An issue of variance
AMD has changed the landscape again with a driver refresh that changes the fan speeds of R9 290X and 290 cards which brings performance changes as well.
AMD just sent along an email to the press with a new driver to use for Radeon R9 290X and Radeon R9 290 testing going forward. Here is the note:
We’ve identified that there’s variability in fan speeds across AMD R9 290 series boards. This variability in fan speed translates into variability of the cooling capacity of the fan-sink.
The flexibility of AMD PowerTune technology enables us to correct this variability in a driver update. This update will normalize the fan RPMs to the correct values.
The correct target RPM values are 2200RPM for the AMD Radeon R9 290X ‘Quiet mode’, and 2650RPM for the R9 290. You can verify these in GPU-Z.
If you’re working on stories relating to R9 290 series products, please use this driver as it will reduce any variability in fan speeds. This driver will be posted publicly tonight.
Great! This is good news! Except it also creates some questions.
When we first tested the R9 290X and the R9 290, we discussed the latest iteration of AMD's PowerTune technology. That feature attempts to keep clocks as high as possible under the constraints of temperature and power. I took issue with the high variability of clock speeds on our R9 290X sample, citing this graph:
I then did some digging into the variance and the claims that AMD was building a "configurable" GPU. In that article we found that there were significant performance deltas between "hot" and "cold" GPUs; we noticed that doing simple, quick benchmarks would produce certain results that were definitely not real-world in nature. At the default 40% fan speed, Crysis 3 showed 10% variance with the 290X at 2560×1440:
I found also that increasing the fan speed reduced (or completely eliminated) the variance but only at the expensive of additional cooler noise and overall power consumption.
This original intent of this new driver, Catalyst 13.11 V9.2, was actually targetting another issue that was cropping up at outlets like Tom's Hardware.
The card that AMD sent to me is a stallion. Even if you get it nice and hot before running a test, bringing it down off of that 1000 MHz “wishful thinking” spec, it’s still faster than GeForce GTX 780, and oftentimes GeForce GTX Titan. But the Radeon R9 290X I bought from Newegg is a dud. It’ll drop to 727 MHz and stay there…and the reference cooler still can’t cool it fast enough. The result is that it violates its 40% fan speed ceiling as well. The craziness, then, is that my R9 290 press board is typically faster than my R9 290X retail card. In the benchmarks, you’re going to see numbers for all three.
The claim here was that press sample cards were running at higher sustained clocks, and higher performance as a result, than the retail cards our readers were actually buying. I was currently in the process of getting retail versions of both the R9 290 and the R9 290X from retailers, not partners or AMD, when AMD starting talking about this driver fix they were working on.
Part of the problem for the retail cards appeared to be that fan speeds (measured by RPM rather than percentage) were different from card to card. This graph shows how (before today's driver) the press sample R9 290 card compared to the retail MSI R9 290 card we received today compared in fan RPM, out of the box. We also added in a "normalized" fan speed: increasing the fan speed percentage on the MSI card to 50% that closely matched what our press sample card was spinning at.
The MSI card was running at 2350 RPM under a full gaming load (driver claiming 47%) while the press sample was instead running at nearly 2550 RPM under the same 47% indication. Clearly a 200 RPM difference means a lot! It allows the Hawaii GPU to stay cooler, longer, to create higher sustained clock speeds at 95C and that will result in better overall gaming performance.
Keep in mind that even though both the MSI and press sample card shown here are reporting 47% fan speeds, they are clearly NOT running at the same speed. In fact, that delta is about 15%. It concerns me that this was able to slip through the cracks of AMD's QA process. And, be assured, this is not an isolated occurance. I have many reports of other retail cards showing similar (and occasionally lower) fan speeds.
This graphic shows the clock speed comparisons (again, BEFORE today's driver). The green line represents the press sample we were given and shows a flat 947 MHz clock rate – the maximum rated clock of the R9 290. But the R9 290 from MSI, even when running at the same fan RPM speed, was dropping to ~850 MHz after just 5-7 mins of gaming. Out of the box, the MSI card was hitting as low as 660 MHz core clock.
So what does the driver today change? I installed it with both the MSI retail card and our press sample of the R9 290 to find out.