Implementation of liquid coolers on GPUs makes far more sense than on the standard CPU. We've shown in testing that actual performance can improve as a result of a better cooling solution on a GPU, particularly when replacing weak blower fan or reference cooler configurations. With nVidia cards, Boost 3.0 dictates clock-rate based upon a few parameters, one of which is remedied with more efficient GPU cooling solutions. On the AMD side of things, our RX 480 Hybrid mod garnered some additional overclocking headroom (~50MHz), but primarily reduced noise output.
Clock-rate also stabilizes with better cooling solutions (and that includes well-designed air cooling), which helps sustain more consistent frametimes and tighten frame latency. We call these 1% and 0.1% lows, though that presentation of the data is still looking at frametimes at the 99th and 99.9th percentile.
The EVGA GTX 1080 Hybrid has thus far had the most interesting cooling solution we've torn down on an AIO cooled GPU this generation, but Gigabyte's Xtreme Waterforce card threatens to take that title. In this review, we'll benchmark the Gigabyte GTX 1080 Xtreme Water Force card vs. the EVGA 1080 FTW Hybrid and MSI/Corsair 1080 Sea Hawk. Testing is focused on thermals and noise primarily, with FPS and overclocking thrown into the mix.
A quick thanks to viewer and reader Sean for loaning us this card, since Gigabyte doesn't respond to our sample requests.
NVIDIA GeForce GTX 1060 Specs vs. GTX 1070, GTX 1080, GTX 960
| NVIDIA Pascal vs. Maxwell Specs Comparison | ||||||
| GTX 1080 | GTX 1070 | GTX 1060 | GTX 980 Ti | GTX 980 | GTX 960 | |
| GPU | GP104-400 Pascal | GP104-200 Pascal | GP106 Pascal | GM200 Maxwell | GM204 Maxwell | GM204 |
| Transistor Count | 7.2B | 7.2B | 4.4B | 8B | 5.2B | 2.94B |
| Fab Process | 16nm FinFET | 16nm FinFET | 16nm FinFET | 28nm | 28nm | 28nm |
| CUDA Cores | 2560 | 1920 | 1280 | 2816 | 2048 | 1024 |
| GPCs | 4 | 3 | 2 | 6 | 4 | 2 |
| SMs | 20 | 15 | 10 | 22 | 16 | 8 |
| TPCs | 20 | 15 | 10 | - | - | - |
| TMUs | 160 | 120 | 80 | 176 | 128 | 64 |
| ROPs | 64 | 64 | 48 | 96 | 64 | 32 |
| Core Clock | 1607MHz | 1506MHz | 1506MHz | 1000MHz | 1126MHz | 1126MHz |
| Boost Clock | 1733MHz | 1683MHz | 1708MHz | 1075MHz | 1216MHz | 1178MHz |
| FP32 TFLOPs | 9TFLOPs | 6.5TFLOPs | 3.85TFLOPs | 5.63TFLOPs | 5TFLOPs | 2.4TFLOPs |
| Memory Type | GDDR5X | GDDR5 | GDDR5 | GDDR5 | GDDR5 | GDDR5 |
| Memory Capacity | 8GB | 8GB | 6GB | 6GB | 4GB | 2GB, 4GB |
| Memory Clock | 10Gbps GDDR5X | 4006MHz | 8Gbps | 7Gbps GDDR5 | 7Gbps GDDR5 | 7Gbps |
| Memory Interface | 256-bit | 256-bit | 192-bit | 384-bit | 256-bit | 128-bit |
| Memory Bandwidth | 320.32GB/s | 256GB/s | 192GB/s | 336GB/s | 224GB/s | 115GB/s |
| TDP | 180W | 150W | 120W | 250W | 165W | 120W |
| Power Connectors | 1x 8-pin | 1x 8-pin | 1x 6-pin | 1x 8-pin 1x 6-pin |
2x 6-pin | 1x 6-pin |
| Release Date | 5/27/2016 | 6/10/2016 | 7/19/2016 | 6/01/2015 | 9/18/2014 | 01/22/15 |
| Release Price | Reference: $700 MSRP: $600 |
Reference: $450 MSRP: $380 |
Reference: $300 MSRP: $250 |
$650 | $550 | $200 |
The Gigabyte GTX 1080 Xtreme liquid cooled card runs the highest stock clock-rate of the AIO GTX 1080s we've tested thus far, though normal Boost functionality muddies that higher clock-rate. It's this same Boost function that pushes GPUs to nearly the 2GHz mark, despite an advertised clock-rate in the 1700-1800MHz range. If there's no thermal limit, the GPU will push its clock-rate higher (without requiring manual overclocking) until another limit is reached – usually power or voltage. With the higher-end cards, like the FTW Hybrid and the Xtreme Water Force, enough power is available that Boost will often hit about 1999-2012MHz with no user-applied changes to the clock-rate.
The point isn't to say that a higher stock clock-rate is less significant, but is to illustrate that some of these devices may perform nearly identically to one another regardless of the paper spec. This same Boost functionality includes some variance from one GPU to the next, resulting in occasional swings in leaders on FPS charts (within about 1-2FPS). Ultimately, though, those swings won't much matter at this level of the playing field. We'll see that more later.
Gigabyte Xtreme Water Force Tear-Down & PCB
Our tear-down video provides a walk-through of the disassembly process, and an upcoming PCB analysis (featuring overclocker 'Buildzoid') will go live shortly on the YouTube channel. In the meantime, here's a short version of the PCB analysis:
We believe that Gigabyte is pulling 3 PWM signals from the UP9511 and using quadruplers to create its 12-phase VRM, potentially because the card's routing of video outputs to the front of the board (consuming more PCB layers).
Regardless, the cooling solution is the most interesting for today.
Gigabyte is cooling the GPU, VRAM, and VRM with its solution. The VRM cooling is somewhat unique, preceded in recent years only by the Fury X cooling design. We believe Cooler Master to be the OEM for Gigabyte; Gigabyte's Xtreme card is the only AIO cooled GPU we've looked at this generation that does not use an Asetek CLC. The CM solution uses a wide pump block with a larger coldplate surface area, and an isolated copper protrusion in the center. This protrusion contacts the GPU directly, and is able to keep the GPU diode temperature lower by isolating itself from the VRAM & VRM cooling plate.
The copper plate extending outwardly from the CLC coldplate is in direct contact with the VRAM, sinking heat there and sharing thermal dissipation with the GPU through the liquid cooler. Strictly looking at GPU diode readings, this will increase the liquid temperature and saturation of the cooling solution and increase diode readings. That's not necessarily a bad thing, since the end result is significantly cooler VRAM and a cooler VRM.
Above: Gigabyte VRAM + coldplate + VRM cooling solution.
Below: EVGA Hybrid, for comparison.
Again, the above is EVGA's cooler.
Speaking of, the VRM is also cooled by this solution. The copper plate attached to the VRAM is connected via heatpipe (soldered) to an aluminum heatsink, attached to the MOSFETs, inductors, and capacitor bank. That heatsink pulls heat away from the VRM components and transfers the heat to the heatpipe, which transfers to the copper VRAM plate, which terminates in the CLC coldplate and gets wicked away through normal liquid cooler functionality.
We believe that the outflow barb is the one located closest to the VRM, important only insofar as mitigating the impact on GPU temperature by the VRM components. It is also possible that CM is doing some flow-routing under the hood that helps strategically guide cooler liquid (“new” coolant) through the GPU first, but that seems less likely. We were unable to dismantle the cooler as this is a reader's card on loan, but maybe in the future.
The fan pushes a pretty high RPM – upwards of 3000 – because no VRM/VRAM blower fan is present on this board. Gigabyte has opted to reduce total fan count in favor of a complete liquid solution; in that regard, the Xtreme Water Force is the closest to the full coverage blocks used with open loop coolers. Noise overall is theoretically reduced by using just one fan and liquid cooling everything, but we'll talk about that more later.
PET tubing is used on the cooler, making it more rigid and similar to a bendy straw in its movement. This type of tubing has an inner coating along the plastic that helps insulate the coolant, aiding in reducing permeation and keeping lower temperatures. The downside is that PET coatings can be easily cracked if the tube is bent too much, increasing permeation and liquid loss down the line. This is contrary to the other common tubing material, a low-permeation rubber used for the EVGA and Sea Hawk coolers by Asetek. Both are commonly used materials at this point, and play the primary role of extending life of the unit by reducing permeation.
Continue to Page 2 for testing methodology.
