Over the past few days there have been some claims that retail samples of the Radeon R9 290X are clocking down once they hit the 90~95 Degrees C temperature margin. Now first let me eliminate a rumor, this might have been reported first by Tom's Hardware, but they didn't discover it. This was actually mentioned in an NVIDIA presentation and from there on some editors rightfully picked up on it. It seems that some retail samples indeed clock down. Today I just have gotten my hands on a Radeon R9-290X from HIS which is a retail sample. With that sample at hand I figured that there's only one way to see if there is any validity in the claim. And yeah it is partly true, the BIOS setting needs to be at Quiet mode (which is the default). We fired up a hefty setting for Unigine Heaven and once the GPU starts heating up the clocks will go down a bit. Now I can not replicate the reports that are rumored on other websites though, some are even reporting drops as low as 750 MHz. Find out more after the break.
Our sample however starts to clock down to as low as 850 MHz coming from 1 GHz. You'll notice that the first two minutes rendering performance sticks at roughly 1 GHz as expected, and then after a while the power or temperature limiter seems to kick in with the result being a noticeable drop on the core clock freuqency. This could mean a drop of 5% in performance quite easily. Have a peek at what we measured with our retail sample. Also keep in mind that this behavior is pretty much non existent in the BIOS 'Uber' setting mode. There the fan RPM reacts and increases (with more noise as a result). Btw the tiny larger dips are scene changes in Heaven or that's me firing up software / screenshots.
Right now I have the demo looping for an hour making the card as hot as can be, we'll add another chart later on to see how much the core clock frequency drops.
And here a second update with the GPU still up and running making its demo loops. The core clock drops to 840 at worst, again the small dips are scene changes. So yeah, there definitely is some validity to the claim. A drop of 160 MHz with 125 MHz on average is quite a lot. This is driver fixable though as I am confident that AMD simply opted to not increase fan RPM to preserve on noise levels (hence this happens in quiet mode). Instead they are lowering the clock dynamically to adapt to a certain heat and power requirement.
So with the GPU really heated up I started another 3DMark FireStrike session to see if there was a noticeable performance score. It's really not that bad it seems and some websites are blowing this way out of proportion. But hey, we don't judge. We just report what we can measure and see. This might be simply the tradeoff you have for a more silent preference. And if you don't care, in uber BIOS mode you'll get some more performance. Now also please keep in mind that this is happening with the reference products only, once the board partners apply better 3rd party cooling this will become a non-issue.