Again, GTX460 768MB.

Before the newest releases, I was seeing a 24hr average of 87% GPU usage. I am now seeing almost 97%, running 2WU's.

Nice work gents.

---

Indeed :) This is what I've attributed my decrease in WU time to. My GTX216 has had ~10% solid increase in GPU load, and it's nowhere near as spiky as it was before. Very consistent and never drops below 80% usage with 1 WU (GTX2xx can't handle more than 1 WU at a time :( )

I've got a GTX460 too, so that is interesting to me. I guess this would reduce or eliminate the need to run Fred's Priority tool, then? Does this benefit only apply to Fermi cards or is this a general improvement for all CUDA cards? (I haven't updated yet)

Yeah I'm seeing the same on my GTX480. With three WUs going I'm getting 97% usage and 933MB of vRAM being used out of 1536! Getting better! Now if they could just get it to where 1 WU would use 100% or so on these cards and really speed the time up we would be golden!

---

Traveling through space at ~67,000mph!

You all talk about x32f_cuda30 vs. x38g_cuda32 apps?

And GTX4xx+?


I see between stock 6.09_cuda23 (this was the latest fastest app on my system (E7600 & GTX260 OC, WinXP)) and x38g_cuda32 no GPU load difference. Hmm.. maybe it's look like x38g_cuda32 use little bit less.. but 5 MB more VRAM (260 vs. 265 MB).

But it's O.K., the x38g_cuda32 app is little bit faster than stock 6.09_cuda23 on my system. But use little bit more CPU time.

Test, 6.09/6.10/V11/V12/V12b/x32f/x38g


- Best regards! - Sutaru Tsureku, team seti.international founder. - Optimize your PC for higher RAC. - SETI@home needs your help. -

---

You all talk about x32f_cuda30 vs. x38g_cuda32 apps?

And GTX4xx+?


I see between stock 6.09_cuda23 (this was the latest fastest app on my system (E7600 & GTX260 OC, WinXP)) and x38g_cuda32 no GPU load difference. Hmm.. maybe it's look like x38g_cuda32 use little bit less.. but 5 MB more VRAM (260 vs. 265 MB).

But it's O.K., the x38g_cuda32 app is little bit faster than stock 6.09_cuda23 on my system. But use little bit more CPU time.

Test, 6.09/6.10/V11/V12/V12b/x32f/x38g


- Best regards! - Sutaru Tsureku, team seti.international founder. - Optimize your PC for higher RAC. - SETI@home needs your help. -

On my system with the 2X GTX 480's if I ran 3 units on each card I wouldn't really see a speed improvement either. If I ran 2 per card then I saw the improvement because the system wasn't pegged at 99% GPU usage all the time.

---

On my system with the 2X GTX 480's if I ran 3 units on each card I wouldn't really see a speed improvement either. If I ran 2 per card then I saw the improvement because the system wasn't pegged at 99% GPU usage all the time.

I'm seeing an improvement while running 3. Nothing to write home about but it is faster and is using a higher amount of vRAM. Working wonderfully here.

---

Traveling through space at ~67,000mph!

On my system with the 2X GTX 480's if I ran 3 units on each card I wouldn't really see a speed improvement either. If I ran 2 per card then I saw the improvement because the system wasn't pegged at 99% GPU usage all the time.

I'm seeing an improvement while running 3. Nothing to write home about but it is faster and is using a higher amount of vRAM. Working wonderfully here.

If I would have a GTX4xx+ GPU I would let run 1 WU/GPU first.
Then I would look with help of e.g. GPU-Z how high the GPU load is.
Then I would increase to 2 and I would look again to the GPU load.
If I would see now ~ 95 - 100 % average usage I guess it's enough.
If not, then I would increase to 3 and I would look again to the GPU load.
If continuously 100 %, it's too much.

The x38g_cuda32 app detect if non-/ or Fermi GPU and decide what/how to use.
On Fermi GPUs the GPU load is now higher with the new CUDA app.

From what I heard, the max WUs/GPU is now 2 @ GTX4xx and 3 @ GTX5xx with the new CUDA app.
But from machine to machine it could also vary.
Everybody should look how it's on the own machine (now)..


It's well that I have no GTX4xx+ GPU, if yes I would make a few bench-tests what would be better.. ;-)
OTOH, I guess it's currently not possible with the Lunatics bench-tool to let run 2+ WUs/GPU.


- Best regards! - Sutaru Tsureku, team seti.international founder. - Optimize your PC for higher RAC. - SETI@home needs your help. -

---

Windows Vista and 7 have a new driver model which also plays into the effectiveness of running multiple tasks on a GPU. IIRC it even allows running multiple tasks on older GPUs, though since they lack some Fermi improvements it's probably a net loss in productivity.

                                                                 Joe

From what I heard, the max WUs/GPU is now 2 @ GTX4xx and 3 @ GTX5xx with the new CUDA app.
But from machine to machine it could also vary.
Everybody should look how it's on the own machine (now)..

Good morning Sutaru,
Just wanted to brag a little bit about my GTS 450. I was running two at a time with the old 0.32f when the 0.38e came out to beta. ( I helped test it in beta at Lunatics) I noticed a dramatic speed increase on my machine so I decided to try three at a time. The third work unit slowed me back down to about the same speeds I was getting with two at a time on 0.32f. I'm showing GPU usage around 95% with drops to 85% and spikes to as high as 99%. My memory usage is staying pretty stable at 976MB. My temps have gone up from ~67c to ~72c. My RAC has gone from barely breaking 8k to now over 10k. I'm very happy with the results.

---

PROUD MEMBER OF Team Starfire World BOINC

If I would have a GTX4xx+ GPU I would let run 1 WU/GPU first.
Then I would look with help of e.g. GPU-Z how high the GPU load is.
Then I would increase to 2 and I would look again to the GPU load.
If I would see now ~ 95 - 100 % average usage I guess it's enough.
If not, then I would increase to 3 and I would look again to the GPU load.
If continuously 100 %, it's too much.

The x38g_cuda32 app detect if non-/ or Fermi GPU and decide what/how to use.
On Fermi GPUs the GPU load is now higher with the new CUDA app.

From what I heard, the max WUs/GPU is now 2 @ GTX4xx and 3 @ GTX5xx with the new CUDA app.
But from machine to machine it could also vary.
Everybody should look how it's on the own machine (now)..


It's well that I have no GTX4xx+ GPU, if yes I would make a few bench-tests what would be better.. ;-)
OTOH, I guess it's currently not possible with the Lunatics bench-tool to let run 2+ WUs/GPU.

I've already ran those tests.....that's why I'm running 3 at a time. It averages out ~2 minutes faster running 3 than running 3 units one at a time. It's faster time wise to run 2 at a time but when averaged out per unit it's slower. Only running one unit it only utilized about 25% of the gpu, 2 it's 75%, 3 it's 97% at 4 it chokes since they are fighting for clocks.

As far as limits, I know of none. I can set my 480 to run 4 at a time however it slows everything down. There are people here who have ran up to 8 work units on a 590 and fast.

As far as it varying from machine to machine you are absolutely correct, differences in operating systems, ram speeds, cpu speeds, number of cores, and usage of the machine will make results vary a lot.

---

Traveling through space at ~67,000mph!

OTOH, I guess it's currently not possible with the Lunatics bench-tool to let run 2+ WUs/GPU.


- Best regards! - Sutaru Tsureku, team seti.international founder. - Optimize your PC for higher RAC. - SETI@home needs your help. -

What would prevent it?
Just add bat or cmd file for launching few tests from different directories and all. what GPU will be used governed by -device N param that BOINC passes to GPU app.