AMD's Radeon R9 Fury has been called everything from a hot plate to a space heater. This doesn't have to be the case, though. We massively improved the card's efficiency by undervolting it.
Sapphire recently sent us a sample of its Nitro-branded Radeon R9 Fury card, and the timing couldn't be better. A few weeks ago, AMD opened the door to voltage adjustments on these cards, which prompted us to exploit the feature in a way you probably wouldn't expect. Our analysis won't focus on overclocking. After all, the card is already pushing its limits. We'll instead try to get it to a more sensible place.
Undervolting With MSI Afterburner
MSI's Afterburner tool, which is still based on Riva Tuner, now allows enthusiasts to adjust the voltage up or down on a number of AMD graphics cards that couldn't be tweaked previously. How many millivolts can be pared back before the card becomes unstable? How does that impact performance? How much power consumption can we shave off in the process?
What we're doing is often called undervolting. However, we generally prefer the more exact description, since voltage isn't actually changed; instead, the firmware is given an offset. This doesn't result in a fixed voltage adjustment, but rather a modification of how the telemetry regulates the voltage supply.
To this end, we're comparing our Sapphire Radeon R9 Fury Nitro to MSI's GTX 980 Gaming 4G. The two graphics cards offer very similar performance. Before we started our experiment, we downloaded and installed the current version of MSI Afterburner, which looks like this:
The first slider is what we're looking for—it adjusts the voltage offset. Don't be surprised if the number changes once you enter it. The figure needs to be a multiple of six. This means -100mV isn't possible so you're stuck with -96mV (-6mV x 16).
Note: Every GPU is unique. This means that the lowest number at which a processor still functions without errors can be different from one to the next. What's more, errors don't always become apparent immediately. All of a graphics card's features need to be used for a prolonged period of time to determine if the configuration is stable.
Cards made during AMD's early production cycle are particularly prone to display problems after undervolting. These GPUs were only made into Furies by a somewhat adventurous unlocking process. The newer ones are a lot easier to undervolt. However, even with these, your mileage will vary.
Our review sample managed -96mV stably enough, other than an occasional drop under extreme load. Values from -48mV to -72mV should be possible for any newer GPU. However, older cards, which we tried with the help of forum members, couldn't achieve the same low numbers. These graphics cards were almost exclusively bought right after the launch, and could have been unlocked.
When we compare the different voltage targets to MSI's GTX 980 Gaming 4G, for instance, in a Metro: Last Light loop running at 4K, then the value of our endeavor is plain to see. The Sapphire Radeon R9 Fury X Nitro's power consumption plummets from 279W all the way down to a much more moderate 213W. Of course, that only works when your GPU is willing to cooperate at such a low voltage.
If you're aggravated and want to see the same gains from Nvidia's GPU, get ready for disappointment. Lowering the GPU's voltage just isn't in the cards because it's achieved by decreasing the internal power target. Since GPU Boost is a very fragile system, every little drop has a negative impact on clock rate. In turn, this results in a massive performance hit. It's not something to complain about, per se. Nvidia simply has its mechanism optimally balanced, so there's practically no room for improvement. Consequently, MSI Afterburner doesn't even offer the option to lower the voltage. It can only be increased.
Power Consumption With Different Games
We're using our usual graphics test system, per the following table:
|Contact-free DC Measurement at PCIe Slot (Using a Riser Card)|
Contact-free DC Measurement at External Auxiliary Power Supply Cable
Direct Voltage Measurement at Power Supply
Real-Time Infrared Monitoring and Recording
|2 x Rohde & Schwarz HMO 3054, 500 MHz Digital Multi-Channel Oscilloscope with Storage Function|
4 x Rohde & Schwarz HZO50 Current Probe (1 mA - 30 A, 100 kHz, DC)
4 x Rohde & Schwarz HZ355 (10:1 Probes, 500 MHz)
1 x Rohde & Schwarz HMC 8012 Digital Multimeter with Storage Function
1 x Optris PI650 80Hz Infrared Camera + PI Connect
|Intel Core i7-5930K @ 4.2GHz, Water-Cooled|
Crucial Ballistix Sport, 4x 4GB DDR4-2400
MSI X99S XPower AC
1x Crucial MX200, 500GB SSD (System)
1x Corsair Force LS 960GB SSD (Applications, Data)
Be Quiet! Dark Power Pro, 850W Power Supply Unit
Windows 10 Pro (All Updates)
|Water Cooler||Alphacool VPP655 Pump (Lowered Speed)|
Alphacool NexXxos CPU Cooler
Alphacool 24cm Radiator
2x 12cm Noiseblocker eLoop Fan @ 400 RPM
|Drivers||AMD: Crimson Edition 16.1|
Nvidia: ForceWare 361.43 WHQL
Since the curve in the graph above already shows power consumption at different levels, we're limiting the following results to just those for the -96mV maximum offset, the -48mV medium offset, the stock value and the Nvidia card's stock value. We measured all of this using a total of nine games with very unique performance requirements.
A comparison of the two competing graphics cards shows that their 4K performance is similar. Sapphire's Radeon R9 Fury Nitro does edge out Nvidia's offering, but the differences are marginal except for The Witcher 3, Shadow of Mordor and Thief.
Efficiency With Watt Per FPS
AMD's GPU can be almost as efficient as Nvidia's when the company isn't flogging it. Fiji even edges out GM204 in a couple of games, namely The Witcher 3 and Thief. This result is more than acceptable. No more hot plate!
Nvidia's GeForce GTX 980 beats AMD's Radeon R9 Fury at Full HD. However, the enthusiasts buying either graphics card are bound to use it for higher resolutions than that, either due to a monitor with a higher native resolution or downsampling. Since AMD's GPU starts pulling ahead at QHD and WGHD, our measurements should be representative of real-world results. The only exceptions would be games lacking good driver optimization.
Lowering the voltage using a suitable tool is something to most definitely take into consideration. The fact that AMD tends to set extremely high voltage targets for its graphics cards' GPUs can almost be considered standard at this point. Its strategy does maximize the number of GPUs that can be used, especially at the beginning of its production cycle, thereby increasing yield. However, as nice as that might be for AMD's bottom line, it does hurt the company's image with customers since they're the ones measuring higher power numbers and coping with more heat.
On the next page, we'll take a closer look at undervolting and its consequences. This analysis yields some interesting results. We strongly recommend reading this part to achieve a better understanding of and appreciation for the results, even though we've already gone over them at a high level. Believe us, it's worth the read.