Interesting info about power consumption with CUDA apps

Timmah!

Golden Member
Jul 24, 2010
1,510
824
136
So i found this topic on Nvidia forums, its about GTX590 and its power consumption with some scientific CUDA app...

http://forums.nvidia.com/index.php?showtopic=197878

Apparently the GPU has much lower power consumption, when under load, than expected/while playing games/testing with benchmarks. The guy speculates, this might be down to GPU not using some parts like rasterizers or polymorph engines, despite being loaded to full...

For me its pretty interesting, as i bought gtx590 for work with Octane Render, which is CUDA based as well. Unfortunately i cannot confirm anything, as i do not own the consumption measuring device, but i OCed my card to 670 and rendered with it for hours and it was rock solid....

while you have full internet of reports, how the card cant be overclocked, cause it barely manages to run on default clocks due to weak power cascade. I did not try to run any game with both cores on, nor any benchmark or stress test, so it might prove to be unstable on those clocks there.

With Octane it works like charm though, i even tried it (only for 5 minutes as i was afraid after reading those horrible stories of dead cards on stock volts) to run at 700MHz and it seemed ok, there was obvious performance improvement as well (from 5,82 to 6,03 Megasamples)... indeed 5 minutes long stress test is irrelevant, i know.

Still i have a feeling, it would run ok and it might be really down to the lower power consumption of the card at the full load than in games, as the test suggests.
 

Cerb

Elite Member
Aug 26, 2000
17,484
33
86
Not much to speculate on, there. How do they keep the power scaling so good, when the TDPs are high? By not powering, and not switching, parts of the chips.

This is nothing new in general, only new that nVidia is doing it well.

Now, being able to OC higher, if not gaming, because of power limitations...that could be an interesting phenomenon.
 

3DVagabond

Lifer
Aug 10, 2009
11,951
204
106
Keep in mind that it's not the GPU's that are the problem. They will run fine at 670, 700, or even higher. People O/C the 580 all of the time, and it's the same GPU's. It's the power stage that limits the card. The OCP also throttles the card if power draw goes too high.

Possibly, and I don't know because I haven't tested it, while doing compute tasks the card(s) throttle and that's what is reducing power consumption. Just speculating as to probable causes. Not stating that it is what's happening.
 

Timmah!

Golden Member
Jul 24, 2010
1,510
824
136
Keep in mind that it's not the GPU's that are the problem. They will run fine at 670, 700, or even higher. People O/C the 580 all of the time, and it's the same GPU's. It's the power stage that limits the card. The OCP also throttles the card if power draw goes too high.

Possibly, and I don't know because I haven't tested it, while doing compute tasks the card(s) throttle and that's what is reducing power consumption. Just speculating as to probable causes. Not stating that it is what's happening.

Well that is possible, but the guy did not overclock his card and he states the performance is expected:

Speed wise, the GTX590 is just as fast in CUDA as expected from its core count and clock rates.
The GTX480 has 480 cores at 1.4GHz, the GTX 590 has 2 GPUs each at 512 cores and 1.26 GHz (for the EVGA Classified GTX590 I have).
The run times I got show the same ratio.. the GTX590 is about 7% slower than the GTX480 (from cores alone you'd expect 5% slower, but the GTX590's RAM is slower too.)
As expected, using dual GPUs of the GTX590 is twice as fast as a single GPU. That's more an measure of my CUDA code's efficiency, but it's good to see there's no unexpected hardware slowdown.
These speed measurements are not a surprise.


Based on this i doubt the card throttled and really it would be weird if it throttled on default clocks/voltage. There is lot of bad talk about it, but its not that bad:D
 

Timmah!

Golden Member
Jul 24, 2010
1,510
824
136
Not much to speculate on, there. How do they keep the power scaling so good, when the TDPs are high? By not powering, and not switching, parts of the chips.

This is nothing new in general, only new that nVidia is doing it well.

Now, being able to OC higher, if not gaming, because of power limitations...that could be an interesting phenomenon.

Yep, if the card cant overclock higher due to OCP stepping up in case of excessive power draw (while playing games), this could theoretically mean, that less power-hungry CUDA app could allow to run your GPU faster (overclock more, given the cores are pretty capable as 3DVagabond says)