Clock Rates And Performance
Clock Rates In A Closed Case
Let’s get straight to the point: do these cards maintain the clock rates that AMD uses in its marketing materials, or do they slowly bleed performance like the reference-cooled models? For the sake of monitoring, I used the last 10 minutes of our 20-minute run into a line graph, which should tell the story.
Aside from a few drops, all of the cards do manage to sustain their rated frequencies more than 97% of the time. I can live with that, particularly since PowerTune appears to alter settings much more frantically than Nvidia's GPU Boost. This is very evident when you break power consumption down into 1 ms samples the way I did in our initial Radeon R9 290X review.
The clock rate fluctuations are likely a result of scene changes in the Metro benchmark. So, let's take a more general look at gaming performance to gauge how the cards compare.
Gaming Performance
Sapphire's Tri-X OC Radeon R9 290X maintains open-air performance levels even in a closed chassis. More specifically, the board sheds .4% of its average framerate, which is within a margin of error.
Asus' R9 290X DirectCU II OC fares worse, losing 8% of its performance in the closed case.
Gigabyte lands in the middle. Its R9 290X Windforce OC sacrifices 4% when we get it running in our Enermax enclosure. That's alright, but not nearly as good as Sapphire's effort.
Smoking Gun Or Aggressive Power Management?
Are these cards running cooler and still not as fast as we'd expect them to be? You might want to take that up with PowerTune. An analysis of the technology in 1 ms samples shows some pretty crazy fluctuations in power consumption (enough so that two cards that seemingly run at similar frequencies aren't necessarily performing the same). PowerTune's frenetic intervention is stunning indeed. The faster the fans need to spin in order to keep up with a target temperature, the more PowerTune slows things down.
And what's the impact on thermals? When Power Tune increases fan speed, it does so in such coarse of steps that one bump up might register as a drop in temperature, even though the GPU is actually trying to bring its clock rate down. Unfortunately, I can't measure that rate in the same 1 ms intervals. That would certainly be interesting though, helping us shed more light onto how PowerTune does its business.
With all of this information collected, we can draw seven conclusions from our experimentation:
- A graphics card in a closed case doesn’t necessarily get hotter than one on an open test bed, so long as the enclosure is configured correctly.
- In fact, it can even run more coolly if the intake, exhaust, and pressure are right.
- Generally, though, you need faster fan speeds to maintain the same GPU temperature, which means more noise. Exceptions prove the rule, as illustrated by Sapphire's card.
- We observed a performance hit in a closed case, even in situations where the graphics processor's temperature was similar (or even lower). Again, PowerTune appears to be the culprit. Every change to frequency, power, and fan speed costs a bit of performance.
- Certain vendors are leaving performance on the table due to their half-baked coolers. Direct-contact heat pipes that don't touch the GPU aren't effective in cooling it. In essence, a case is going to accentuate all of the shortcomings of a given thermal solution, some of which aren't evident during open-air testing.
- Unoptimized fan cover designs negatively affect airflow, which we saw from Asus' card.
- If AMD's Hawaii GPU does get too hot, it's due to high thermal density, compounded by insufficient thermal management from AMD and its board partners. A commendable exception is Sapphire's effort.
Now, back to working on our round-up of aftermarket-cooled Radeon R9 290-series cards!
