Sign in with
Sign up | Sign in
PowerColor LCS AXR9 290X: Water Makes Hawaii Comfortable
By ,
1. Radeon R9 290X Performance Without The Noise

Our series of Hawaii-based (Radeon R9 290 and 290X) coverage caused more than just a bit of controversy, beginning with Chris’ observations concerning noise and followed by revelations that the AMD-supplied Radeon R9 290 he received for review generated benchmark scores that couldn’t be replicated by retail parts. We eventually found that variations in fan speed at the same PWM setting were part of the problem, but that the inconsistent application of thermal paste was also an issue. Taking apart your heat sink is a good way to get your warranty voided though, and AMD eventually addressed our most pressing concerns by overriding BIOS-determined fan profiles with a new driver. Even more noise was generated by the company's reference design, and simply had to be tolerated. It could only be solved by replacing the stock cooler.

Several of AMD’s partners now offer multi-fan-cooled cards that quietly dump all of the big GPU's heat into your case without the warranty worries of a do-it-yourself upgrade. Frankly, that still sounds like a bad idea to me, since extra heat affects every other component inside your case.

PowerColor has a better solution though, and it's currently selling for $800 on Newegg.

Liquid cooling helps enthusiasts move heat away from performance-oriented components to a large radiator, with full control over airflow and noise. Typically, when you build a liquid-cooled system, it's heavier, more intricate, and, of course, at risk of a leak shorting out expensive hardware. All of that is to say, simply, open-loop coolers aren’t for everyone. But if you're confident enough in your assembly technique, PowerColor will back you up with a two-year warranty.

In addition to the custom-cooled card, anyone buying a PowerColor LCS AXR9 290X 4GBD5-WMDHG/OC gets a bundle containing two 3/8” and two 1/2" line fittings, matching hose clamps, replacement seals, a mini 6 mm Allen wrench, a six-to-eight-pin PCIe power adapter, and a Battlefield 4 redemption code.

2. An EK Block And Custom Clock Rates

Most enthusiasts associate water cooling with overclocking. Given that this is indeed an enthusiast-oriented product, PowerColor must have felt obligated to overclock its LCS AXR9 290X from a stock clock rate ceiling of 1000 MHz up to 1060 MHz. But in order to realize that speed-up, you need to change the stock Quiet firmware to PowerColor's customized Uber mode using the switch on top of the card.

Of course, Quiet mode is named as such because it corresponds to a lower fan speed ceiling. This is a liquid-cooled model that runs silently, so it makes more sense to call the shipping BIOS Default mode instead. In any case, remember to flip the switch for maximum performance.

PowerColor uses a custom-engraved version of EK’s FC R9-290X cooler in the Copper/Acetal version. EK still applies its own logo, assuring customers of the quality within. Fittings and locations are compatible with EK’s other parts, including its EK-FC Terminal multi-card cooling links.

The EK website indicates that the matching aluminum back cover isn’t required. But that cover provides another excellent place for PowerColor to brand its board. It also improves overall appearance, while protecting the components beneath from minor impacts.

Switching to an angled view of the I/O panel allows us to see all of the Radeon R9 290X’s standard connectors as we evaluate the card’s overall thickness. The cooler is thin enough to allow single-slot spacing, but the card’s extra DVI connector prevents that configuration.

3. Test System And Benchmark Configuration
Test System Configuration
CPUIntel Core i7-4770K: 3.5 to 3.9 GHz, 8 MB Shared L3 Cache, LGA 1150
Overclocked to 4.50 GHz, 1.25 V at 100 MHz BCLK
MotherboardAsus Z87 Pro: Intel Z87 Express, UEFI 1707 (12/13/2013)
Reference GraphicsAsus R9290X-4GD5 Retail Boxed: 1000 MHz GPU, 4 GB GDDR5-5000
RAMMushkin Redline Ridgeback 997121R 16 GB Dual-Channel Kit
2 x 8 GB XMP-2133 CAS 9-11-11-28, 1.65V
Hard DriveSamsung 840 Pro MZ-7PD256, 256 GB SSD
CPU CoolingThermalright MUX-120 w/Zalman ZM-STG1 Paste
SoundIntegrated HD Audio
NetworkIntegrated Gigabit Networking
PowerSeasonic X760 SS-760KM: ATX12V v2.3, EPS12V, 80 PLUS Gold
System Software
OSMicrosoft Windows 8 Professional RTM x64
GraphicsAMD Catalyst 13.12

CPU bottlenecks inhibit proper graphics analysis, and so we're overclocking Intel’s Core i7-4770K to 4.50 GHz to squash any platform-based limitations. Asus’ Z87-Pro gets us there easily, at low voltage, and at a low enough price to win a value award.

With CPU bottlenecks out of the way, RAM becomes the next concern. Though high-density DIMMs usually have trouble hitting aggressive data rates at reasonable timings, Mushkin’s 997121R gives us the best of all three worlds. Yes, that made it another award winner.

Liquid-cooled graphics cards require a separate pump and radiator. Swiftech’s old MCP350 pump still provides the volume and pressure needed for a high-end single device, its MCR220-QP radiator still provides an integrated reservoir to simplify our cooling loop, and a pair of its MCB-120 spacers allow the other parts to be screwed together as a hang-able external unit. Two of Swiftech’s fans complete the package.

Game Benchmarks
Arma 3
Version 1.08.113494, 30-Sec. Fraps "Infantry Showcase"
Test Set 1: Standard Preset, No AA, Standard AF
Test Set 2: Ultra Preset, 8x FSAA, Ultra AF
Battlefield 4Version 1.0.0.1, DirectX 11, 100-Sec. Fraps "Tashgar"
Test Set 1: High Quality Preset, No AA, 4X AF, SSAO
Test Set 2: Ultra Quality Preset,  4X MSAA, 16X AF, HBAO
Far Cry 3V. 1.05, DirectX 11, 50-sec. Fraps "Amanaki Outpost"
Test Set 1: High Quality, No AA, Standard ATC., SSAO
Test Set 2: Ultra Quality, 4x MSAA, Enhanced ATC, HDAO
F1 2012Steam version, in-game benchmark
Test Set 1: High Quality Preset, No AA
Test Set 2: Ultra Quality Preset, 8x AA
Metro Last LightSteam version, Built-In Benchmark, "Frontline" Scene
Test Set 1: DX11, Med Quality, 4x AF, Low Blur, No SSAA, No Tesselation, No PhysX
Test Set 2: DX11, High Quality, 16x AF, Normal Blur, SSAA, Tesselation Normal, No PhysX
Tomb RaiderSteam version, Built-In Benchmark
Test Set 1: High Quality Preset (8x AF, FXAA), Motion Blur, Screen Effects
Test Set 2: Ultimate Quality, (16x AF,  FXAA), Tesselation, TressFX
Synthetic Benchmarks
3DMark ProfessionalVersion 1.1, SystemInfo 4.17.0.0, Fire Strike Benchmark (Extreme Off/On)
4. The Definition Of Insanity

Four days into testing, I contacted Chris to let him know I was writing up our review of PowerColor’s LCS AXR9 290X. I also sent him a copy of the benchmark data, and that’s where the trouble began.

Cool all year long and isolated from noise, an underground shelter might sound like the perfect place for a benchmark lab. It's even better with a building on top of it, which absorbs some heat in the summer. And using one of the heating ducts from the top floor appeared to be a great plan as well. The only problem is that an unreasonably cold winter in Michigan leaves me unable to heat that large space above 15° C (59° F).

Similarly, 4000 W at the breaker should be plenty of power for my office. But that also doesn't leave a lot for a dedicated electric heater. And there's not enough ventilation for a portable gas heater, either. I was eventually able to isolate the work space enough to push it to a range from 18° to 19° C.

The card I originally planned to use for this comparison, AMD's press sample, happily ran at full speed through all of my benchmarks at those low temperatures (which shouldn't have come as a surprise, given its strong performance in Angelini's Bakersfield, CA lab). Chris suggested that I toss two days worth of work and start over with a retail sample.

Well, of course, my office space remained fairly frigid, so the retail sample ran just as fast. But I was on an open-air test  bench, which we know doesn't reflect the thermals experienced by an enthusiast in the real world. Chris sent me a link to Igor’s article and suggested a closed case.

As air temperatures climbed into the 20s (Celsius) inside the case, I was comfortable knowing that my test results would at least be heat-affected enough to represent what most folks would see in a chassis with plenty of airflow. And if that doesn't reflect your real-world uses, we’ve seen a few extremely-quiet, well-ventilated alternatives.

5. Results: F1 2012 And Tomb Raider

Unlike our canned Metro: Last Light benchmark, F1 2012 and Tomb Raider don’t generate their own frame rate-over-time charts. With fewer images on a page, I bundled both tests together.

High Quality isn’t an adequate preset for gauging high-end graphics performance in F1 2012, since the game is either CPU- or RAM-limited all the way through 5760x1080. That's surprising given our 4.5 GHz CPU and DDR3-2133 CAS 9 memory. But it goes to show how fast these graphics cards are.

Given the tight bottlenecks at other settings, I find F1 2012’s results at 5760x1080 using the Ultra preset incredible. I simply can’t believe that air-cooled card is being pushed hard enough to throttle back at this setting.

Tomb Raider’s overall performance reflects some of the differences we'd expect to see between air- and liquid-cooled cards at low ambient temperatures, though a broader spread at lower resolutions doesn't make as much sense.

The good news in that Tomb Raider is smooth on a single Radeon R9 290X all the way through 5760x1080. My test notes show that minimum performance doesn't drop below 34 FPS on any of the tested configurations.

6. Results: Arma 3

The 60 MHz advantage wielded by PowerColor’s LCS AXR9 290X shows up at 4800x900 and 5760x1080 using Arma 3’s standard quality preset. At 1920x1080, the results aren’t as consistent.

Minimum frame rate isn’t a problem when you play at this title's standard preset using such powerful graphics hardware.

Arma 3’s Ultra preset appears to make 1920x1080 results somewhat more relevant, though.

Performance drops to around 30 FPS for all four configurations at 5760x1080, raising the question of whether the game can actually be played smoothly at this combination of high resolution and the Ultra detail preset.

Our frame rate over time chart shows that 5760x1080 really is playable on a single card, if barely.

7. Results: Battlefield 4

The PowerColor LCS AXR9 290X's clock rate advantage again barely outpaces the air-cooled card at low ambient temperatures, which are ideal, but rarely faced by folks in warmer climates. And dropping to the card's Quiet mode appears to trigger slight throttling in this title.

So why ship this liquid-cooled card with AMD's Quiet firmware selected by default? We're not certain; what we do know is that this switch uses the unaltered firmware from the original air-cooled Radeon R9 290X.

Gamers looking for minimum frame rates around 40 FPS will want to use PowerColor’s Uber firmware and the High preset, while those looking to maximize the title's graphics detail will likely prefer the Ultra setting.

At an average of 33 FPS, we have to start wondering if the Ultra detail level at 5760x1080 is a reasonable combination for playable performance.

The good news is that minimum frame rates aren’t much lower than averages in our test. The LCS AXR9 290X's 27 FPS floor is largely acceptable as we proceed through this taxing graphics-bound benchmark sequence.

8. Results: Far Cry 3

Far Cry 3 responds a little more eagerly to PowerColor’s mild overclock, and the lower temperatures appear to help it through the game, even when using AMD’s Quiet firmware (the card's default switch setting).

Minimums are close to averages in this title, and PowerColor’s top-performing factory-overclock maintains its lead throughout.

I personally prefer averages above 30 FPS and minimums above 20 FPS. At 5760x1080 and Ultra quality, Far Cry 3 takes the average frame rate of all three cards dangerously close to what I'd consider playable.

Frame rate over time shows that only the liquid-cooled card can keep its nose above 20 FPS.

The 60 MHz-overclock that PowerColor applies through a BIOS selector barely helps at these settings.

9. Results: Metro: Last Light

The LCS AXR9 290X maintains its small lead in Metro: Last Light, appearing to perform well across all resolutions.

That lead remains as we increase quality settings, though 5760x1080 begins to stutter as frame times exceed visual fluidity.

Dropping back to 4800x900 while using our higher test settings, we find that the air-cooled card stays above 20 FPS except for two frames.

The LCS AXR9 290X runs into a similar issue, though it manages to maintain at least 20 FPS.

10. Power, Heat, And Efficiency

In today’s test, the liquid cooling system was powered by the platform, pulling around 12 W continuously (18 W at most). Hot cards consume more energy, and that helps explain why the LCS AXR9 290X uses less power, even with the extra cooling parts added to system load.

With AMD’s target temperature of 94 °C, full-load temperatures don’t mean much for air-cooled cards. This chart only indicates an ambient temperature of 18.5 to 19.4 degrees, where I’ve already noted the 18-19 °C temperature as a potential consistency issue.

Temperatures for the liquid-cooled card are exceptionally low. And PowerColor's solution, on its own, pulls around 30 W less as a result (though it gives some of that efficiency back to power the pump).

Now that AMD is overriding the Quiet firmware settings in its driver, it makes less sense that the air-cooled card would still run faster using the Uber firmware (particularly since the Uber setting merely allows faster fan speeds). Even more testing suggested that the source of the added performance was a 0.5° drop in room temperature! It's almost impossible to maintain a workspace with a tighter thermal tolerance, and it's pretty telling that these Hawaii-based boards are so sensitive.

The only performance gains that do make clear sense happen to the LCS AXR9 290X. It always runs cool enough to maintain AMD's peak clock rate ceiling, and rarely encounters a taxing enough load to trigger throttling. Shipping with its BIOS selector set to AMD’s default for air-cooled boards, the PowerColor card picks up a 60 MHz-higher core clock and 100 MHz-faster memory after enabling Uber mode.

Posting greater performance and lower power consumption, the liquid-cooled LCS AXR9 290X naturally achieves better energy efficiency. The reference-class Radeon R9 290X is so inconsistent (in both power and performance measurements) that I no longer wish to use it for any testing purposes, at least at its default fan settings. The only way to make the stock card perform consistently is to push its fan fast enough to maintain consistent clock rates. Of course, then it no longer behaves the way it shipped out from the factory.

11. Overclocking

Extensive testing shows that AMD necessarily sacrifices performance in order to force the fan to operate at or below specific rotational speed (almost certainly to guarantee a certain "quality" of experience). Should the Hawaii GPU heat up beyond the fan's ability to maintain safe operating temperatures, the card will increase fan speed even more to keep pace.

The company's stock cooling solution can keep temperatures low enough to allow for overclocking, but only if you're willing to accept louder-than-stock noise levels. Overclocking tools offer different types of fan control, and my personal favorite is MSI Afterburner.

With temperatures in check, our retail Asus Radeon R9 290X with reference cooling pushes a 12% GPU overclock and 15% memory overclock.

The GDDR5 used on retail Radeon R9 290X cards doesn’t overclock consistently, though. I’ve seen various samples of the same model facilitate anything from 1280 to 1600 MHz (a 5120 to 6400 MT/s data rate). I shouldn’t have been surprised, then, to find this specific LCS AXR9 290X sample sustaining lower clock rates, in spite of its improved GPU headroom.

The battle between exceptional memory overclocking and a better GPU frequency played out in all of our benchmarks, and the liquid-cooled card came out on top.

12. Putting A Price On Silence

With a GPU that’s factory-overclocked by 6%, PowerColor’s LCS AXR9 290X 4GBD5-WMDHG/OC sets the example of liquid-cooled value by being priced only $100 higher than the card we compared. Yet, one of the reasons this $800 offering looks like such a good deal (relatively) is that its $700 air-cooled rival is currently overpriced. We'd hope to see the gap between them widen as the prices on 290X cards fell back to AMD's launch prices, but we're not holding our breath, either.

Price volatility aside, our LCS AXR9 290X sample was also able to reach a 1200 MHz overclock silently. The optimized fan profile required to hold the air-cooled card at 1120 MHz pushed system noise up from 31 dB (stock) to 45. And that's a best-case scenario using Nanoxia’s Deep Silence 1 enclosure to stifle noise, in an 18 to 19 °C room. I pity the enthusiast in a warm environment.

A 6% performance improvement is hardly worth a 15% price increase, but the above chart also includes the $180 worth of liquid-cooling hardware that we added to make the LCS AXR9 290X function. If you already own a large enough loop to add this card, you’ll find value parity between PowerColor's $800 board and the $700 air-cooled retail card when you add them both to a $1200 system.

Amazingly, the price-per-performance of the card alone is similar to its value when added with a $180 pump and radiator combo to a $1200 PC.

The real added value of liquid-cooling a Radeon R9 290X doesn't necessarily come from increased performance though, but decreased noise. You can get most of the speed-up by simply turning up the clock rates and cranking the fan speed on an air-cooled card to annoying levels. Unfortunately, the worth of less noise depends on the liquid-cooling components you're using, making any attempt to quantify this completely arbitrary. Our charts are the closest approximation we can find without dollars-to-noise compensation.

The liquid-cooling sink used on PowerColor’s LCS AXR9 290X is definitely worth something over AMD’s maligned reference heat sink and fan, but is the card ultimately worth $800? We believe that all Radeon R9 290X graphics cards are currently overpriced, unless you’re making money off them in crypto-currency mining, in which case there's a good chance you're getting hammered on power costs. Really though, any Hawaii-based card has the potential to re-emerge a value leader once supply catches back up to demand. 

Sidestepping the question of value, which market forces affect on an almost-daily basis, we find that it's better to ask if we'd spend an extra 14% to rid ourselves of the reference card’s heat and noise issues. The answer is an easy yes.