for example, were found beside the plate rather than inside of it, and there’s no included wrench for the tension nuts.

Xigmatek labels its Intel and AMD brackets, though the AMD brackets are also drilled to fit Intel socket hole spacing.

This is how Enermax's installation would have appeared at the half-way point, had we used the other bracket set (minus the Xigmatek and Intel labels, of course). Cross brackets are mounted to LGA 2011-specific standoffs using thumb nuts.

With no wrench to play with, we tightened Xigmatek’s tension nuts using an electric screwdriver. The results are the same, with a slight space found between the cross brackets and the pressure plate when both nuts are fully tightened.

Like SilenX, Xigmatek uses rubber tacks to hold its Venus XP-SD1266 fan and heat sink together. Xigmatek’s design is better in a couple ways, including the fact that the sink is actually tall enough to hold the fan. The tacks are also slotted and hooked on from the sides, rather than forced in from above and below.

Installing Zalman's CNPS12X

Zalman’s CNPS12X installation kit supports AMD’s Sockets AM2 through AM3+, in addition to Intel’s desktop LGA 775 through LGA 2011. LGA 2011-specific standoffs let you get rid of the irksome support plate in favor of the interface's integrated mounts.

Standoffs are threaded in first (foreground), followed by cross brackets and thumb screws (background).

At the edge of a factory-installed pressure bracket, factory-installed spring screws engage holes on the cross brackets to provide even pressure against the CPU.

Ah, but wait…isn’t that backwards? Zalman designed the CNPS12X with ¼” offset to clear the top of your case, your power supply, or whatever else is above your motherboard. Unfortunately, that causes the cooler to overlap our motherboard’s main graphics card slot.

We were left to either put our graphics card in the wrong slot and partially discredit the results (as we will with one other cooler) or mount it in such a way that some of the air pushed through it would already be warmed by our graphics card and power supply. Choosing the latter means that the test results represent a performance floor, by which we know it performs "at least this well."

A final installation issue that affected our build was that the CNPS12X overlapped our second DIMM slot and was just low enough to the motherboard that the fan tips barely touched our 1.55” DIMMs. Though we were able to make things work with a little adjustment, the solution wasn't long-term enough that we'd recommend this cooler to anyone with modules even a hair taller than 1.50”.

Test Settings And Benchmarks

Zalman’s ZM-STG2 thermal paste is used in this test not to favor its cooler, but because Zalman has long been our thermal paste provider. It works equally well with other cooler brands, and settling on a single substance reduces variability in testing.

Our CPU and voltage regulator pull around 300 W at full load at this O/C, using an old C0 stepping Core i7-3960X at 4.60 GHz and 1.40 V. We used this specific processor because it’s the one we can most easily afford to lose (due to heat damage) after long bouts of overclocking.

Twelve threads of Prime95 provide 100% load. We were forced to test noise at extremely close distances to stay within our meter’s calibration range.

Benchmark Results: Cooling And Fan Speed

Noctua and Zalman are in a dead heat at the top of our CPU temperature chart, and its here where many readers will remind us that Zalman’s cooler was mounted backwards (where its intake air might have been slightly warmer).

We’d give Zalman the benefit of the doubt if the fitment issue wasn't such a concern. Besides, some builders (and some cooler manufacturers) prefer to see chassis exhaust fans mounted as an extra intake, though we can only suggest doing that with cases that have extremely good top-panel exhaust.

SilenX’s temperature here has to be recorded as an estimate as a result of processor throttling. You see, the displayed temperature is an average for all cores, and this processor begins throttling at 90° C. Without the ability to achieve lower (and less realistic) ambient temperatures, we simply couldn’t keep the warmest core below its throttle point.

After gauging the amount of time it took to reach throttle temperature at various ambient temperatures, we looked at our rough time to ambient temperature curve and estimated that this cooler could only keep our system at full speed with a room temperature below -4° C.

Enermax's ETS-T40 similarly required an unrealistically-low ambient temperature of 20° C at these overclocked settings. We say unrealistic because we never cool our offices below 22° C in the summer, and even our coolest cases are slightly warmer inside. Yet, we can say with certainty that the ETS-T40 is at least good enough to use with this processor at a lower frequency, since it barely crossed the borderline while using a fairly high 1.40 V core voltage and a full CPU load.

Thermalright offers the lowest fan speed, so we expect it to give off the least amount of noise. Yet, the same product manages to achieve respectable cooling performance in the middle of the pack thanks to its fan’s enormous size. Unfortunately, those same larger dimensions forced us to move our graphics card to a different slot. Proper fitment is an important variable, so this small configuration change negates any possibility for a cooling-to-noise award. As with SilenX, Thermalright’s cooling performance is simply an estimation.

Benchmark Results: Noise And Acoustic Efficiency

Single-fan coolers have the lowest noise at any given speed, since each doubling of identical sound sources adds an extra 3 db to the reading (in audio theory). That same theory explains why Arctic Cooling and SilenX’s mid-speed individual fans take second and third place in silence, ahead of Zalman’s three low-speed fans.

We’d consider anything below 30 db in open air to be quiet, mostly because we use quiet cases to cut back on ambient noise even more. True seekers of silence might settle for nothing less than Thermalright’s Archon, however, even though it doesn’t fit our motherboard properly.

The Archon SB-E is so quiet, in fact, that it tops our Acoustic Efficiency charts, in spite of its mid-pack thermal performance. At the other end of the chart, the SilenX EFZ-120HA5’s cooling performance can't compensate for its second-place noise level.

Which LGA 2011 Cooler Would We Buy?

The top three cooling performers, in order, are Noctua and Zalman, followed by Phanteks. The three coolers with the highest cooling-to-noise ratio are from Thermalright, Zalman, and Noctua. Wouldn’t it be nice if one of these companies could also top our cooling-to-price chart?

Alas, the least-expensive models take that prize. All three of the best cooling-per-dollar contenders use mid-speed fans, and only Corsair needs a pair of fans to achieve this value rating.

If we exclude Enermax from this overclocking-oriented cooler comparison due to its relatively low thermal performance, we’re left with the 31 db Cooler Master Hyper 212 Evo battling Corsair’s A70. A temperature difference of only 1° C certainly makes it hard to argue against Cooler Master’s lower noise level, yet price parity will probably push many enthusiasts to take that 1° to the bank.

It’s at this point where we must wonder what would happen if we compare acoustic efficiency to price.

Cooler Master wins this time, and by a wider margin. If we were basing this review on Intel’s upcoming Core i7-3820, we might even hand out a value award and be done with the story.

But we’re not using a low-cost processor, and prices on Intel’s six-core models justify higher standards of cooling and noise. Putting cost aside, we like the low temperature and noise of Zalman’s CNPS12X. But it also doesn’t fit our motherboard in the desired orientation, or even provide proper memory clearance for our 1.55” tall DIMMs. Instead of using the CNPS12X backwards or trying to remove its fans and face them the other way, and then installing shorter memory that we didn’t even want, we’re forced to consider Noctua’s second-place performer as a perennial champion ever since its took the performance lead two years ago.

The problem with “Best of” awards is that the best often changes before we can even get a review published. Determining that something's the best also means we have to test all of its competition. If we trust that Noctua’s competitors put forth their best efforts for this comparison, two years of performance leadership make the NH-D14 deserving of our highest honor.

But how valuable is this award, really? The NH-D14 just earned this editor’s very first “Best of Tom's Hardware” in the award's four-year history!

A Foundation For Case Cooling: Fans

After explaining the theoretical background in Part One, we're ready to wrap things up in our quest to help you plan and pick the parts for your next air-cooled PC. Admittedly, this excursion also includes a bit of theory. But our main goal here really is to push forward with an actual build. We'll be talking about case fans, heat sinks, thermal grease, and graphics card cooling.

Why Do We Care About Case Fans?

In the last piece, we briefly discussed the chimney effect. However, convection on its own is not sufficient for cooling a desktop PC. The more heat your hardware dissipates, the more air is needed to exhaust it out of your enclosures. This is largely achieved using case fans, which come in many different sizes.

The right combination of case fans plays a huge role in determining the cooling performance of your PC, plus the noise it generates. There are a few rules to follow for maximizing air volume at modest noise levels, and we'll get into those as well.

Will A Small Fan Work, Or Is Bigger Better?

The size of your case fans is often dictated by the chassis you choose and the mounting holes it includes. Fans come in several standard sizes, but we're focusing on 60, 80, 92, 120, and 140 mm models. Larger fans do exist, but a majority of those come factory-installed.

Fans move air using an array of spinning blades, similar to an airplane propeller. When a fan has to spin faster to move more air, it makes more noise. Conversely, blades that turn slower are also more quiet. You can compensate for the loss in air volume from a slower-spinning fan by increasing its diameter. Here's the takeaway: whenever possible, favor a large, slow fan to a small, fast fan. Most case vendors follow this line of thinking and include 120 mm and larger coolers with their enclosures. In general, smaller 80 mm fans are falling out of vogue as a result of the noise they create.

Of course, you don't have to shun 80 mm fans entirely. High-quality coolers can easily run more quietly than less-precisely-manufactured fans, even when they're smaller. We include an affordable 80 mm model in our forthcoming recommendations, which could easily replace a noisy model, if that's all your chassis can accommodate.

Fan Connectors

Fans are either speed-controlled or they're not, and a fan's connector tells you the complete tale. We'll cover voltages, pin-out changes, and simple ways of controlling fan speed. But, in general, case fans run on 12 volts. This voltage is supplied either by the motherboard or directly from the power supply. In the latter case, big four-pin Molex connectors are used (though only two of the four pins, ground and 12 V, are actually needed). Smaller fan connectors are also standardized by Molex. They plug into outputs built onto motherboards or emanating from a dedicated fan controller.

The three-pin plug includes a tachometer feedback signal, which lets the motherboard read a fan's rotational speed. This can then be controlled by varying supply voltage. Fans with four-pin connectors are more common on CPU coolers, and their speed can be controlled with PWM (pulse-width modulation), typically temperature-dependent.

Case Fans: Air Flow And Noise Level

What Does Air Flow Really Mean?

The simple answer is that air flow corresponds to the volume of air moved through the fan in a set interval. Thus, it is a very important parameter for us. Unfortunately, it's often conveyed using different units, making it potentially confusing for the beginner. In English (and in most data sheets), the typical parameter is CFM, or cubic feet per minute, while in Europe, m³/h (cubic meters per hour) is more common. Yes, it'd be easy to mix the two up, but here's a handy conversion chart to keep your comparisons straight.

Example Conversion:

At an end-of-summer clearance, we bought a desk fan sporting 840 m³/h at an appliance store in Germany. What would be the corresponding CFM value for the same fan purchased in the U.S.? We apply the conversion factor and get:

840 x 0.58867 = 494.4828 CFM

See? That wasn’t too hard, after all.

What Is Fan Noise, And How Does It Come About?

Typically, noise is a mix of several frequencies, which makes it hard to define and compare. Looking at the data sheets of fans, noise levels are either specified in dB (decibel), dB(A), or in Sone (loudness). The drawback of merely looking at a specified acoustic spec is that it's hard to imagine how unbearable that number is in the real world.

Whether noise is considered annoying is a multi-faceted issue affected by several factors. Is the disturbance slight, is it a humming motor, or a squeaky bearing? Unfortunately, a fan that boasts a low noise level on its data sheet can be more annoying than one with a less attractive spec as a simple consequence of its tone.

This looks like it could potentially be pretty noisy. First and foremost, a fan's blades impact its acoustic profile through more or less audible flow separation, which depends on build quality and rotational speed. To that end, surface quality, the blade's angle, and the number of blades all directly influence whether or not a fan is noisy. Many manufacturers claim to minimize acoustics with curved blades and golf ball dimples. But more than anything, you can help mitigate noise by not placing fans too close to case openings and meshes, which can result in the dreaded siren effect.

Sleeve, Ball Bearing, Or Something Entirely Different?

A fan's motor can contribute to noise output as well. Many times, motor noise is most annoying at low speeds, and is manifested as humming. Grinding or clanking sounds from the motor bearings are tough to quantify in a specification, but those are super annoying as well. Ball bearings endure less friction than sleeve bearings, and they typically last longer as well (depending on lubrication and design). However, some cheap ball-bearing fans only employ a single bearing, and the rotational forces (plus air pressure) may cause torsion, leading to premature bearing failure. Fans with double ball bearings do not exhibit this problem.

Sleeve bearings are typically quieter than ball bearings, and a ceramic sleeve bearing has the longest life span of all bearing types, since ceramics with a high-quality surface finish offer the least amount of friction. In general, however, double-ball-bearing fans present the best compromise between noise and longevity. Other design features fall into the category of personal preferences; in general, most high-quality fans perform equally well.

Keep in mind that even experts have a tough time predicting how a given fan will perform in the real world.

Interim Conclusion

Regardless of marketing proclamations about innovations like liquid bearings, the best product isn't always the most expensive one in the shiniest box. The true quality of a fan is determined by the ratio of its air flow to measured and perceived noise level. The lower the noise level for a given (high) air flow, the better we consider it for our purposes.

Case Fans: Decoupling Done Right

A Few Notes about Structure-Borne Sound

Most typically, sound levels are cited on data sheets in db(A). This means the sound level is adjusted by a curve that reflects the sound response of the human ear, the A-weighting curve. But this value only encompasses sound emitted through the air. Alas, it becomes almost meaningless for assessing the sound level of a fan mounted in a case.

In addition to emitted sound, there is also structure-borne sound to consider. What is structure-borne sound? It consists mostly of low-frequency vibrations that emanate from the motor and bearing, and are conducted from the fan to the case via the fan mount. A large chassis surface then dissipates these vibrations into the air like a loudspeaker membrane. Stiff, high-quality cases bleed out less sound from this phenomenon than flimsy ones. In addition to the loudspeaker effect, an enclosure's interior can act as a resonance chamber, aggravating the effect.

Correctly Decoupling Case Fans

Here’s the bad news: decoupling frames only benefit one entity, the vendor who sells them, since their mounting screws conduct most of the sound anyway. Rubber/silicone bolts or screws are much better because they barely conduct sound. Used together with a thin, flat gasket from the plumbing aisle of Home Depot, you get a perfectly decoupled fan. Be sure to use soft bolts instead of regular fan screws.

Fans with a Soft Rubber Frame

Companies like Xilence, Deepcool, and Noiseblocker offer case fans with a fan frame made of rubber, which can be screwed to a case without further ado. The whole frame decouples the fan noise, rendering rubber bolts unnecessary. A modern example is the 2CF from Xilence:

Case Fans: Speed Control

Option 1: Do It Yourself For Free

Your power supply already has 12 V and 5 V rails. As a result, you have the option of employing the difference between them (yielding 7 V) to drive a fan using three different voltages. As mentioned, fans that don't support the four-pin PWM connector can be speed-controlled by means of their supply voltage. The lower the voltage, the lower the fan speed. Lower fan speeds typically result in lower noise.

But be aware that every fan has a start-up voltage, required for the fan to start spinning. The start-up voltage is slightly higher than the minimum voltage for the lowest possible fan speed. For instance, some 12 V fans run at 5 V once they have started up. There's a chance they wouldn't fire up at 5 V, though. Thus, you have to experiment in order to figure out if your fans really start up at the voltage you want to use. In order to use all three possible voltages, you can build an adapter yourself using the following illustration:

You can buy a pin extractor tool to pull pins from Molex connectors. Once removed, they are easily reinserted at any position. The pins are held into place by two barbs. You can use tweezers to squeeze the barbs and then gently slide the pin out by tugging on its cable. The magic word here is gentle. Don’t rip the cable from the pin. Do not cut cables and reconnect them with duct tape; that's not professional and might even cause a short inside your gaming rig.

Option 2: The 7 V Adapter with Series Resistor

You can find fan cables with series resistors in PC stores. However, series resistors are matched to certain fan wattages. The higher the wattage, the more voltage drops at the resistor, at which point attaining 7 V for the fan is hit and miss. A powerful fan can even cause the resistor to burn out, so be careful not to exceed a fan power of 1.5 W or so.

Option 3: Aftermarket Fan Controller

PC shops seem to stock fan controllers by the bucketful. Before buying, check the maximum power per channel and perhaps the total peak power level. Do you want to control a PWM-based fan as well? Better make sure the controller supports it.

Whether you want to control your fans with old-fashioned knobs or digital touch panels is totally a matter of personal preference. Let your fashion sense and wallet guide you.

Case Fans: Should You Worry About Positive Or Negative Pressure?

Much Ado About Dust

If your case is more or less airtight (fan openings excluded), the ratio between intake and exhausted air determines whether positive or negative pressure develops inside. Let’s take a look at schematic drawings showing the use of fans with different air flow capabilities, which also applies to groups of fans with different cumulative totals:

Negative Pressure

Benefits

• Good cooling performance for performance-oriented situations
• Amplification of natural convection
• Linear and direct air flow
• Compatible with all direct heat exhaust graphics cards (the ones that blow dissipated heat out of a vent built into the shroud)
• Augments the cooling performance of downward-facing CPU heat sink fans

Drawbacks

• Without a dust filter, dust gathers inside the case
• Graphics cards without the ability to exhaust heat don’t benefit much

Positive Pressure

Benefits

• Only average cooling performance in enthusiast-oriented configurations
• Better support for graphics cards without the ability to exhaust dissipated heat
• All case openings contribute to getting heat out
• Less dust in the case

Drawbacks

• Counters convection
• Air flow is determined by size/position of case openings
• Counteracts the cooling performance of downward-facing processor coolers
• Graphics cards with DHE (direct heat exhaust) can partly counter the overpressure

Analysis

If you want maximum PC performance, which means turning massive amounts of electricity into heat, you should carefully consider the cooling strategy you want to adopt and the best-suited case for it.

Keep in mind that you cannot achieve positive pressure in a case with meshed surfaces or lots of openings. If you want to go that route, you should also be sure that the heat sink on your processor isn't being artificially handicapped.

Case Fans: Recommendations

The Right Fan for the Right Purpose

This page isn't intended to drive your purchasing decision in the way one of our detailed reviews would. We have a very comprehensive case fan round-up in the works right now that should help there. But we have tagged the following fans based on their quality, suitability, and recommendations from Tom's Hardware staff and our forum members. You can browse this table, look at the data, and pick a fan that suits you. Or, simply wait for the upcoming round-up, where we put a wide range of models to the test.

60 mm Fans

        Scythe Mini KazeSize: 60 x 60 x 20 mmWeight: 48 gSound level: 19.99 dB(A)Speed: 2500 RPMAirflow: 20.09 m³/hStart-up voltage: 9 VNominal Voltage: 10.8 - 13.2 VConnector: Two-Pin/Three-PinExpected life: 30 000 hours   

80 mm Fans

Noctua NF-R8Size: 80 x 80 x 25 mmBearing Type: SSO-BearingSpeed: 800/1300/1800 RPMSound level: 7/10/17 dB(A)Air Flow: 26/39/53 m³Blade Geometry: Straight-Blade DesignPower Consumption: 1.44 WOperating Voltage: 4-13 VMTBF: > 150 000 hours

120 mm Fans

Scythe Slip Stream SY1225SL12SH 1900 RPMSize: 120 x 120 x 25 mmWeight: ca. 115gSpeed: 1900 RPMAirflow: max. 186.86 m³/hPower Consumption: 6.36 WSound level: 37.0 dB(A)Connectors: Three-Pin/Four-Pin AdapterBearing: SleeveExpected life: 30 000 hours

Noiseblocker Multiframe S-Series M12Size: 120 x 120 x 25 mmWeight: ca. 155 gMaterial: Plastic, SiliconeSound level: 7-23 dB(A)Speed: 600-1500 RPMAirflow: 40-100 m³/hStartup voltage: 5.5 VPower Consumption: 1.56 W max.Connector: Four-Pin PWMExpected Life: 130 000 HoursWarranty: Six Years

Enermax T.B. Vegas Trio UCTVT12PSize: 120 x 120 x 25 mmMaterial: PlasticColor: Black transparent (frame), transparent (rotor)Material: Metal (fan blade leading edge)Sound level: 14 dB(A) min.Speed: 500-1600 RPMAirflow: 34.7-114.65 m³/hAir Pressure: 0.618-1.912 mm-H2OVoltage: 12 VPower Consumption: 8.4 W Connector: Four-Pin PWMExpected Life: 100 000 HoursLED: 54x (18x Blue, 18x Red, 18x Green)

We tried to pick quiet fans, which include powerful coolers, inexpensive ones, high-end models, plain fans, and colorful models. There should be a fan here for every need and taste.

CPU Coolers: Selection And Installation

There are a great number of aftermarket CPU coolers available, and many are more capable than the boxed solutions you get with a retail processor from AMD or Intel. Don't make the mistake of assuming that they're all better, or even compatible, though. Before you pick your next heat sink and fan, ask (and then answer) the following questions:

• Does this cooler fit my processor interface?
• Is its specified cooling performance sufficient for my needs?
• Does the cooler’s air flow direction match my case?
• Does the cooler fit my case (and over my memory modules) with regard to its dimensions?
• Can the cooler be turned, if needed?
• Does the cooler come with a back plate for support (if it weighs more than a pound)?
• Does the cooler's specified noise level fit my needs?
• Does the cooler’s price make sense for this PC?
• Are benchmarks for this cooler available?

Once you've answered the important questions, then you can start considering how each potential winner looks.

Suitability And performance Always Trump A Pretty Face!

Typically, picking a cooler depends principally on the constraints of a build budget. It doesn't make sense to buy more heat sink than you need, though it's nice to keep an overclocked processor nice and cool with performance reserves. It doesn't make sense to go overboard, just like it doesn't make sense to pinch pennies and end up with stability issues. Most importantly, your cooler of choice needs to live up to its specs when it's running inside a case, in the real world. A healthy mix of solid engineering and raw airflow usually does the trick, regardless of whether you pick a tower cooler, a downward-facing blower, or some more exotic design.

Without a good fan, the best CPU cooler simply cannot live up to its potential. Some companies deliberately don't bundle their heat sinks with fans, giving the user freedom to pick the right complement. As an example, we found that the Prolimatech Super Mega combined with a PWM-controlled Noiseblocker Multiframe is ideal for running a Core i5-2500K at up to 5 GHz. At 4.5 GHz, the system whispers at a barely-measurable 22 dB(A).

Massive Base Or Exposed Heat Pipes?

This seems to be a question shrouded in marketing. Many high-end coolers (like Prolimatech) have a massive base plate and no flattened heat pipes. Xigmatek's products, however, do. They feature the cryptically-named D.L.H.D.T. (Double Layer with Heat-pipe Direct Touch) technology, and seemingly derive no real advantage from it. We think that other criteria are more important in choosing a cooler.

Seating the Cooler

Using the right amount of pressure is essential. The more-is-better approach doesn't work here because you don't want to warp your motherboard and create an uneven fit. Also, don't test for correct cooler placement by turning it, as doing this may rupture the thin layer of thermal paste. Instead, fasten the screws until you feel resistance, and then take a look at the CPU temperature in the BIOS. If that number is in the safe zone, odds are good that the cooler is properly mounted.

Ease Of Installation Instead of Knotted Fingers

As with any other piece of hardware, it makes sense to do your research before buying. Browse cooler reviews for a model that's easy enough to install if you're worried about the complexity of an enthusiast-oriented product. And don't be afraid to take to the forums for help from our community of power users.

CPU Coolers: The Right Thermal Paste

Is There Such A Thing As The Perfect Thermal Paste?

From the user’s perspective, we have to answer this question with an emphatic no. There are suitable and unsuitable, bad, run-of-the-mill, and excellent thermal pastes. Certain options are suitable for specific scenarios and a range of budgets.

There are two common categories of thermal compounds: metal-based and metal-free, each of which can be liquid, creamy, or almost solid. Specialty products like nano pastes, liquid-metal pads, and metallic liquid are intended for professionals with skills, experience, and occasionally even nerves of steel.

For the beginner, the number of options that are both easy to use and completely effective seems to be quite narrow. Based on our experience, the best pastes for a neophyte are the simple, semi-liquid ones. It doesn’t matter whether the paste is silver-based or loaded with nano ceramics. The achievable cooling performance is quite similar.

In order to test one high-end paste from the liquid-metal family, we had to replace Xigmatek's Aegir on our test bench, since liquid metal pastes cannot be used on a cooler with exposed flat heat pipes. This is due to the exposed aluminum.

Thus, we picked the Xilence M606 (which features fairly similar performance) teamed with the 2CF cooler, and tested a few of the currently-available thermal pastes.

Test Results for Six Thermal Pastes

Once again, we are using the test rig from Part 1 of this short series and an old AMD Athlon 64 FX-62 processor, which can operate at three distinct power levels. This processor makes it easy to evaluate paste performance at different thermal power levels. The test bed's case is closed, its power supply is at the bottom of the enclosures, and the case fan arrangement leads to negative air pressure (fans on top and in front).

Not unexpectedly, we see that the high-end paste achieves lower temperature than the solutions we'd consider ideal for beginnings by a margin of 3 to 5 kelvins.

It is worth mentioning that you can easy erase any benefit inherent to an enthusiast-class thermal compound by handling and applying it improperly. On the other hand, if you use a mainstream product properly, it'll give you solid results.

The Xilence X5 and the Arctic MX2 are both non-conducting liquid pastes, which can be applied and spread easily. Below 100 W thermal power, the X5 slightly beats the MX2. Above that, the picture changes and the MX2 takes the lead. But keep in mind that differences of 1 kelvin can barely be measured consistently anyway. Both products are inexpensive and easy to use. Since the Xilence X5 is also suitable for graphics cards, we chose the inexpensive X5 as a reference for other tests and recommend it.

The enclosed spatula is suitable for spreading the paste, but in the next section we will illustrate an even simpler and cleaner way.

CPU Coolers: Applying Thermal Grease

Is There A Perfect Method?

Just as every BBQ fanatic claims to know how to grill a perfect steak, enthusiasts all have opinions on the best way to apply thermal compound. Do you spread it with a spatula, flatten it with a razor blade/credit card, use the one-fingered glove method, or merely apply a drop of paste in the middle of your processor? Heated forum discussions suggest that there is no such thing as the method. However, because this article is tailored to beginners, we want to focus on consistency and repeatability. Nobody wants to fry their CPU, anyway. After field tests with a couple of novices, we settled on the following:

A Drop Of Paste Method: Foolproof

By squeezing the tube, we put a small drop of paste right in the middle of the CPU. The correct quantity is about the size of a lentil (not a pea). Refer to the pictures below for the end result using different quantities of paste.

In order to take these pictures, we put an ultra-thin transparent film between the cooler base and the CPU. We installed and then uninstalled the cooler. The paste remained between the CPU heat spreader and transparent film. Thus, these pictures illustrate the paste spread just as if the cooler was invisible. Let’s take a look at the results when different quantities of thermal paste are used:

Upper Limit (Maximum)

It is important to install the cooler evenly and to push it down a little bit. Hooking the cooler in on one side and then tilting it into position will result in a lopsided spread. Screws should be tightened in a diagonally alternating way.

Sins, Mistakes, And A Brief Afterword

There are many methods that achieve similar results, when properly used. However, spreading the paste with a gloved finger seems problematic, as it would be difficult to suggest the proper quantity. Furthermore, just like the credit card method, this technique is too difficult and uncertain, as the thickness of the compound layer is tough to gauge without lots of experience.

Cheesecake? Way too much paste! Apart from the fact that everything oozes out and makes a big mess, the thermal conductance is negatively impacted and the cooler does not achieve its optimal performance.

Mister Scrooge. It’s OK to be frugal, but not when it comes to thermal paste. This is how people buttered their bread during the Great Depression. For a CPU, it’s not enough. When you can read the type and date code, the layer of compound is too thin.

CPU Coolers: Initial Startup And Test Run

The Very First Test Run

You can never be 100% certain that the cooler is installed correctly until you power up for the first time. Thus, it is important to quickly check the CPU temperature. Boot up the PC, enter the BIOS, and seek out the monitoring information. One advantage of checking the processor's temperature in the BIOS is that its power-saving technologies aren't yet enabled, forcing the chip to run at full speed. Keep an eye on both the fan speeds and the vital temperatures.

Adjusting the Fan Speed

Once you have verified that the CPU temperature isn't spiraling out of control, and that the cooler is doing its job, you can go ahead and optimize the fan speeds. If you are not familiar with your firmware, check the manual for more information on where to find each setting. PWM-controlled fans with a four-pin connector can be slowed down based on thermal thresholds by setting a target temperature and a fan speed. Even fans with a three-pin connector can sometimes be controlled, albeit through their voltage. In either case, the fan speeds up in response to processor loading and heat, saving your ears from a constant drone.

Stress And Stability Test

After setting each fan parameter, you can conduct a stress test. In Windows, you can use Linpack (Windows executable: LinX) or Prime95, and monitor the CPU temperature with a program like CoreTemp or HWMonitor.

When reading the core temperatures, make sure that the Tjunction parameter is set correctly; otherwise the readings won’t make much sense.

VGA Coolers: We Rescue A GeForce GTX 480

Unconventional Repair Job Instead Of Expensive Replacement

Swapping or upgrading a graphics card's cooler is not a job for beginners. Thus, we omit that in this entry-level tutorial. However, even a neophyte can reproduce our unconventional experiment.

Let’s assume you own a factory-overclocked GeForce GTX 480 (like the one in the picture), and let’s further assume that the Accelero Xtreme's fans just died on you. Naturally, your warranty already expired and a quick search on eBay doesn't turn up any results for spare parts. Now what?

Broken is broken, and a new Accelero fan assembly costs something like 50 bucks. Thus, the only option is to dig up a fan out of the spare parts bin. The new blower can't be any thicker than the original because we don’t want to block an additional PCI slot, and it should at least boast the same performance.

Two Slip Stream Fans Cool The GeForce Like Never Before

Now, granted, we attached both fans in a somewhat haphazard manner, since we weren't even sure if this would work. We didn't take a boatload of pictures, either.

However, our measurements would make an Eskimo proud: the home-grown solution is slimmer, quieter, and significantly cooler! As an added benefit, the air stream that passes through the protruding fans also cools the top of the card. An experiment with 92 mm fans, which don’t stick out the same way, wound up 5 kelvins warmer, and thus we didn’t even bother to take pictures.

Rigged Fans Beat Nvidia's Factory Setup

This is amazing, but true. Our ugly ducklings beat the Accelero Xtreme's factory-installed fans in all respects. This puts the fun back into big old Fermi! Take a look at the measurement data:

We could have taken full-sized 120 mm fans and seen even better results. However, the card would have then monopolized three slots, which would have been too much for our tastes. 

MacGyver would applaud us, even though these fans won’t detonate. Some digging in the spare parts bin saved an expensive graphics card from the garbage heap, and we even managed to improve its cooling. Had we bought the fans brand new, a ~$20 total for both wouldn’t have broken the bank.

VGA Coolers: Single-Slot Whisper Cooler

DIY Improvements

Curiosity may have killed the cat, but it also got us to play around with another graphics card. We recently reviewed Afox's Radeon HD 6850 low-profile card on our German site. In that story, we used two 80 mm fans to improve cooling versus the OEM ones. Here, we want to up the ante in terms of performance and noise. The result is a completely usable low-profile card that doesn't generate much noise at full load. Once again, we dipped into the Tom's Hardware parts bin for a couple of spare fans.

Conversion in Two Stages

In the original review, we used a pair of Enermax T.B. Silence fans, which protrude slightly. Since the cooler is merely an inexpensive extruded aluminum slab, protruding fans do not offer any advantage. To the contrary, the fans do not live up to their potential and the lid of the graphics card doesn't fit properly.

Thus, we dug around once more and found two 60 mm Scythe Mini Kaze fans:

Updated Results and a Chilly Surprise

Two 60 mm fans are good enough to cool this card quietly and effectively. As the fans are uncontrolled, the noise level stays the same regardless of the graphics workload.

Conclusion

Our do-it-yourself experiment got the manufacturer thinking about a new revision of this card, which would replace its two cheapo and loud fans. If the missing power plug also makes a comeback, this card could be among the few price/performance leaders for HTPCs. Cost of the 60 mm fans: less than $20.

Think About Cooling Early

Summary

It is essential to think about cooling issues before you start buying hardware. Picking the right chassis, CPU cooler, and case fans lays the foundation of a successful build. As much as we're trained to snap our necks for attractive parts, good looks simply have to play second fiddle here.

We hope that we clarified a few things for beginners. Even the of applying thermal paste is unfounded. Squeeze out a drop and you’re done.

Our goal here wasn't to make purchasing recommendations, but to provide a tutorial. As time passes, there will always be new and improved products. In the end, though, only the best will do. The best doesn’t necessarily mean most expensive, either.

Instead Of Closing Thoughts, Bon Appétit

We've done some extreme cooling coverage in the past (that link is in German), so we wanted to leave you with a humorous recipe for enthusiasts who prefer turning their PCs into Dutch ovens rather than cooling them properly:

Tom's Menu - Cheese Fondue á la Technical AuthorWe changed the ingredients slightly, but the result turned out to be appealing to the eyes and taste buds.Ingredients: • Swiss cheese • Dry Prosecco (make sure you're at least 21, of course) • Salt and Pepper • Garlic piecesSeasoning: • Worcester sauce • Lemon juice (optional)

Step 1We round up the ingredients

Step 2We use a Radeon HD 6990 with its overclocked BIOS enabled to create a 450 W hot plate

Step 3We mix Prosecco into the melted cheese and add some seasoning

Step 4Bon Appétit

Nettop Nirvana

Intel’s Atom CPU might have been the driving force behind the popularization of the nettop form factor, but manufacturers are squeezing more powerful hardware into these tiny machines. Sometimes, more potent graphics performance is added via a mobile chipset. Sometimes, processing m muscle is emphasized instead with a CPU designed for powerful notebook. And now, APUs belonging to AMD's Fusion initiative are an option, serving up efficient power use and higher-performance graphics on a single processor die.

Truly, the nettop is no longer a stripped-down machine barely capable of Web browsing and word processing. These tiny PCs are tailor-made to excel in specific applications, which often includes use in a home theater.

We’re taking a close look at four very different nettops. Before we examine them, here are their specifications:

Arctic MC001-BD

The foci of Arctic’s MC001-BD are summed up with two words: video and silence. This is the only passively-cooled nettop in our roundup, and the little noise it does generate comes from the hard disk; if you choose to upgrade that to an SSD, it's completely silent. The system is driven by a dual-core Hyper-Threaded Atom D525. But instead of leaning on the integrated Intel GMA 3150 graphics core, the engineers opted to add a Mobility Radeon 5430 with 512 MB of dedicated DDR3 memory. This is the only nettop in our roundup with dedicated graphics RAM, and AMD's Radeon GPU is the key component that should make the MC001-BD a capable media center.

Arctic's entry sells for $714 at www.arctic.ac. That price is higher than some of the other models in our round-up, but keep in mind that it includes Windows 7 (the Home Premium 32-bit flavor). It also comes with 4 GB of DDR3 RAM, a 500 GB hard disk, and an LG slimline Blu-ray ROM drive.

The aluminum and glossy black finish is simultaneously classy and understated. Arctic Cooling's MC001-BD certainly looks at home in the living room. At 10.5” x 5.5” x 1.5”, it consumes the most shelf space of any unit in our round-up, but it’s certainly not large compared to more familiar desktop PCs.

On the back, we see HDMI and VGA video outputs, an optical S/PDIF audio output, five USB ports, a gigabit Ethernet port, six 1/8" analog audio jacks, and the power supply input. This is a lot of connectivity options for a nettop, especially in regard to all of the USB ports.

The unit's front side hosts a memory card reader, two USB 3.0 ports, and requisite front-panel speaker and microphone jacks.

Arctic's MC001-BD comes with an external DC power supply, an HDMI cable, a wireless antenna, a manual, and the CyberLink BD Blu-ray playback software. There is no remote, which we consider inexcusable for a media-oriented PC. You can always add a third-party solution later, though. Arctic Cooling is also in the process of certifying a remote control application for Android-based phones and Apple's iPhone. We weren’t able to try the iPhone app, since it isn't available, but we expect it in the App Store soon.

The inside of the MC001-BD is dominated by its heat sink, which is no surprise considering that the unit is passively cooled. Otherwise, upgradeable components like the hard drive and memory are relatively easy to access.

ASRock CoreHT 252B

Priced at $800 on Newegg, ASRock’s CoreHT 252B is the most expensive contender in our round-up. That doesn’t suggest a lack of value, though. This unit boasts the most powerful CPU, the fastest hard disk, and the ability to play Blu-ray 3D movies over HDMI. The only thing you don't get for $800 is an operating system.

Despite its diminutive size, the CoreHT 252B compromises very little when it comes to performance. The Core i5-2520M CPU is based on Intel’s Sandy Bridge architecture, and unlike the rest of the competition in this round-up, it runs at a desktop-like clock rate (2.5 GHz stock, with a maximum Turbo Boost frequency of 3.1 GHz).

Although it's clearly powerful, this system is also quite small at 7.5” x 7.5” x 2.75”. Its classic glossy black finish is attractive, yet unassuming.

The rear output panel hosts HDMI and VGA video outputs, an eSATA port that can also be used as a USB port, six USB ports (two of which support USB 3.0 transfer rates), a gigabit Ethernet jack, the power supply input, optical audio output, and five 1/8" audio jacks.

Up front, you get a power button, speaker and microphone jacks, and a pair of USB 3.0 ports. That gives the CoreHT 252B a total of 11 USB connectors, which we consider to be very impressive expansion on a nettop. The only thing missing is a memory card reader, but surely there's enough scalability here to accommodate a USB-based card reader.

Bundled you'll find a no-slip pad, an external DC power supply, a driver CD, a manual, a remote with two batteries, an HDMI-to-DVI dongle, and extra cables for connecting a second hard disk (though storage expansion is rare in a chassis this small).

We’re quite fond of the media center remote that ASRock previously included with its Core100 family. The new, smaller model works just as well, despite the awkwardly tiny buttons and labels.

Disassembling the CoreHT 252B is a very simple affair, with two screws holding the case cover and two more screws securing the drive bay insert. With these removed, access to all of the components is easy.

Zotac Zbox AD03BR-PLUS

We're familiar with Zotac’s family of tiny Zbox HTPCs, but the AD03BR-PLUS model adds AMD’s E-350 APU to the mix. Combining low power usage, minimal thermal output, and an integrated Radeon HD 6310 graphics processor, this should make for a solid nettop media center platform. As tested, armed with 2 GB of DDR3 and a 250 GB hard disk, this is the lowest-priced option in our round-up at $504 from www.zotacusa.com. If you prefer to choose your own memory and storage, a barebones model sells for $399. Just remember that neither model includes an operating system.

The Zbox is long and flat, measuring 11.5” x 7.5” x 1.5”. Its glossy black and brushed metal chassis are attractive, and we dig the large blue illuminated circle (though it might not be the most welcome addition in a dark theater room).

The back side features HDMI and DVI video outputs, a combo eSATA/USB port, a USB 3.0 port, gigabit Ethernet, the power supply input, and an optical S/PDIF audio output. After testing two USB-rich nettops, the single USB port is disappointing; you're forced to use the eSATA combo port for USB if you need a separate keyboard and mouse (and don't want cabled dangling off the front). The most obvious workaround is to use a wireless keyboard and mouse setup that requires one USB port for its receiver.

Zotac's Zbox doesn’t come with a remote, detracting from its allure in a home theater capacity. If you want to add one, you might have to plug its receiver into the front of the unit.

More than half of the Zbox's narrow front is consumed by the optical drive slot. You get two more USB ports up there, though, one of which transfers at USB 2.0 speeds and the other supporting USB 3.0. Audio and microphone jacks are also routed to the front.

Zotac’s entry comes with an external DC power supply, a support (driver) CD, a manual, a quick start guide, a VESA mounting bracket for attaching the Zbox to the back of a monitor, a DVI-to-VGA converter, and a full version of CyberLink’s BD Solution software for Blu-ray playback.

As mentioned, no remote comes bundled with the Zbox. While we accept that some folks have a third-party favorite they'd prefer anyway (or even an app like Mobile Mouse Pro), we still think remote control should be standard on a product so obviously intended for HTPC duty.

Accessing the hard disk and memory is quite easy via a rear access panel. 

Giada i50 B5541

Giada has a reputation for producing extremely tiny PCs, and the i50 is no exception. Not satisfied with the mediocre performance of Intel’s Atom, the company opts for a dual-core, Hyper-Threaded Core i3-430UM. At 1.2 GHz stock with a 1.73 GHz peak Turbo Boost clock rate, this processor isn't as potent as ASRock's Sandy Bridge-based solution, but it should wipe the floor with competing E-350- and Atom D525-based platforms.

With no space for an optical Blu-ray drive, Giada's i50 isn’t equipped for kicking back in the living room and watching the latest blockbuster. Instead, it's geared for productivity and Web surfing in environments where desktop real estate is at a premium, like you'd see in a dorm room.

Armed with 4 GB of RAM and a 500 GB hard disk, the i50-B5541 sells for $659 at shop.polywell.com, which includes a copy of Windows 7 Home Premium. A stripped-down version with 2 GB of RAM and a 320 GB drive goes for $465 from www.amazon.com, but that package comes with Ubuntu Linux, a free operating system. Incidentally, the Giada i50 is the only system in our round-up with a built-in Bluetooth controller.

As we've seen in the past, Giada’s nettop is by far the smallest of the four we're testing (7.5” x 5.9” x  1”). The finish is glossy and dark, with silver around the edges. Though the lack of actual aluminum is missed, the unit still looks nice (albeit plastic-y).

HDMI and VGA video outputs are exposed on the back, along with four USB ports, a gigabit Ethernet port, and the power supply input. Giada learned that more USB ports are important after we pointed this out on its older N20 model, and we really appreciate the consideration of our feedback there.

Up top, a small door conceals a memory card reader, a USB 3.0 port, and the expected speaker and microphone jacks. It’s a shame that there aren’t any 1/8" audio jacks around back because, if you're not sending audio over HDMI, you have to keep this door permanently open to output an analog signal.

The Giada i50 comes with a plastic base, an external DC power supply, a driver CD, a manual, a VESA monitor mount, a remote guide, and a remote with two AAA batteries.

The remote is large enough to fit in our hands comfortably, incorporating the requisite directional keys and playback controls. The large music button launches Windows Media Center.

Giada’s slim nettops aren't meant to be disassembled, and they consequently don't include instructions on how to take them apart. We managed to get the unit open without breaking it, thanks to help from the company itself. Essentially, you need to remove all of the screws you can see, and then pry the side off of the case off by inserting a small tool (a screwdriver) through the heat sink port. Once the cover is off, the other side is freed by removing additional screws, exposing the rest of the components.

This is no simple affair, and I agree that casual users should steer clear of pulling the i50 apart. But for those who need access to the nettop's internals, the job is doable so long as you don't mind pushing heat sink fins back into place. As you can see from the pictures, Giada wastes no space in its nettops.

Test Systems And Benchmarks

These are very different systems with varying target markets. We need to compare performance though, so we’re putting them through a battery of tests that allow us to compare their strengths and weaknesses. We’re even benchmarking a bit of light gaming.

Benchmark Results: Synthetics

SiSoft Sandra and PCMark 7 are good synthetics for comparing these four systems. Given a quartet of different architectures, it'll be interesting to see how the numbers fall. Let’s start with the SiSoft Sandra benchmarks, which focus on processor and memory performance:

The Sandra scores don't offer up any surprises; the Intel Core i5-2520M leads, and the Core i3-430UM falls into a distant second. The AMD E-350 and Atom D525 are tied for third place.

The multi-media capabilities of these processors mirror the arithmetic results we already saw.

The memory bandwidth results aid in demonstrating why ASRock's CoreHT dominates our benchmarks. The other competitors fall way behind in this metric.

The Core i5-2520M in ASRock’s CoreHT 252B leaves its competition in the dust when it comes to PCMark 7, which is no surprise considering this processor's relatively high clock rate and desktop architecture. The real surprise is that the Atom D525, E-350, and Core i3-430UM are so closely matched.

Now that we have an idea how these nettops perform, bear in mind that they're not built for high performance. They’re built for media playback and general productivity. Despite their modest results, these systems still provide workable performance as media centers and Web machines.

The fast 500 GB Western Digital Scorpio Black serves ASRock's nettop very well. The other systems have relatively similar storage performance, though Zotac's Zbox isn't helped by its 5400 RPM disk compared to the 7200 RPM competition.

Now, let's consider gaming performance:

The graphics benchmark doesn’t flatter any of these solutions, and even ASRock’s CoreHT can’t provide a 30 FPS rate. The Radeon-based graphics chipsets in the Zbox and MC001-BD don’t fare well, but the HD Graphics engine in the Giada i50 is hit hardest.

Benchmark Results: Productivity

Now we can dig into the individual productivity-oriented PCMark 7 benchmarks, including Web browsing, data encryption, image manipulation, and video transcoding performance.

When it comes to Web browsing, the ASRock nettop maintains its first-place position. The Core i3-430UM in Giada’s i50 has little trouble muscling its way past the more anemic E-350 and Atom D525 processors. Really though, any of these competitors is sufficient for general usability.

The Giada i50 and Zotac Zbox run close together, while the Arctic Cooling MC001-BD falls behind. The CoreHT takes its now-familiar first-place finish way up in front of the pack.

Photoshop users should be interested in the results of PCMark 7's image manipulation test. The Giada i50 demonstrates a big improvement over the i50 and Zbox. By now, though, there's no questioning which one of these boxes is fastest.

When it comes to video transcoding, cores and clock rate are king, and ASRock leads by a mile. Arctic’s MC001-BD is held back most severely by the low-power Atom processor.

Benchmark Results: StarCraft II

None of these nettops were designed for gaming, but we still want to know their limitations, and whether or not casual gaming is even a possibility. We've added results gathered from testing an Atom D525 nettop equipped with Nvidia's second-gen Ion chipset, originally generated for our Low-Power Gaming: AMD's E-350 Vs. Intel Atom D525 And Ion 2 article.

Let’s start with StarCraft II:

This CPU-limited game is only viable on ASRock’s CoreHT 252B; none of the other solutions can deliver a playable minimum frame rate.

Benchmark Results: Left 4 Dead 2

Left 4 Dead 2 is a great game with a huge multi-player following, and we want to see if these nettops can handle playing it, since the game doesn't apply a heavy graphics load compared to other modern titles.

Once again, ASRock’s entry delivers great performance all the way up to 1080p, while the other options are less impressive. Zotac’s Zbox and Arctic Cooling’s MC001-BD might be considered passable at 720p and below, but Giada’s i50 is out of the running entirely.

Benchmark Results: World Of Warcraft

If these nettop PCs will ever be used to play a game, there’s a good chance it'll be World of Warcraft. Let’s see how they handle the land of Azeroth:

Once again, the ASRock pushes respectable frame rates right up to 1080p. The Zotac and Arctic models handle 720p well enough, but it looks like Giada’s i50 simply cannot handle a 3D workload at all.

Benchmark Results: Video Playback

The following test from PCMark 7 measures how well these platforms play back a 1080p, H.264 video file at 24 FPS:

The Giada i50 suffers a little, likely the result of its entry-level HD Graphics engine. We’re a little surprised to see the Zbox fall behind as well, as we didn't choppiness during actual Blu-ray playback testing. As a matter of fact, it's surprising that none of these solutions manage to achieve a perfect 24 FPS. Despite this, HD video appears smooth in our tests on the Zbox, MC001-BD, and CoreHT.

Now let's look at video playback quality using the HQV 2.0 benchmark. Giada's i50 is left out of this test because it doesn't come equipped with a Blu-ray drive:

The Zbox and MC001-BD share an identical 170-point score, which we expected since both solutions employ similar Radeon-based graphics hardware. The Intel HD Graphics 3000 chipset in the ASRock CoreHT scores slightly less, but still manages a fine result.

Blu-ray 3D Playback

The only one of these nettops advertised with Blu-ray 3D support is ASRock's CoreHT, and it performs this task well. This is our first experience with 3D over HDMI using an Intel graphics chipset, and its performance is surprisingly good thanks to hardware-based decode logic able to handle the otherwise taxing workload.

The Zbox and MC001-BD don't fare as well. Arctic's entry plays back Blu-ray 3D, but without MVC decode acceleration on its Mobility Radeon HD 5430, the resulting slide show is caused by the Atom D525 CPU trying to keep up. As for Zotac's Zbox, the E-350 APU is able to decode MPEG-2, VC-1, and H.264 in hardware. However, AMD didn't see fit to arm it with MVC decode acceleration, which is required for Blu-ray 3D playback. As a result, its two Bobcat cores struggle right alongside Intel's Atom, too slow for a usable experience.

Other Video Considerations

The ASRock, Arctic, and Zotac nettops all feature a Realtek 7.1-channel HD Audio codec and the ability to bitstream lossless audio data over HDMI. ASRock’s CoreHT 252B also features THX TruStudio PRO certification. While the Giada i50 does not have an optical audio output, it can deliver digital audio via a coax adapter that does not come with the bundle.

Benchmark Results: Networking

All of these nettops come equipped with gigabit Ethernet controllers and Wi-Fi, so we used SiSoft Sandra’s networking benchmark to see how they compare.

As far as wired networking goes, ASRock's first-place finish is followed by Giada's i50 and Zotac's Zbox. Arctic’s MC001-BD has a little trouble keeping up.

When wireless networking is tested all of these solutions perform closely, although the ASRock system doesn't perform as well as its contemporaries.

Unlike bandwidth, when it comes to latency, less is better.

Power, Temperature, And Noise Benchmarks

With a well-defined idea how each of these nettops performs, it's also important to consider power use and noise. Especially when it comes to enclosures without much airflow, cooling is a particular challenge that we need to evaluate.

We've finally found something that the ASRock CoreHT 252B struggles with in comparison to the other three competitors. The results are expected given a desktop-class processor. However, all of the machine's components still use less than 60 W with a full GPU and CPU load applied, so that power number hardly seems extreme.

The rest of these systems just sip power. It’s interesting that Arctic’s Atom D525/Mobility Radeon HD 5450 combo is more power hungry than the E-350- and Core i3-430UM-based machines.  

The temperatures are so close that the differences are barely worth mentioning. Arctic's MC001-BD does exceptionally well when you consider that it’s passively cooled.

Although Arctic is the shoo-in given its passive cooling system, the other models hardly generate enough noise to notice. We wouldn’t mind having any of these systems in close proximity on our desks.

Have Nettops Transcended Productivity?

If nothing else, this comparison shows us that the nettop space offers something for everyone, so long as your pockets are deep enough. Each PC introduces unique features in a very small form factor. As a result, the four contenders give us strengths and weaknesses not shared by the rest of the field.

Arctic MC001-BD

Arctic realizes its idea of complete silence in the MC001-BD, a product that tries to fit in with the HTPC crowd. While the Atom D525 processor driving this thing is the least-powerful CPU represented, we're not looking at demanding transcode workloads or threaded rendering projects here. If you need heavy lifting, a nettop is not on your radar, we're sure.

Instead, the video playback capabilities of AMD's Mobility Radeon HD 5430 and the included Blu-ray drive let the MC001-BD shine in its target segment: the home theater.

We do have a few reservations, though. There is no bundled remote, although Android and iPhone owners should get access to free remote control application in the near future. Moreover, this system is incapable of Blu-ray 3D playback because the mobile GPU can't accelerate it and the CPU is too slow to handle it in software.

Priced at $714 including Windows 7 Home Premium, this is the second-most expensive offering in our round-up, too. There aren't a ton of choices for completely silent nettops, though. So, if you can live with this machine's few shortcomings and demand passive cooling, the MC001-BD is at least worth considering.

ASRock CoreHT 252B

With a Sandy Bridge-based Core i5-2520M driving its platform, ASRock’s entry is significantly faster than the other nettops in our round-up. In fact, the CoreHT 252B has enough processing power to stand up against desktops (despite the fact that its CPU is a mobile model). In productivity-oriented benchmarks and light gaming tests, the CoreHT 252B walks on by the rest of the field without consuming much more power or generating much more noise. The included remote and Blu-ray drive make it a capable HTPC with the chops to play back Blu-ray 3D over HDMI.

Not surprisingly, the highest-end features are accompanied by the highest price tag, too. Selling for $800 without an operating system, you could build one heck of a desktop with a lot more horsepower for less money. It wouldn't be as small, as power-friendly, or as quiet as the CoreHT 252B though, and it probably wouldn't hide in a home theater rack, either. Go into this one knowing you're paying a premium for its physical dimensions.

ASRock’s interpretation of a power media-oriented nettop is very capable, and it's ideal for folks who don't want to compromise on performance but still need the very small footprint.

Giada i50-B5541

Giada does big things in tiny spaces. Despite its miniscule stature, the Core i3-430UM in the i50 double the raw performance of Intel's Atom D525 and AMD's E-350 in many compute-intensive disciplines. Bereft of a DVD or Blu-ray drive, the i50 is not ideal as a traditional media center. However, with Flash decode acceleration, it’s at least capable of playing back streaming video over the Web. Its bundled remote control is a bonus for that task, though probably not as useful when you factor in the need for a keyboard. In addition, this is the only nettop in our round-up with an integrated Bluetooth controller. Its only clear weakness is a lack of 3D gaming potential, a consequence of relying on Intel's HD Graphics engine. But gaming isn't what this nettop was designed to do, anyway.

Yes, ASRock's submission is a lot faster. It's also more expensive. You can buy Giada’s i50 for $659, complete with Windows 7 Home Premium. If you can live with Ubuntu, 2 GB of RAM, and a 320 GB hard disk, it sells for as little as $465. Both options are solid values for anyone prioritizing size in their search for a nettop able to handle typical online tasks and basic productivity.

Zotac Zbox AD03BR-Plus

We liked Zotac’s Zbox HD-ID34, and the newer AD03BR-Plus offers even better specifications in the same elegant package. The E-350 APU performs well in applications that don't require a lot of processing power. And the integration of AMD's Radeon HD 6310 is a boon for video playback. Equipped with a Blu-ray drive, the AD03BR-Plus is ideal in a media center environment.

Yes, the dual-core Bobcat-based E-350 processor is fairly weak, only marginally better than Intel’s Atom D525. This system could also use more USB connectivity. And the lack of a bundled remote is strange for a piece of hardware so obviously geared toward the HTPC space. 

Unfortunately, the fact that the graphics component in AMD's E-350 doesn't support MVC acceleration means the two processing cores are tasked with trying to handle Blu-ray 3D playback in software (ultimately, they're too slow for this). But at $504 with no operating system (and $399 for a version without an operating system, hard drive, or memory), the Zbox AD03BR-Plus is still a considerable value proposition. It looks good and performs well in the living room. As far as HTPCs go, the only thing missing here is Blu-ray 3D support.

Arctic Cooling, known mostly for its cooling products, introduced at Computex this year a small form factor media center PC it calls the MC001 that packs an Intel Atom D525, ATI 5430 512 MB graphics card and 4 GB of RAM.

Helping its claims as a media PC is that the MC001 is completely passively cooled, meaning that you'll never hear any fan noise at all while watching a video.

The other specifications include 7.1-channel audio output, two USB 3.0 ports located on the front, IR receiver, 4 in 1 card reader, VGA, HDMI, Ethernet, and 5 additional USB ports at the rear panel. As an option, you can install a DVB-T/DVB-C.

There are four different versions of the MC001 which are differentiated by storage options:

•          500 GB HDD and DVD-RW 24X

•          500 GB HDD and Blu-Ray 4x

•          SSD 120 GB SATA II and DVD-RW 24X

•          SSD 120 GB SATA II and Blu-Ray 4x

All models come with Windows 7 Home Premium 32-bit. Addition accessories are an IR remote and IR keyboard, but thankfully each one comes with an HDMI cable.

Prices start from $249 and can head north of the $600 mark with the SSD and Blu-ray options. Look for it in the coming weeks.

Manufacturers, Designers, And Labels

Between gobs of reader feedback and our own data compiled over many years, we've managed to put together a fairly comprehensive list of power supply brands and manufacturers. Despite the fact that it consists of more than 150 manufacturers, though, this list still doesn't reflect the entire market, which always seems to be in a state of flux. It can, however, be used as a guide to finding the difference between a bad deal and a bargain.

Who is Who?

Let’s start by dividing the manufacturers into three large groups so we can better understand the database and how these companies are connected:

1. The OEMs (Original Equipment Manufacturers)

OEMs manage all of their production internally. They either exclusively design and manufacture their own PSUs (like Enermax) or design and manufacture their own brands, as well as manufacture PSUs designed by other companies (such as FSP, HEC, and SeaSonic). Some of them focus heavily on worldwide exports and provide a range of models, which are later sold under different labels. It's common to find otherwise-identical models marketed under many different names and labels. The industrial areas around Shenzhen, China, are the cradle of the lowest-priced PSUs sold all over the globe.

2. Designers: Without Their Own Production

The second group of companies also develops and designs their own products. However, they have to outsource either some or all of the manufacturing to other companies. One example of this is Be Quiet. Those familiar with the brand noted how Be Quiet P7 models were suddenly much better than the disappointing P6. The answer was simply a manufacturer change, from Topower to FSP. Other examples of designers include SilverStone, Corsair, PC Power & Cooling, and Tagan.

3. The Labels: With or Without Any Technical Involvement

Arguably, this group could be subdivided. Some importers of foreign PSUs that resell models under their own labels have a certain influence over the quality and choice of components, while others simply bring in some very cheap products, relabel, and resell them.

This third group is the most interesting one for price-oriented customers, though also the most uncertain for quality. You're as likely to score a bargain by getting a relabeled high-quality product at a lower price as you are to be disappointed by being too tight-fisted. Some good examples of products to watch are new models from Aerocool, which are essentially the Cougar units from Compucase/HEC with a discounted price and completely restyled exterior.

After many tests and inspections of budget models (by us, our readers, and friendly computer stores), we would advise you to steer your piggy banks clear of the labels Rasurbo, Inter-Tech (Sinan Power, Coba), Tech Solo, LC Power, RaptoxX, Tronje, Xilence, Ultron, World Link, Q-Tec, etc. We were able to identify some of these models without looking at the UL number simply by having a look at the installed components. These were almost exclusively the simplest work of such manufacturers as Enhance, World Link, Andyson, Topower, Casing Macron, and Channel Well.

Lack of protection circuits, low efficiency, and bad build quality were major points of criticism. The lowest of the low was a European label called Hardwaremania24, targeted at OEM PCs. While still in standby mode, the PSU heated to about 176 degrees Fahrenheit, spent the next six hours billowing smoke, and finally made what might be described as a trumpeting sound before dying. The host computer was never even turned on. After analyzing the PSU, we found no protection at all save for a single slow fuse.

Firecracker Or Power Plant?

How do you identify a bad power supply before buying it?

• Extremely high wattage claims at a comparatively very low price are suspicious. There are simply no decent 750 W power supplies for $50. For every product class based on performance and features, there must be a minimum price. When a product is significantly below that price, be cautious. You can get a "400 W PSU" for $20, and such fire hazards are installed in budget PCs every day by unscrupulous companies that know exactly the risk they're handing off to buyers.
• Check the specifications. For example, if a PSU claims high performance on the 3.3 and 5 V rails while the 12 V rail numbers are low, then you know something is wrong.
• The manufacturer does not specify any combined maximum performance, but instead only shows the maximum load for each rail separately. This is done without specifying how much real power would be available if all rails are used at the same time. Avoid PSUs without this information.
• Be careful with juicy marketing expressions and commercial lingo: Super, Extreme, Gaming, Combat, etc. Using superlatives to describe something quite normal should arouse suspicion and have you double-check specification details.
• Passive rather than active Power Factor Correction (PFC) leads to lower power efficiency.
• Very few or short power connectors and cables might be an issue. A 750 W PSU usually has four PCIe connectors for graphics cards (2 x 6-pin and 2 x 6+2-pin), so think twice if a model only offers two (or at least consider your upgrade options).
• With cheap PSUs, the quality of the cable insulation may be poor, or the cables may not be insulated at all. The power cable grommet may also be insufficiently padded.
• Be careful if there are few or no indications of protection circuitry. If the PSU specification only says OPP (overload protection) or perhaps SCP (short circuit protection), this points towards a normal fuse. If the specification also says OVP (overvoltage protection), this probably means that it is equipped with a simple metal oxide variable resistor. These security measures by themselves are absolutely insufficient and cannot replace any kind of digital safety chip.

Unfortunately, you can't always tell at first glance whether you're dealing with a high-quality PSU or whether there's nothing but disappointment waiting behind the pleasant facade. Therefore, we decided to open up two budget PSUs representative of what you can find in many of today’s OEM PCs and illustrate the points and features you should be examining.

A First Look At the Inside: Primary Capacitor and PFC

First, look at the storage capacitors in the primary circuit. These act as buffers and help protect the PSU and computer from voltage fluctuations. The electrolyte used in them is key, because it evaporates or dries out through a combination of heat and time. As a general rule, capacitor lifetime is halved for each 10 degrees Centigrade increase in temperature over the specified normal load. Using higher-quality capacitors that can handle 105 degrees instead of 85 degrees (C) should almost double their lifetime, greatly contributing to the PSU's durability.

A PSU equipped with chokes (a kind of inductor) is a clear indicator of passive Power Factor Correction (PFC). Passive PFC plays a significant role in the efficiency of the PSU. Only active circuitry allows for factors close to the optimum value of 1, while passive components can reach 0.7 to 0.8 at best, meaning they only achieve 70% to 80% efficiency. PSUs with a passive PFC may be cheaper to buy in the short run, but poor efficiency can swallow savings over time in the form of higher electricity costs.

Protection Circuits

Even without opening the PSU, a data sheet can reveal some of the safety measures taken (or not) by manufacturers. A decent PSU should contain the following safety measures:

• OCP (Over Current Protection): protection against power spikes
• OVP (Over Voltage Protection)
• OPP (Over Power Protection): overload protection, sometimes called OLP 
• OTP (Over Temperature Protection): protection from overheating 
• UVP (Under Voltage Protection) 
• SCP (Short Circuit Protection) 
• NLO (No Load Operation): this isn’t exactly protection in the same sense as the other features, but it allows the PSU to power up and function normally, even with no load.

Without this information, you have to look inside the PSU to find out what you need to know.

We found no protection at all on this unit, except for a simple fuse. Sadly, this PSU is still available on the market under a couple of different labels.

Passive components do not guarantee sufficient protection. Without a digital security chip, the computer hardware is severely exposed to risks.

The security chip PS223 from Silicon Touch is popular, and you should avoid PSUs not using it or similar products, such as the PS332S.

Cables and Short Circuits

You can tell a lot about your PSU by looking at its internal wiring. A lack of heat shrink tubing, carelessly exposed solder joints, and components fastened with a glue gun are symptomatic of cheap and hazardous manufacturing. If unprotected cables are placed next to hot components, a PSU failure is nearly assured.

Boards

A final quality indicator is the circuit board material. Impregnated laminated paper (like the yellow boards in the pictures) is a sure sign of cost cutting. Fiber materials are much more durable and, perhaps more importantly, non-flammable.

How Do You Recognize The Manufacturer?

UL Numbers From Underwriters Laboratories

"Underwriters Laboratories, an independent firm working with product safety certification, has been active in the field of product testing and preparation of safety standards for more than a century. UL evaluates more than 19,000 types of products, components, materials and systems annually. Every year more than 20 billion UL marks are placed on products from 66,000 different manufacturers. The UL Group and its network of service providers include 68 testing and certification labs worldwide, serving customers in 102 countries."--from the About UL page at www.ul.com

Essentially, this means that all PSUs sold on the North American market must be marked with an UL number. This number should identify the actual manufacturer of a product. However, not every PSU has such a number. UL number omission on a North American product might indicate poor quality.

Step 1: Reading the UL Number

To find the UL number, there's no need to open your PSU or desolder anything. Simply open your PC and look at the power supply's label. The UL number usually begins with an "E" followed by a string of numbers. See the three examples below.

Online UL Number Query:

Now let's find out more about your PSU. Go to the UL Online Certifications Directory and enter the UL number in the UL File Number field. If the number exists in the database, you should immediately see the result. If not, the number is invalid, fake, or the manufacturer no longer exists.

Link: Online Certifications Directory

PSU Makers: 0-9 And A-B

Companies: 2the Max Asia Pacific, 3Y Power Technology, Ablecom, AcBel, Achme, AeroCool, Akasa, Amacrox, AMS, AOpen, Antec, Apex, Arctic Cooling, Aspire, Astec Int. Ltd., Asus, Athenatech, ATNG, Atrix, AXP, Be Quiet!, Bestec, BFG, Boser Technology Inc., Broadway

PSU Makers: C-D

Companies: Casing Macron, Channel Well, Chenbro, Chieftec, CompUSA, Coba, Codegen, Cooler Master, Coolmax, Cooltek, Corsair, Deer, Dell, Delta, Dongguan Zhangmotou, Dynapower

PSU Makers: E-H

Companies: Eagle Tech, Elwin Technology Ltd., Enermax, Enhance, Enlight, Etasis, Forepoint, Fortron, Foxconn, Foxlink, Fujitsu Siemens, Gigabyte, GPS, GTR, Hama, HEC, Hiper, Highscreen, High Performance Group, High Performance PC, High Power, Himere, Hipro, HP, Huntkey

PSU Makers: I-M

Companies: Impervio, Inter-Tech, In Win, I-Star, ICP, Jersey, JOU JYE, JPAC, Just PC, Key Mouse, Kingwin, L & C, LC-Power, Leadman, Levicom, Li Shin, Lian-Li, Lite-On, Linkworld, Logisys, Mad Dog, Mapower, Masscool, Mean Well, MGE/XG, Mitac, MSI, Mushkin

PSU Makers: N-R

Companies: Newton, Nexus, nMediaPC, NorthQ, NZXT, OCZ, Okia, PC Power & Cooling, Point of View, Powerman, Powertek, Powmax, Raidmax, RaptoxX, Rasurbo, Revoltec/Listan, Rosewill

PSU Makers: S-T

Companies: San Hawk, Scythe, Seasonic, Seventeam, Sharkoon, Shuttle, SilenX, Silverstone, Sinan Power, Sino Tech, Sirtec, Solytech, Soyo, Sparkle, Spire, Startech, Sunbeam, Sun Pro, Sunfone, Super Flower /SFC, Tagan, Taiwan Youngyear, Thermaltake, Topower, TTGI, Tyan

PSU Makers: U-Z

Companies: Ultra, Unitek, Vantec, Win-Tact, Wintech, Wiseframe, XClio, Xigmatek, Xilence, Xion, X-Spice, Yesico, Zalman, Zippy

Conclusion

We want to stress two things in particular:

A PSU is not necessarily bad because it is cheap. But the lower the price, the more you'll likely have to skimp on safety and performance features. A high-quality PSU contains more expensive components. If you buy cheap, you might have to buy twice--or more.

A PSU is not automatically inferior because it was made by a contractor with which you aren't familiar. These companies make all kinds of products, from luxury brands to pure garbage. Again, it's a question of hitting different production targets. Is the product built for quality or aggressive cost reduction? The responsibility always lies with the buyer to recognize and reject products that are obviously unsuitable.

We will keep collecting data and update this list every once in a while, and we’d love to include your findings. Please use our feedback form to forward information that isn’t available in our listing!

The Quest For A Bigger Gun

If you care about tweaking and tuning your PC, then you already know that heat is your enemy. Too much heat in your computer means limited overclocking ability at best and premature hardware failure at worst.

There are two components in a modern computer that are responsible for the lion’s share of heat output: the central processing unit (CPU) and the graphics processing unit (GPU).

Modern graphics cards can create an exceptional amount of heat, and standard cooling systems struggle to keep temperatures low. Nvidia's GeForce GTX 480 is perhaps the most poignant reminder of the interaction between power, heat, and noise since Intel's Pentium 4 based on the Prescott core. We're using the GTX 480 as our example here because, frankly, it stands to benefit from aftermarket cooling the most.

To remedy this problem, you can bypass the reference cooling solution on your graphics card altogether and lean on an aftermarket solution designed to bring GPU temperatures far below what the standard models can accomplish. Today we’re testing three monster VGA coolers that promise to do just that: Arctic Cooling’s Accelero XTREME Plus, DeepCool’s V6000, and Zalman’s VF3000.

Let’s have a look at the vital statistics for these coolers before we discuss them in detail: