Sign in with
Sign up | Sign in
Building A Liquid-Cooled MicroATX Gaming Monster
By ,
1. Maximum Performance From MicroATX

It's not uncommon for PC enthusiasts to gravitate toward enormous cases that sit under their desks. Not only do those enclosures command attention, but they also have the extra space for installing add-ins and keeping them nice and cool.

And yet, our own case comparisons prove that extra space doesn't always translate into better airflow. Moreover, even a majority of enthusiasts don’t add more than a trio of three hard drives to their potent builds.

With that in mind, we're going to try something different today.

Harangued by lovers of multifaceted fascias, bright lights, and big windows, we decided to see how far we could push performance, while keeping our box's footprint as unassuming as possible.

As a frame of reference, we're comparing today's effort to our recent $2000 System Builder Marathon configuration. And like that build, we begin today's tale with a lovingly-chosen shopping list.

Ultimate Performance Mini PC Components
MotherboardAsus Maximus IV Gene-Z: LGA 1155 Intel Z68 Express $170
GraphicsPNY XLR8 VCGGTX580XPB-LC-CPU GeForce GTX 580
PNY XLR8 VCGGTX580XPB-LC GeForce GTX 580 (for SLI)
 $1080
ProcessorIntel Core i7-2600K: 3.4 GHz-3.8 GHz, 8 MB Shared L3 Cache $315
MemoryG.Skill F3-17600CL9D-8GBXLD: DDR3-2200 C9, 4 GB x2 (8 GB) $300
System DriveCrucial CT256M4SSD2CCA 256 GB, SATA 6Gb/s SSD $380
Storage Drive2 x Seagate Momentus ST9750420AS 750 GB, 7200 RPM Hard Drive
 $80
OpticalLG WH12LS30: 12x BD-R 2x BD-RE 16x DVD±R 8x DVD+RW $80
CaseFractal Design Arc Mini $100
PowerSeasonic SS-850HT: 850 W, ATX12V v2.31, 80 PLUS Silver $120
Heat SinkPNY XLR8 / Asetek Integrated Liquid Cooling   
  Total Cost  $2625


Unlike our System Builder Marathon machines, which are all sponsored by Newegg, this build was “off the books.” We pieced it together through some very helpful press contacts and and used a few parts we simply had laying around.

Our test results prove that we still didn’t cut any corners, even if some of the parts we picked aren't for everyone.

2. The Impetus: PNY’s Liquid-Cooled GeForce GTX 580 Graphics In SLI

GeForce GTX 580 round-ups are “so last-season,” and PNY’s parts arrived just as we were putting away the white summer wear. Fashion jokes aside, we simply didn’t have any more comparisons lined up when these parts arrived in our lab, but knew we could do great things with any GeForce GTX 580 factory-overclocked by 11% (to 857 MHz).

Simplified maintenance is a key advantage sealed liquid cooling has over do-it-yourself open loops, and increased radiator size is its key advantage over standard air-cooled parts. The biggest problem with sealed systems is that they can’t be disassembled and combined. So, if you have two kits, you have to find room for two radiators. Often, that simply isn't possible.

In case you didn’t notice (we didn’t at first), PNY included two different products for this build. Its XLR8 VCGGTX580XPB-LC-CPU is equipped with two fans, a double-thick (48 mm) radiator, and a CPU cooling block, while its XLR8 VCGGTX580XPB-LC has only a single-thickness radiator, a single fan, and no CPU cooling components.

The –CPU version of this GeForce GTX 580 includes mounting hardware for AMD’s Sockets AM2/AM2+/AM3 and Intel’s LGA 1155/1156/1366, plus an extra set of larger mounting screws for an upcoming Intel socket that uses a built-in support mechanism.

3. The Smaller Footprint: Fractal Design’s Arc Mini

A 16” case is barely tall enough to host a full ATX motherboard and PS/2 form-factor power supply. Fractal Design is a little kinder to its builders however, designing the Arc Mini for microATX motherboards, while leaving extra space at the top for cooling and cable management. At 8.25” wide and 19” deep, its footprint is nearly identical to that of competing 13.8” by 11.2” cube designs.

All that glitters is not gold, or even metal for that matter. Fractal designed its Arc Mini with a high-definition brushed-finish molding into a plastic face to give the appearance of aluminum at reduced cost.

Though the front panel might not win any design awards, the steel parts behind it are just as thick and sturdy as those of the Arc Mini’s bigger siblings. Scratch-resistant textured paint is also used, as durability and longevity are hallmarks of every Fractal Design chassis we’ve sampled.

A single 120 mm fan fills one of two front-panel mounts, while another resides in the back. Similarly, the top panel’s dual 140 mm fan mount includes a single fan.

The Arc Mini is designed to host a 2 x 120 mm dual-fan radiator, which is offset approximately 2.5” from the motherboard support plate to increase clearance for tall memory modules and large motherboard heat sinks. It’s still not wide enough to hold PNY’s pair of single-120 mm radiators side-by-side, but be assured that there is a method behind our apparent madness.

4. Overcoming Overclocking Ordeals: Asus’ Maximus IV Gene-Z

This editor always tries to keep his Sandy Bridge-based CPU core voltage at or below 1.40 V to assure longevity with above-ambient cooling. That usually means choosing 1.38 V for additional safety factor, though voltage fluctuations on some lower-range boards have occasionally still blown processors up over time. Even more of that built-in “safety factor” was subsequently added to our motherboard round-ups, taking that test voltage down to 1.35 V.

Our most recent System Builder Marathon revealed that an award-winning motherboard simply wasn’t able to cope with the increase from 1.35 V to 1.38 V, and that kind of mistake wouldn’t be tolerated for this build. Asus’ Maximus IV Gene-Z comes with a larger voltage regulator.

Assuming the same amount of current per PWM phase, the Maximus IV Gene-Z’s twelve-phase regulator would surely outpace that of the seven-phase product we had problems with in the SBM. As an advanced overclocking model, we’re actually counting on the Maximus IV Gene to have better-quality parts on each of its twelve phases.

Oh, and of course the board had to support SLI, which it does without breaking the Z68’s support for Lucidlogix's Virtu software and the Intel Quick Sync video enabled by it (even with our display natively attached to the discrete graphics cards).

Most enthusiasts think of microATX as a big, fat compromise. And yet, experienced overclockers are familiar with Asus’ smaller Gene line of “Republic of Gamers” motherboards, which incorporate the same design DNA as the company's full-sized ATX boards. Asus certainly doesn’t think of its product as lacking seriousness, providing a special USB cable that allows the system to be overclocked remotely, on the fly, by another machine system.

It’s also nice to see six SATA cables included with a motherboard that supports six internal drives.

5. Lightening The Load With Crucial And Seagate

When it comes to running real-world benchmarks, SSDs don't make a big impact, which is why they're  generally given a lower priority in our System Builder Marathon series. But we know they have a much bigger experiential impact than the numbers give them credit for. So, with no budget cap on this PC, we decided to go with the parts list that'd yield top-end usability, and that includes an SSD.

Storage guru Andrew Ku’s tests have shown that we can’t go wrong with Crucial’s 256 GB m4.

Of course we couldn’t get by with just 256 GB, as this machine was made for more than just games. Video and large image files would require at least one conventional hard drive.

A pair of Momentus 750 GB drives give us the option of super-fast 1.5 TB RAID 0 or redundant 750 GB RAID 1 arrays. This 2.5” drives would increase airflow and reduce weight compared to their 3.5” siblings, and we still have a BD-R upon which to dump any accumulation of old files.

We were extremely pleased with the performance of LG’s external drive, but disappointed by its selection of interfaces. The interface problem goes away when you install a drive internally, so we were even more pleased to find the internal version for $80. BD-R performance isn’t among today’s benchmarks however, so we substituted the new drive for a used part we had laying around.

6. Busting The Remaining Barriers

What do you call half of the legendary DDR3-2200 quad-channel 16 GB kit that G.Skill sent for our motherboard testing? G.Skill calls it F3-17600CL9D-8GBXLD, with two 4 GB modules making up our desired 8 GB configuration.

We chose these modules specifically to overcome any and all overclocking limitations, even though we know that performance gains over low-cost DDR3-1600 are minimal.

At $110, there’s absolutely no doubt that Seasonic’s SS-85HT is a value-oriented or even a low-cost part. And yet, it’s still capable of cranking out 80 A across two 12 V rails. We saw the feedback on $2000 System Builder Marathon configuration, where some folks suggested that this unit might have been the cause of its poor overclocking, and one of the reasons we kept it here was to put those rumors to rest. The only two things that cheapen this part are its lack of removable cables and its unpainted steel shell (though we could have spent nearly twice as much to get a full-featured part with the added benefit of 3% greater efficiency).

One such full-featured part would have been Seasonic’s own X850. We didn’t have one, but we did have our trusty X760 handy, and decided to use it for further evaluation.

7. Hardware Installation

The Asetek-supplied installation kit included with PNY’s XLR8 VCGGTX580XPB-LC-CPU graphics card sandwiches the motherboard between a base plate and top plate. The base plate is drilled for LGA 775 in addition to 1155/1156 and 1366, but the top plate is only compatible with the later standards.

Four threaded inserts fit marked holes on the base plate, while the top plate uses snap-in plastic spacers to align mounting screws.

Four fine-thread screws hold the base plate to the top plate. The CPU water block twists into place (so leave these screws loose until the water block is positioned correctly). A thermal interface material is factory-applied to the water block, negating the need for thermal paste.

Fractal Design’s six drive trays support both 2.5” and 3.5” form factors, though smaller drives mount without noise-dampening grommets. We used three of these trays for our build and removed the center drive cage entirely for increased airflow.

Another option to increase airflow would have been to mount the center drive cage parallel to the motherboard, but doing so would have forced us to remove its trays for added graphics card clearance. The idea of mounting a cage without its trays seemed pointless to us.

PNY’s XLR8 VCGGTX580XPB-LC-CPU cooling system uses a matched pair of fans in a push-pull configuration. We removed the unmatched Fractal Design rear fan before installing PNY’s dual fans and radiator in its place.

The thinner XLR8 VCGGTX580XPB-LC radiator was then installed in the front half of the dual-120 mm radiator top mount. Had the thicker radiator been mounted on top, it would have blocked access to DIMM slots.

We may have removed the case's original rear fan, but we were still able to put it to good use. A gentle pull on the front panel snapped it away for easy access to its empty bottom fan mount, and the left-over rear fan just as easily snapped into that mount.

Removed from the top panel during XLR8 VCGGTX580XPB-LC installation, the Arc Mini’s 140 mm top fan also fits the empty vent on its side panel. Putting it there prevents the side panel from closing, however, since the rear-mounted VCGGTX580XPB-LC-CPU cooling system overlaps its mounting space.

Asus provided two USB 2.0-based Wireless N adapters to complete our build. We chose the model USB-N10 for its lower profile, even though network performance is not among today’s benchmark set.

8. Overclocking

The Asus Maximus IV Gene-Z defaults our CPU to approximately 3813 MHz, regardless of the number of active cores. But it then idles down to 1606 MHz. Enabling XMP Profile 1 prevents Intel's SpeedStep technology from spinning the processor down when it's idle. That's good news for performance fanatics, and bad news for anyone who cares about their power bill.

A core voltage of only 1.38 V provides the extra stability needed to push the CPU multiplier to 47x and the CPU base clock to 101.1 MHz.

The motherboard’s 1.375 V setting pushes our CPU to 1.38 V with ± 4 mV of fluctuation.

As with most Asus motherboards, “Load-Line Calibration” reduces voltage “droop” under full load. The Maximus IV Gene-Z takes this feature a little further by adding several steps, with its “75%” setting appropriate for our CPU voltage and clock frequency.

A stable 4.75 GHz frequency is the result.

The Maximus IV Gene-Z responds to the altered base clock by changing from 9-11-9-28 to 10-10-10-30 timings. G.Skills DDR3-2200 remains stable, even though 10 is less than 11.

Liquid cooling allows our GeForce GTX 580s to reach an incredible 950 MHz frequency, even under “combined load” testing with the CPU and one of the two graphics cards on the same radiator.

9. Test Settings
Test Hardware Configurations
 Ultimate Performance Mini PCSystem Builder Marathon $2000 PC
Motherboard
(Overclock)
Asus Maximus IV Gene-Z
LGA 1155, Intel Z68 Express
101.1 MHz BCLK
Gigbyte Z68XP-UD3
LGA 1155, Intel Z68 Express
102 MHz BCLK
Processor
(Overclock)
Intel Core i7-2600K
3.40 GHz, Four Physical Cores
O/C to to 4.75 GHz, 1.38 V
Intel Core i7-2600K
3.40 GHz, Four Physical Cores
O/C to to 4.48 GHz, 1.36 V
Memory
(Overclock)
8 GB G.Skill DDR3-2200
CAS 9-11-9-28, O/C at 1.65 V
to DDR3-2158 CL 10-10-10-30
8 GB G.Skill DDR3-1866
CAS 9-10-9-28, O/C at 1.60 V
to DDR3-1901 CL 9-10-9-16
Graphics
(Overclock)
2 x PNY GeForce GTX 580: 857 MHz GPU,  GDDR5-4212
O/C to 950 MHz GDDR5-4300
2 x EVGA 015-P3-1580-AR: 772 MHz GPU,  GDDR5-4008
No O/C
CaseFractal Design Arc MiniAntec Three Hundred Illusion
CPU CoolerPNY XLR8 Integrated Liquid CoolingCooler Master Hyper 212 Plus
Hard DriveCrucial m4 256 GB, SATA 6Gb/s SSDAdata S511 120 GB, SATA 6Gb/s SSD
PowerSeasonic SS-850HT: 850 W, ATX12V v2.31, 80 PLUS Silver
Software
OSMicrosoft Windows 7 Ultimate x64
GraphicsNvidia GeForce 280.26
ChipsetIntel INF 9.2.0.1030


We compared today’s microATX build to our previous ATX build to see how much more performance its higher clock speeds would provide.

Benchmark Configuration
3D Games
CrysisPatch 1.2.1, DirectX 10, 64-bit executable, benchmark tool
Test Set 1: High Quality, No AA
Test Set 2: Very High Quality, 8x AA
F1 2010V1.01, Run with -benchmark example_benchmark.xml
Test Set 1: High Quality Preset, No AA
Test Set 2: Ultra Quality Preset, 8x AA
Just Cause 2Version 1.0.0.2, Built-In Benchmark "Concrete Jungle"
Test Set 1: Medium Details, No AA, 8x AF
Test Set 2: Highest Details, 8x AA, 16x AF
Metro 2033Full Game, Built-In Benchmark, "Frontline" Scene
Test Set 1: DX11, High, AAA, 4x AF, No PhysX, No DoF
Test Set 2: DX11, Very High, 4x AA, 16x AF, No PhysX, DoF On
Audio/Video Encoding
iTunesVersion 9.0.3.15 x64: Audio CD (Terminator II SE), 53 minutes, default AAC format 
Lame MP3Version 3.98.3: Audio CD "Terminator II SE", 53 min, convert WAV to MP3 audio format, Command: -b 160 --nores (160 Kb/s)
HandBrake CLIVersion 0.94: "Big Buck Bunny" (720x480, 23.972 FPS) 5 Minutes, Audio: Dolby Digital, 48 000 Hz, Six-Channel, English, to Video: AVC Audio: AC3 Audio2: AAC (High Profile)
MainConcept ReferenceVersion: 2.0.0.1555: MPEG-2 to H.264, MainConcept H.264/AVC Codec, 28 sec HDTV 1920x1080 (MPEG-2), Audio: MPEG-2 (44.1 kHz, 2-Channel, 16-Bit, 224 Kb/s), Codec: H.264 Pro, Mode: PAL 50i (25 FPS), Profile: H.264 BD HDMV
Productivity
Adobe Photoshop CS5Version 12.0 x64: Filter 15.7 MB TIF Image: Radial Blur, Shape Blur, Median, Polar Coordinates
Autodesk 3ds Max 2010Version 12.0 x64: Space Flyby Mentalray, 248 Frames, 1440x1080
WinZipVersion 14.0 Pro: THG-Workload (464 MB) to ZIP, command line switches "-a -ez -p -r"
WinRARVersion 4.0 Beta 4: THG-Workload (464 MB) to RAR, command line switches "winrar a -r -m3"
7-ZipVersion 9.2: THG-Workload (464 MB) to .7z, command line switches "a -t7z -r -m0=LZMA2 -mx=5"
ABBYY FineReaderVersion 10.0.102.82: Read PDF save to Doc, Source: Political Economy (J. Broadhurst 1842) 111 Pages
Synthetic Benchmarks and Settings
3DMark 11Version: 1.0.1.0, Benchmark Only
PCMark 7Version: 1.0.4 x64, System, Productivity, Hard Disk Drive benchmarks
SiSoftware Sandra 2011Version 2011.1.17.15, CPU Test = CPU Arithmetic / MultiMedia, Memory Test = Bandwidth Benchmark
10. Benchmark Results: 3DMark And PCMark

Factory-overclocked graphics cards give the microATX system a huge performance advantage in 3DMark, which gives us the illusion that it's 3.8 GHz stock processor (running that fast because Asus' UEFI settings sneakily make it so) thrashes the full-sized SBM build’s 4.5 GHz overclock.

We noted that the full-sized system’s cards could not be overclocked by a noticeable amount beyond their stock 772 MHz setting. But the liquid-cooled cards shoot all the way up to 950 MHz with complete stability. A huge 4.75 GHz CPU frequency pushes the microATX machine’s performance even further, making it absolutely unapproachable in 3DMark by the ATX machine.

A small bump in SSD performance gives the microATX build a similarly small bump in its overall PCMark score. A better overclock is worth around three times as much performance leadership.

Overclocking doesn’t help these single-SSD machines, though older System Builder Marathon systems have realized better storage scores from improved (software-based) RAID performance.

11. Benchmark Results: SiSoftware Sandra

Standard and microATX machines show similar performance in non-overclocked trim, suggesting that both machines push the same Intel Turbo Boost ratios when memory is set to “XMP” mode. XMP mode is an overclock-oriented setting, so this type of CPU optimization is par for the course.

Overclocking is a far different story, as our full-sized motherboard had great difficulty supplying the increased voltage we needed to reach a high CPU frequency. A microATX win doesn’t surprise us, since the board had a larger voltage regulator.

Though several of our memory articles have shown little performance advantage for data rates exceeding DDR3-1600, Sandra Memory Bandwidth always shows the performance improvements that we might not otherwise be able to realize in real-world apps.

12. Benchmark Results: Crysis And F1 2010

Crysis has a history of benefiting from nearly any improved component, from drives (faster progress saves) to CPUs and memory upgrades. This is especially true at medium settings (High Quality), where the graphics cards are not taxed enough to reach their limits.

The order of performance persists at Very High quality, though we actually expected the smaller machine’s higher-frequency graphics cards to prove more advantageous.

F1 2010 shows the results we were hoping to see in Crysis: at the game’s higher test settings, the microATX machine’s higher GPU clock allows it to lead over the overclocked SBM PC’s faster CPU. Overclocking the microATX build makes it a true performance monster.

13. Benchmark Results: Just Cause 2 And Metro 2033

Just Cause 2 appears CPU-bottlenecked, with the baseline microATX system’s faster graphics outpacing the overclocked SBM machine’s higher-frequency CPU only when screen resolution is pushed to 2560x1600. This observation applies even at our highest test settings.

Conversely, Metro 2033 appears to be GPU-bottlenecked even at our lower test settings, with the baseline microATX machine's faster GPUs leading the overclocked SBM machine’s faster CPU from 1680x1050 upward. Higher details further aggravate that bottleneck as the microATX machine always leads.

14. Benchmark Results: Audio And Video Encoding

The same CPU in both machines demonstrates the same iTunes and Lame MP3 encoding times when set to the same standard frequency. The microATX machine’s superior overclock leads from there.

A slight lead among standard-frequency machines in MainConcept appears nothing more than a hiccup favoring the microATX build, whereas differences in overclocking capability are what really separate the (little) men from the (big) boys.

15. Benchmark Results: Productivity

While 3ds Max shows a slight preference for the microATX build overall, Adobe Photoshop simply reflects the differences in CPU frequency.

What we said about 3ds Max also applies to WinRAR and WinZip. We were surprised to see the lack of performance differentiation between overclocked configurations in ABBY FineReader, however.

16. Power, Heat, And Efficiency

Power consumption is where the microATX build literally sucks, as in the amount of extra current it draws from the wall. Though factory-overclocked graphics cards take part of the blame, the Maximus IV Gene-Z’s habit of disabling low idle power states when XMP memory mode is enabled also hurts the smaller machine’s overall efficiency.

While the microATX machine’s liquid-cooled graphics cards reduce GPU temperature, CPU temperatures are relatively well-matched between both configurations. We should also point out that both cases have similar airflow, in spite of the smaller machine’s radiators.

Higher graphics card frequency and better overclocking capability put the microATX system in the driver’s seat in front of the full-sized SBM build.

The reference-speed SBM machine sets the baseline in our efficiency charts, so that the chart reflects only how much better or worse other configurations are in comparison. This is done by starting off with the reference machine at 100% then subtracting 100% from the results.

Power consumption favors the bigger machine by such a large amount that it completely diminishes the microATX machine’s performance lead in our efficiency comparison.

17. Getting More Performance From A Smaller PC

Our experiment in microATX performance illustrates a few concepts that seam to have escaped many enthusiasts in recent years. The first of these is that you don’t really need a big case to achieve stellar performance.

Of course, we didn’t stop there. Every time our System Builder Marathon’s $2000 build sheet crosses its budget limit, I drop down to Antec’s Three Hundred Illusion case to save money. The rationale for that decision is that the case is sturdy enough, and ventilated  better than many competing products priced the same or even more.

The idea is to achieve high-end performance without the accompanying high-end price, and the most technically-charged argument against those builds has always been that bigger cases allow higher CPU overclocks. And yet, today we achieved a higher overclock using similar ventilation in an even smaller case.

Now, you could argue that the CPU in the smaller machine received an unfair benefit in this comparison from its liquid cooler, but we’ve already proven that liquid coolers this small have no benefit over big air coolers (on several occasions). The processor in this liquid-cooled machine is further forced to share its cooling with a graphics card, and the real cooling benefits were reserved for that high-end GPU.

In our recent System Builder Marathon, we concluded that the $2000 machine's overclocking issues were most likely caused by an improperly cooled and/or undersized CPU voltage regulator. Today, we showed that a similarly-priced microATX motherboard is capable of both hosting a larger voltage regulator and using it effectively. We also showed that any GPU overclocking deficiencies we ran into in the System Builder Marathon were a result of the lower-binned cards we chose. Picking more overclocking-friendly cards for today’s test opened up quite a bit of headroom.

We also heard some negative feedback about our previous-build’s SSD. But today’s test showed a top competing model providing only 2% better performance. While we love the upgraded SSD's increased capacity and performance, we wouldn’t sacrifice other parts of the machine to keep it within a reasonable budget. Like the DDR3-2200 memory we used in our microATX machine, a large, high-end SSD should be reserved for builders without budgetary restrictions.

Finally, several readers voiced concerns about the 850 W power supply we chose, which we readily admit isn’t the most efficient part in the world. It’s hard to argue against its 80 PLUS Silver rating for a mere $110, but we did experiment with an 80 PLUS Gold unit as well.

This is the same Seasonic X760 used in many of our reference builds, and we expected it to run at approximately 100% load in this follow-up piece. The 80 PLUS organization says that it pulls 863.6 W at full load and 88% efficiency. Our full-load test showed 875 W wall load with both GPUs and the CPU overclocked and maxed-out, so we’re pushing roughly 10 W beyond its rating.

The X760 saved 19 W at full load compared to the less efficient, lower-cost 850 W part. While that is enough power to light a small room in fluorescent glory, it’s not much by performance PC standards. We’d pick a more efficient model only if we had the money to burn. Though 850 W is our minimum recommendation for similar builds, this microATX configuration is already pushing the case’s maximum capacity with the parts we used. Anyone making room for big upgrades should upsize both their case and power supply accordingly.

All of these careful tests and considerations lead us to two conclusions. First, we were spot-on by stating that the only deficiency in the SBM $2000 PC was the motherboard’s voltage regulator, which only limited its overclocking capability. Second, it’s now obvious that most power uses can get even more performance in an even smaller size through careful parts selection.

We’d like to thank Asus, Crucial, Fractal Design, PNY, and Seagate for supplying the extra parts needed to test all of the concepts proven today. While we’re at it, we’d also like to know which of today’s parts you’d like to see in future builds.