The EVGA 980 Ti Hybrid is the key differentiator for this benchmark. Both the Fury X and 980 Ti Hybrid utilize CLCs for thermal dissipation, but execute their cooling solutions in a markedly different fashion.
In the case of the R9 Fury X, AMD liquid cools its GPU, high-bandwidth memory (VRAM), and VRM with a Cooler Master CLC, defined here. The EVGA GTX 980 Ti Hybrid uses an Asetek-sourced CLC to cool the GPU, but combines (see: “hybrid”) a blower fan and alloy heatsink for VRM and VRAM cooling.
Cooler Master's CLC uses FEP tubes – more rigid and plastic in design, but capable of insulating liquid for lower thermals – while Asetek's solution deploys rubber tubing. FEP tubes are less pliable than the rubber alternative and the internal teflon coating can crack when kinked or bent with a great enough force. Unlike the EVGA / Asetek solution, this is not coupled with a more traditional on-card air cooler. The EVGA unit is still considerably quieter and cooler than a 980 Ti reference cooler and, given its reliance on liquid, deserves to be checked for whine in a similar fashion to AMD's Fury X.
AMD R9 Fury X Cooling Design and Impact on Power
AMD's R9 Fury X utilizes a Cooler Master CLC for its GPU cooling. No board partner modifications are permitted with the Fury X – they're all entirely regulated by AMD.
The video card itself is roughly 7.5” in length, though the somewhat stiff FEP tubing restricts tube bending to a point of adding an effective 1” to card length. Underneath a user-serviceable shroud rests the Cooler Master pump, situated atop a copper coldplate using standard liquid cooling methodology. A copper coldplate extends into the pump block, where microfins are cooled by liquid pumped through a lower channel.
FEP tubing employs a plastic inner lining that reduces loss of liquid through interface permeation and evaporation, but – as with all technologies – comes with a few downsides. Bending the tubes beyond their minimum bend radius can cause the plastic to crack, meaning liquid loss over time that will contribute to overheating and early death of the GPU. Installing the radiator must be done with great care to ensure that the tubes aren't tweaked in any particular direction. For comparative purposes, many of the most popular CLCs on the market are made by Asetek (including Corsair's, NZXT's, the EVGA Hybrid, and others), who use a thick rubber wall that offers more flexibility and an easier install. It's our understanding that the cost of Asetek hardware is greater than what's made by Cooler Master, likely an indication of AMD's choice to opt for the FEP solution.
The radiator uses a bulked-up liquid chamber to ensure longevity of the cooling solution with a back-up supply of liquid. This combined with AMD's installation requirements, make for a somewhat irritating install process in some enclosures. It is strongly advised by AMD to install the radiator with the tubes at the lower side of the radiator, a difficult task at times given the inflexibility of the FEP tubing.
Unlike EVGA's Hybrid, AMD's R9 Fury X completely ditches on-card fans. The Fury X uses only the CLC and radiator + fan setup, using liquid to cool the VRM and VRAM; competing CLC-cooled solutions still utilize a VRM/VRAM fan with the GPU coldplate.
Using a CLC helps keep temperatures down, obviously, but has some more important value-adds that aren't necessarily as obvious. Keeping the transistors at lower operating temperatures will ensure less power leakage and ensure longer lifespan of internal components. The liquid-cooled transistors' reduced power leakage goes hand-in-hand with AMD's attempt at a smaller power envelope overall, keeping the R9 Fury X under 300W.
If it weren't for radiator install restrictions – like in situations where a user might already have a CPU CLC – I'd say that liquid-cooled high-end video cards were assuredly the way of the future. For the most part, our team does feel this is true for the current “Big GPU” architecture, but it's also not entirely necessary if enclosure cooling is efficient enough. Reducing the thermal footprint is good for the entire system, it's just that not every case allows a second radiator.
Radeon R9 Fury X Thermals at Equilibrium
Note: We use a Z97 platform for thermal testing, but an X99 platform for game testing.
This thermal chart shows the temperatures (dT over ambient of 21C) at equilibrium. This is the highest average we collected during thermal benchmarking. See page 1 for test methodology in greater depth.
The R9 Fury X is substantially cooler than its air-cooled AMD counterparts – a marked improvement over AMD's 90C+ recent past. NVidia's combination of lower TDP and Asetek CLC make for a substantially cooler load temperature.
It is worth noting that AMD's increased thermals are likely tied to the company's decision to cool its HBM and VRM with liquid, which will feed more heat into the closed system.
Radeon R9 Fury X Thermals over Time vs. 980 Ti over Time
Again, we're using dT over A for these charts. The thermals shown are over time, providing greater insight to the burn-in time and time required to meet equilibrium for each solution. We only recently started producing these charts, so everything featured is from the Hybrid review. The reference 980 Ti is also shown at clock-for-clock (C4C) ratings against the Hybrid.
AMD's R9 Fury X takes a little longer to warm-up, meaning it's cooling less aggressively during the first few minutes, but continues gaining temperatures throughout the test. Part of this is probably tied to warming of the memory and VRM. Thermals equalize toward the 16-minute mark (1000s). This isn't inherently bad – it just means the cooler doesn't attack thermals as powerfully as the nVidia solution, which should help with noise control.
R9 Fury X Pump Frequency & Noise Levels
This was discussed more heavily in our separate frequency spectrum analysis article, and we'll leave the in-depth test methodology and analysis over there. The short of this topic is as follows:
- AMD has acknowledged a high-pitched whine output from the pump of its R9 Fury X CLCs.
- AMD has done everything short of a recall, promising an immediate resolution for the next batches of retail cards. Our retail cards did not have the fix.
- We do not think the high-frequency output is a viable reason to withhold purchase, though other elements (discussed later) could be. The frequency is irritating, but mostly inaudible in a case.
The below charts show our findings:
As seen here, the EVGA Hybrid CLC emits a similarly audible whine, but is of a deeper nature. AMD's whine pierces the 20KHz range, which becomes separately audible over other PC noise emissions. Both have a similar dB level, and AMD actually has a slight dB advantage, but the less spiky nVidia frequencies are less noticeable given the more “camouflaged” spectrum.
AMD R9 Fury X Power Draw
Note: We use a Z97 platform for power draw and thermal testing, but an X99 platform for game FPS testing.
Our power draw analysis is defined in the methodology section. The below chart shows peak system power draw (averaged at equilibrium to eliminate outliers) when under a gaming load.
The Fury X still runs hotter than its direct competition – 379W for the Fury X vs. 353.5W for the 980 Ti (peak system load) – but is measurably better than previous AMD attempts. CrossFire Fury X cards requires a massive 663W when loaded with real-world applications, making for the hottest configuration on our chart. The SLI 980 Ti cards require a still-immense ~589W, though stand low enough to require a slightly smaller PSU.