Testing GPU Power Draw at Increased Refresh Rates using the ASUS PG279Q
Subject: Graphics Cards, Displays | October 24, 2015 - 08:16 PM | Ryan Shrout
Tagged: ROG Swift, refresh rate, pg279q, nvidia, GTX 980 Ti, geforce, asus, 165hz, 144hz
In the comments to our recent review of the ASUS ROG Swift PG279Q G-Sync monitor, a commenter by the name of Cyclops pointed me in the direction of an interesting quirk that I hadn’t considered before. According to reports, the higher refresh rates of some panels, including the 165Hz option available on this new monitor, can cause power draw to increase by as much as 100 watts on the system itself. While I did say in the review that the larger power brick ASUS provided with it (compared to last year’s PG278Q model) pointed toward higher power requirements for the display itself, I never thought to measure the system.
To setup a quick test I brought the ASUS ROG Swift PG279Q back to its rightful home in front of our graphics test bed, connected an EVGA GeForce GTX 980 Ti (with GPU driver 358.50) and chained both the PC and the monitor up to separate power monitoring devices. While sitting at a Windows 8.1 desktop I cycled the monitor through different refresh rate options and then recorded the power draw from both meters after 60-90 seconds of time to idle out.
The results are much more interesting than I expected! At 60Hz refresh rate, the monitor was drawing just 22.1 watts while the entire testing system was idling at 73.7 watts. (Note: the display was set to its post-calibration brightness of just 31.) Moving up to 100Hz and 120Hz saw very minor increases in power consumption from both the system and monitor.
But the jump to 144Hz is much more dramatic – idle system power jumps from 76 watts to almost 134 watts – an increase of 57 watts! Monitor power only increased by 1 watt at that transition though. At 165Hz we see another small increase, bringing the system power up to 137.8 watts.
Interestingly we did find that the system would repeatedly jump to as much as 200+ watts of idle power draw for 30 seconds at time and then drop back down to the 135-140 watt area for a few minutes. It was repeatable and very measurable.
So, what the hell is going on? A look at GPU-Z clock speeds reveals the source of the power consumption increase.
When running the monitor at 60Hz, 100Hz and even 120Hz, the GPU clock speed sits comfortably at 135MHz. When we increase from 120Hz to 144Hz though, the GPU clock spikes to 885MHz and stays there, even at the Windows desktop. According to GPU-Z the GPU is running at approximately 30% of the maximum TDP.
Though details are sparse, it seems pretty obvious what is going on here. The pixel clock and the GPU clock are connected through the same domain and are not asynchronous. The GPU needs to maintain a certain pixel clock in order to support the required bandwidth of a particular refresh rate, and based on our testing, the idle clock speed of 135MHz doesn’t give the pixel clock enough throughput to power anything more than a 120Hz refresh rate.
Pushing refresh rates of 144Hz and higher causes a surprsing increase in power draw
The obvious question here though is why NVIDIA would need to go all the way up to 885MHz in order to support the jump from 120Hz to 144Hz refresh rates. It seems quite extreme and the increased power draw is significant, causing the fans on the EVGA GTX 980 Ti to spin up even while sitting idle at the Windows desktop. NVIDIA is aware of the complication, though it appears that a fix won’t really be in order until an architectural shift is made down the road. With the ability to redesign the clock domains available to them, NVIDIA could design the pixel and GPU clock to be completely asynchronous, increasing one without affecting the other. It’s not a simple process though, especially in a processor this complex. We have seen Intel and AMD correctly and effectively separate clocks in recent years on newer CPU designs.
What happens to a modern AMD GPU like the R9 Fury with a similar test? To find out we connected our same GPU test bed to the ASUS MG279Q, a FreeSync enabled monitor capable of 144 Hz refresh rates, and swapped the GTX 980 Ti for an ASUS R9 Fury STRIX.
The AMD Fury does not demonstrate the same phenomenon that the GTX 980 Ti does when running at high refresh rates. The Fiji GPU runs at the same static 300MHz clock rate at 60Hz, 120Hz and 144Hz and the power draw on the system only inches up by 2 watts or so. I wasn't able to test 165Hz refresh rates on the AMD setup so it is possible that at that threshold the AMD graphics card would behave differently. It's also true that the NVIDIA Maxwell GPU is running at less than half the clock rate of AMD Fiji in this idle state, and that may account for difference in pixel clocks we are seeing. Still, the NVIDIA platform draws slightly more power at idle than the AMD platform, so advantage AMD here.
For today, know that if you choose to use a 144Hz or even a 165Hz refresh rate on your NVIDIA GeForce GPU you are going to be drawing a bit more power and will be less efficient than expected even just sitting in Windows. I would bet that most gamers willing to buy high end display hardware capable of those speeds won’t be overly concerned with 50-60 watts of additional power draw, but it’s an interesting data point for us to track going forward and to compare AMD and NVIDIA hardware in the future.