To call PowerColor’s air-cooled dual-GPU project daring would be an understatement. We know darned well how hot AMD's Hawaii GPU can get. Just one is enough to push most thermal solutions to their limits. That's why AMD went with liquid cooling for its Radeon R9 295X2 (a card we found to be well-built, by the way).
Now, PowerColor is trying to improve upon the first workable reference design we've seen from AMD in years with a gigantic air cooler. Is this an act of deft engineering or blind ambition?
Since AMD's sampling policy (at least over in Europe where they tested these cards) is best described as weird, we went the extra mile to make this head-to-head comparison and bought our own Radeon R9 295X2. It performs a little worse than the card we have in the U.S., but the difference is small. There's no reason to suspect anything other than normal variance between them.
Air versus water. A huge chunk of metal versus hoses and a radiator. We can hardly wait.
The three-slot bracket on the back of PowerColor's Devil 13 Dual Core is the first hint that this card means serious business. Picking it up, the thing appears built like a tank, and you have to wonder if your motherboard's PCI Express slot will take all of the weight. This is a graphics card you need to hold with two hands.
PowerColor is asking the same $1500 that AMD wants for its own Radeon R9 295X2. But the Devil 13 board attempts to differentiate itself by including a Razer Ouroboros in its bundle. That's a nearly $140 add-on. And the mouse is extremely comfortable, too. What's more, at least at the time of writing, Newegg is running a special on the Devil 13 that includes a 120 GB Samsung 840 EVO valued at $100.
|
|
| Read the Review | |
|
|
|
Before we get started, let’s take a look at the specifications we're dealing with. After all, PowerColor promises us the world’s fastest graphics card, even though the Devil 13 operates at 18 MHz below the peak clock rate of AMD's Radeon R9 295X2. But company representatives are sure they can compensate for this small disadvantage with a higher memory frequency.
| Products |
|
|
|
| Pricing |
|
|
|
| GPU | Hawaii | Hawaii | Hawaii |
| Process | 28 nm | 28 nm | 28 nm |
| Transistors | 2 x 6.2 Billion | 2 x 6.2 Billion | 6.2 Billion |
| GPU Clock Frequency | Up to 1000 MHz | Up to 1018 MHz | Up to 1000 MHz |
| Shaders | 2 x 2816 | 2 x 2816 | 2816 |
| FP32 Performance | 11.2 TFLOPS | 11.5 TFLOPS | 5.6 TFLOPS |
| Texture Units | 2 x 176 | 2 x 176 | 176 |
| Texture Fillrate | 352 GT/s | 358.3 GT/s | 176 GT/s |
| ROPs | 2 x 64 | 2 x 64 | 64 |
| Pixel Fillrate | 128 GP/s | 130.3 GP/s | 64 GP/s |
| Memory Bus | 2 x 512-bit | 2 x 512-bit | 512-bit |
| Memory | 2 x 4 GB GDDR5 | 2 x 4 GB GDDR5 | 4 GB GDDR5 |
| Memory Transfer Speed | 5.4 GT/s | 5 GT/s | 5 GT/s |
| Memory Bandwidth | 2 x 345 GB/s | 2 x 320 GB/s | 320 GB/s |
In The Box
There’s a lot of stuff in this box. Apart from the driver CD, there are two 6-to-8-pin adapters and a kind of graphics card stand that can be fastened between the Devil 13 board and chassis floor (or power supply) so the PCI Express slot doesn't have to bear all of the weight. Later, we'll get everything set up to determine if this works. Of course, then there's the previously-mentioned Razer Ouroboros gaming mouse.
A Bonus: Razer's Ouroboros Gaming Mouse
Peripherals are highly personal, so however you feel about the Ouroboros, we can't really pass judgment without an in-depth review. Regardless, this is a high-end mouse with a price to match and plenty of enthusiasts will appreciate its inclusion. Razor gives the device's body a number of configuration options, allowing it to be individualized. It's wireless, too, which comes in handy.
Dimensions and Weight
It’s important to note that the height of the PCIe power connectors and their protruding cables need to be added to the net height of the card.
| Dimensions and Weight | |
|---|---|
| Length, from Bracket to End of Cooler | 310 mm |
| Height, from Slot to Top Edge of Cooler | 132 mm |
| Depth, from Back of PCB to Front of Cooler | 55 mm |
| Depth on Back, from PCB to Top of Screw Head | 8 mm |
| Weight | 2343 g |
A massive weight of more than 2.3 kg makes PowerColor's Devil 13 the heaviest board we've ever reviewed by a wide margin.
The incredibly heavy PowerColor Devil 13 employs a true three-slot design. After all, the two separate heat sinks (one for each of the GPUs) need a lot of space. Using three 86 mm fans was quite certainly unavoidable to provide the needed airflow, but all of that cooling could also lead to audible turbulence, especially from the center fan. We’ve made a video about this that we’ll get to later. Five heat pipes per sink dissipate thermal energy quickly from the base to the fin arrays.
The board itself was developed by PowerColor in-house but, just like the reference design, uses a PLX switch for communication between the two GPUs. Power circuitry is split between both processors, also similar to AMD's reference card, enabling separate power supplies. Part of our testing covers why this doesn't work out as well for PowerColor as it does for the Radeon R9 295X2.
The frame that stabilizes the Devil 13 doubles as cooling for the power circuitry and memory packages. We have a picture of a pre-production sample, but PowerColor's retail offering is finished in black.
Around back, the Devil 13 sports another elaborately-designed frame as a backplate and counterpart to the internal frame.
Up top you'll find an oscillating red Devil 13 logo, auxiliary PCIe power connectors, and an illuminated BIOS switch.
This is going to blow your mind: PowerColor's Devil 13 Dual Core employs four eight-pin power connectors, which is to say two per GPU. That should be plenty (considering AMD only arms the 295X2 with two).
The BIOS switch activates a performance mode when it's pushed down. In essence, the fans react more aggressively, though the clock rates do not change.
There's nothing notable about the card's bottom edge. You'll find a few spots for exhaust, but that's about it.
A closer look at the back gives you a view of the copper heat pipes, as well as the stabilization and cooling frame, the cooler’s cover, and the fans.
The display outputs are standard fare, and include two digital dual-link DVI connectors, DisplayPort, and HDMI.
Measurement Apparatus and Procedure
We’re using a standardized PC platform for our performance and power consumption measurements. It has an overclocked Intel Core i7-4930K processor that’s clocked depending on the task in question at 3.6 GHz (workstation), 4.2 GHz (VGA charts), or 4.8 GHz (extreme).
We switched out our power supply after our experience reviewing the Radeon R9 295X2. The new one gives us a little more headroom.
| Test Bench | Microcool Banchetto 101 |
|---|---|
| Test PC | Intel Core i7 4930K (Ivy Bridge-E), Six-Core, Overclocked to 4.2 GHz Asus Rampage IV Black Edition 32 / 64 GB Corsair Dominator Platinum DDR3-2133 Enermax TLC 240 AOI Compact Water Cooler 1x SSD 500 GB Samsung 840 Pro 1x SSD 256 GB Samsung 840 Pro |
| Power Supply (PSU) | 1200 W, be quiet! Dark Power Pro 10 |
Power Consumption
The German power consumption test system for consumer and workstation graphics cards was developed in cooperation with HAMEG (Rohde & Schwarz) to provide the most exact measurements possible at the smallest time intervals possible. It’s capable of temporal resolutions all the way down to one millisecond.
Only sophisticated technology like this can rise to the challenge posed by the demands of AMD’s PowerTune (which can change the core voltage in a time frame of less than 10 ms) and Nvidia’s GPU Boost, and their resulting fast and large load fluctuations. Consequently, we evaluate all current and voltage measurements with a 500 MHz digital multi-channel oscilloscope, the HAMEG HMO 3054, which also allows the collection and storage of all data in one place, as well as meaningful remote control.
The measurements of the first two of the three high-resolution HAMEG HZO50 current probes are completed using a riser card that was specifically constructed by us for the 3.3 and 12 V rails. The riser card supports PCIe 3.0 and features short signal paths. The third current probe does its job via the PCIe power cable that we’ve modified accordingly.
We’re measuring the voltages directly at the PSU after modifying it a bit. For the most detailed measurements, we’re generally working with a temporal resolution of 2 ms, which is enough to cumulatively record all of the power fluctuations caused by AMD’s PowerTune and Nvidia’s GPU Boost technologies. The duration of these tests is limited to two minutes due to the large volume of data across all the measured channels. We only lower the measurement intervals to their physical minimum if we want to look at something specific in yet more detail.
| Test Method | Contact-free DC Measurement at PCIe Slot (Using a Riser Card) Contact-free DC Measurement at External Auxiliary Power Supply Cable Voltage Measurement at Power Supply |
|---|---|
| Test Equipment | 1 x HAMEG HMO 3054, 500 MHz Digital Multi-channel Oscilloscope 3 x HAMEG HZO50 Current Probes (1 mA - 30 A, 100 kHz, DC) 4 x HAMEG HZ355 (10:1 Probes, 500 MHz) 1 x HAMEG HMC 8012 Digital Multimeter with Storage Function |
ZeroCore Power is an interesting feature that represents the best solution for minimizing idle power draw in dual-GPU systems. At least in theory, the graphics card that isn’t needed can simply be turned off. PowerColor tries to illustrate the concept on the back of its Devil 13 board using two LED panels indicating each GPU's consumption.
Clearly, PowerColor put a lot of emphasis on its power handling, so we were fairly surprised by our measurements. At 36 W during idle, that's higher than we'd expect, especially considering that AMD's Radeon R9 295X2 idles more than 10 W less without ZeroCore Power. Once the 295X2 partially shuts down, it drops under 14 W (which is less than half of PowerColor’s result).
These numbers were repeated several times and the outcome is consistent: PowerColor's power supply is definitely different from AMD’s. Let’s first compare the Devil 13 Dual Core to the Radeon R9 295X2.
The first eye-opening result is that the Devil 13 draws 14.8 W from the motherboard slot alone, which is already more than AMD's dual-GPU Radeon when one of its GPUs is disabled.
Since this portion of the power draw is independent of ZeroCore, it results in an extremely high constant minimum power draw just based on component choice and layout. Conversely, AMD doesn’t use the motherboard's 3.3 V rail at all.

Now we'll compare power consumption at the two pairs of eight-pin auxiliary connectors, each of which corresponds to one GPU. The active processor draws just over 10 W through PCI Express, whereas the turned-off GPU still eats up 8 W. Tack that onto high power use from the motherboard and you end up with use reminiscent of AMD's Radeon HD 6990.
Gaming Power Consumption
Two things jump out when we compare PowerColor's Devil 13 to AMD's reference Radeon R9 295X2. First, PowerColor manages to limit the peaks in power consumption so that they’re barely above the average. Even though the Devil 13's PowerTune-based jumps are a bit more pronounced than the reference card, its sophisticated power supply smooths out the extreme peaks, relieving pressure on the PSU.

Second, it’s very noticeable that PowerColor uses the motherboard's 3.3 V rail to take care of components like the Devil 13's GDDR5 memory, whereas the reference card stays exclusive to 12 V.
Both solutions are very different. But PowerColor comes out ahead in the end. Our measurements also show that the two graphics cards have roughly the same ratio of power draw to performance. In the Devil 13's case, that means we observe lower consumption corresponding to slightly less speed.

Stress Test
Again, a direct comparison shows some differences. Even though the Devil 13 manages to sustain its clock rate in every scenario, some kind of measurable voltage clipping limits the maximum power draw to an average of 409 W (power target), which is barely higher than our gaming benchmark.
Lower compute-based stress test results confirm our findings. The power-to-performance ratio holds steady, which exposes the tricks used to avoid pushing this three-slot air cooler past its limits. Even though AMD sets a power target of 208 W per GPU (and sticks to it exactly with the Radeon R9 295X2), PowerColor's Devil 13 stays below that number.

Looking at the 12 V power supply through the four eight-pin PCIe connectors, the Devil 13's smoother curve is noticeable yet again. This consistent power draw is different from AMD’s reference design, which has been known to push PSUs beyond their limits. Certain supplies that weren't able to properly support AMD's Radeon R9 295X2 due to its peak power draw have no problem with PowerColor's board.

As for the 3.3 V rail through the motherboard slot, it's the same here as it was during our gaming workload. The wheel doesn't need to be reinvented.
Power consumption under load is okay. We observe approximately 200 W per GPU of draw, which corresponds nicely to AMD's Radeon R9 295X2 when you take performance into account. PowerColor’s card pulls a little less power, while providing a little bit less speed; we can live with this compromise.

Temperatures
The Asetek closed-loop liquid cooler AMD uses for its Radeon R9 295X2 performs as well as can be expected in spite of its miserly 120 mm radiator. It beats PowerColor's Devil 13 board without breaking a sweat. Though, that card, in turn, isn't bad for an air-cooled offering with two high-end GPUs.


We decided against testing the PowerColor Devil 13 inside of a case in order to protect our hardware. The card with the PCIe connectors on top is too tall for our reference chassis, and its included holder, the “PowerJack,” is too small for our alternative full tower, Corsair's Obsidian 900D. As you think about the hardware you'd pick to go with this graphics card, bear in mind that if PowerColor's PowerJack fits, the Devil 13 should sit securely. But you have to put more effort into coordination than most other configurations.
Fan RPM
In order to manage thermals effectively, the fans must provide a certain level of air flow, of course. This gives us a nice segue to our noise section. The fans on PowerColor's Devil 13 card behave interestingly in performance mode. A number of slow-downs are apparent in our log of rotational speed; these are reflected in the temperature curves, too.

Noise Measurement
Graphics cards noise measurements are performed using a calibrated high-end studio microphone (supercardioid) positioned perpendicular to the middle of the graphics card in question at a distance of 50 cm. This distance, in conjunction with the very strong cardioid directionality of the microphone, represents a compromise between avoiding noise due to fan turbulence and avoiding ambient noise, which can never be completely eliminated.
We performed measurements after each card reached its operating temperature in our gaming loop. Even though the Devil 13 offers both quiet and performance settings, you don't get a significantly better experience in quiet mode. The two firmware options have PowerColor's card competing with AMD's Radeon HD 6990 to become the loudest board we've tested recently.
Unfortunately, this noise level is truly unbearable, even with quiet mode activated and the card in a case. Again, we have to give AMD credit for its water-cooled Radeon R9 295X2. The Devil 13 ends up coming across as an interesting demonstration of what's possible, or what an air-cooled 295X2 might have looked like.
Comparison Videos
Finally, we'll give you a listen to how PowerColor's Devil 13 Dual Core compares to three generations of dual-GPU reference cards from AMD. The numbers don't tell the whole story, after all.
PowerColor Devil 13 Dual Core R9 290X (Quiet Mode)
Radeon R9 295X2
Radeon HD 7990
Radeon HD 6990
We're using the 2014 VGA Charts database for comparative benchmarking. This gives us a great basis for creating an index with all games and tests taken into consideration. If you want the individual test results, check out the charts section itself, where you'll find all of the individual numbers composing the index and have the option to create comparison tables.
Gaming at 1920x1080
We start by testing at FHD and maxed-out settings, even though that's well below the target market for a flagship dual-GPU board. PowerColor's Devil 13 lands right where we'd expect it relative to AMD's reference Radeon R9 295X2. That is to say the air-cooled model is just a bit slower given its more moderate power consumption. The only benchmark where the Devil 13 inches out its competition is 3DMark Fire Strike, due to the card's tuned memory subsystem. Unfortunately, you can't play a synthetic metric. But we still have to be impressed that traditional heat sinks and fans can keep up with two Hawaii GPUs.
Manually increasing the clock rate of PowerColor's Devil 13 to 1018 MHz and increasing its power target to match AMD's Radeon R9 295X2 yields almost identical benchmark results, interestingly enough.

Gaming at 3840x2160
We’re using this high resolution with optimized settings specifically chosen for our benchmark charts. There's no point in generating unplayable frame rates for all cards; then again, it's equally fruitless to lower the bar so much that higher-end cards don't have a chance to shine.
PowerColor's Devil 13 Dual Core again shows up right behind AMD’s Radeon R9 295X2, which is hardly surprising given the technical differences we've already discussed.

PowerColor proves to us that it's possible to put two Hawaii GPUs on one board, run them without a dramatic reduction in clock rate, and cool them with heat sinks and fans.
The Devil 13 Dual Core's power circuitry stays under 70 degrees Celsius as well, which isn’t bad. As opposed to AMD’s reference Radeon R9 295X2, PowerColor's creation is also free from any extreme peaks in power consumption, even under full load. This suggests that, even though the card's power supply is different, it's implemented well and ideal for handling the quick load fluctuations experienced by high-end GPUs. Smoothing out those spikes takes stress off of the power supply, which is already taxed by such high-end hardware. That's actually an important point, since graphics card vendors increasingly don't temper those big spikes. Instead, they make them the responsibility of PSU manufacturers.
There's also a lot to like about the Devil 13's accessory package, which includes a Razer Ouroboros wireless gaming mouse and the PowerJack, designed to hold up and support this massive graphics card. For a presumably limited time, Newegg is also throwing in a 120 GB Samsung 840 EVO SSD worth $100.
Unfortunately, the Devil 13 Dual Core R9 290X is not quiet. Registering more than 58 dB(A) in quiet mode and more than 61 dB(A) in performance mode, this graphics card is a conversation piece only in the sense that it doesn’t allow conversations to happen anywhere near it. No doubt, this is the same challenge Nvidia faced with its GeForce GTX Titan Z. Ultimately, that company had to turn performance down through lower core clock rates in order to make the board's acoustics acceptable.
High power consumption at idle, which tops 35 W in spite of one of the GPUs being turned off, suggests that board layout wasn't a priority when the Devil 13 was designed.
Bottom Line
PowerColor's Devil 13 Dual Core R9 290X is a limited-edition graphics cards that will find a following, despite all of the reasons to avoid it. It's really not fit for everyday use, partly because of its price, and mostly due to its noise level.
For the same $1500 you'd spend on AMD's surprisingly elegant Radeon R9 295X2, you get that improved power delivery. But then you're forced to contend with the heat of two Hawaii processors dumped in your chassis by fast-spinning fans. Sure, there's a pretty sweet mouse bundled as an extra value. However, in the context of high-end graphics, the closed-loop liquid cooler on AMD's dual-slot 295X2 is more sensible.
Still, PowerColor's innovation deserves mention. The company managed to do something different. It runs two of AMD's fastest graphics processor at full speed on air cooling without throttling back GPU frequencies.
In the end, the Devil 13 Dual Core R9 290X is mostly a technology demonstration, we'd say. It shows that two Hawaiis can be cooled by air and that they'll run without overwhelming a built-up thermal solution. But that's really all it works for. PowerColor earns recognition for going out on a limb, even though liquid cooling would have been the quieter way to cool this monster. Then again, there's already a long list of partners doing that.































