Message boards :
Number crunching :
Long GPU units again?
Message board moderation
Author | Message |
---|---|
Brkovip Send message Joined: 18 May 99 Posts: 274 Credit: 144,414,367 RAC: 0 |
I am getting some GPU units that are taking over 3 hrs. Did we make up some new type of units that are really taking that long on a GTX 480? Is anyone else noticing this too? |
Brkovip Send message Joined: 18 May 99 Posts: 274 Credit: 144,414,367 RAC: 0 |
I am getting some GPU units that are taking over 3 hrs. Did we make up some new type of units that are really taking that long on a GTX 480? Is anyone else noticing this too? Interesting, I shut off SLI and the same units go back to normal speeds. This using the 266.58 drivers and 6.10.58. |
aaronh Send message Joined: 27 Oct 99 Posts: 169 Credit: 1,442,686 RAC: 0 |
Are you talking about ones like this? For some reason, it's showing as "Anonymous platform (NVIDIA GPU)" but it's running the CPU application. Have you modified your app_info.xml recently? (you updated while I was typing this) Ah, yes, I believe SLI has been known to cause issues. Still, I'd check that your GPU WU's are really using the GPU |
perryjay Send message Joined: 20 Aug 02 Posts: 3377 Credit: 20,676,751 RAC: 0 |
Well, glad you found your answer. I have run into a few that are taking close to an hour for me on my 450 so I do believe we do have some kind of new WUs out there but I've never seen any taking three hours. I have also seen others complaining about long work units, I will remember this post and be sure to ask them about SLI next time I see one. Aaron, that's called CPU fallback and it is usually caused by not enough memory on the GPU or so it says. Maybe SLI is causing the GPUs to show a false memory reading. PROUD MEMBER OF Team Starfire World BOINC |
Brkovip Send message Joined: 18 May 99 Posts: 274 Credit: 144,414,367 RAC: 0 |
Are you talking about ones like this? Yeah, they are using the GPU, it is easy to tell with EVGA Precision and I have run the Rescheduler to make sure they are right. |
aaronh Send message Joined: 27 Oct 99 Posts: 169 Credit: 1,442,686 RAC: 0 |
Aaron, that's called CPU fallback and it is usually caused by not enough memory on the GPU or so it says. Maybe SLI is causing the GPUs to show a false memory reading. I'm pretty sure CPU fallback doesn't cause it to run a completely different application. (The task I linked ran with "Version info: SSE4.1 (Intel, Core 2-optimized v8-nographics) V5.13 by Alex Kan" while his other GPU tasks ran with "Multibeam x32f Preview, Cuda 3.0") However, he has mentioned using the rescheduler, and that can change what application a WU runs with. |
perryjay Send message Joined: 20 Aug 02 Posts: 3377 Credit: 20,676,751 RAC: 0 |
Well, usually there is a mention in the STderr about falling back to CPU but in the case you are linking to is the optimized CPU app. It looks like that WU was actually a CPU task. After looking closer, I see it does show as a GPU originally, maybe he has his reschedule set to automatically send VHARs to the CPU or moved it himself. PROUD MEMBER OF Team Starfire World BOINC |
Brkovip Send message Joined: 18 May 99 Posts: 274 Credit: 144,414,367 RAC: 0 |
Well, usually there is a mention in the STderr about falling back to CPU but in the case you are linking to is the optimized CPU app. It looks like that WU was actually a CPU task. You are correct, I send VLAR's and VHAR's to the CPU. The rescheduler runs every 24 hrs normally but when I see tasks taking a very long time I run it to verify something hasn't slipped through. |
ReiAyanami Send message Joined: 6 Dec 05 Posts: 116 Credit: 222,900,202 RAC: 174 |
I noticed the same. I'm not using any extra software. Just SETI@home Enhanced 6.03 windows_intelx86 and SETI@home Enhanced 6.10 windows_intelx86 (cuda_fermi). EVGA precision tells me that GPU's are used (actually GPU usage is higher than usual: ~75% compapred to ~50%). Usually GPU calculations take anywhere between 3 to 14 min, but now 80 to 100 min on GTX-475 (2 units SLI). |
aaronh Send message Joined: 27 Oct 99 Posts: 169 Credit: 1,442,686 RAC: 0 |
ReiAyanami, I definitely see a couple VLARs that got sent to your GPU somehow. 29jn10aa.17756.4975.11.10.3.vlar 29jn10aa.17756.4975.11.10.1.vlar 29jn10aa.17756.4975.11.10.6.vlar 29jn10aa.17756.4975.11.10.7.vlar |
Miep Send message Joined: 23 Jul 99 Posts: 2412 Credit: 351,996 RAC: 0 |
Aaron, that's called CPU fallback and it is usually caused by not enough memory on the GPU or so it says. FYI CPU fallback is disabled in x32f. The poor WUs just error out - like this one CPU fallback only works in stock and you would see stderr output on the lines of: setiathome_CUDA: CUDA runtime ERROR in device memory allocation (Step 1 of 3). Falling back to HOST CPU processing... and really long runtimes because boinc doesn't know a CPU has been taken over. I definitely see a couple VLARs that got sent to your GPU somehow. oops. not good. anybody else seeing .vlar marked tasks on their GPU? Carola ------- I'm multilingual - I can misunderstand people in several languages! |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14653 Credit: 200,643,578 RAC: 874 |
oops. not good. anybody else seeing .vlar marked tasks on their GPU? Not here. A couple (correctly) sent to the host I allow CPU crunching on, but no sign of any in the GPU queues. But I agree, it would be a bad sign. We should watch, just in case. |
kittyman Send message Joined: 9 Jul 00 Posts: 51468 Credit: 1,018,363,574 RAC: 1,004 |
oops. not good. anybody else seeing .vlar marked tasks on their GPU? Just checked the 2 rigs I have running right now, and no VLAR in the GPU caches there so far. The other 6 rigs are shut down for the day because my electric rates are higher during the day. I can check them later tonight when the come back online. I seem to recall there were some situations where resend VLAR work would sometimes slip through the filter that keeps the scheduler from sending VLAR to GPU work requests. I don't know if that is still happening or not, but it could be. "Freedom is just Chaos, with better lighting." Alan Dean Foster |
James Sotherden Send message Joined: 16 May 99 Posts: 10436 Credit: 110,373,059 RAC: 54 |
I see some vlars in my CPU cache but none so far in GPU. But i think Mark is right There might still be some old unmarked vlars from a time out or uncruncehed work units still floating around from months ago. [/quote] Old James |
Miep Send message Joined: 23 Jul 99 Posts: 2412 Credit: 351,996 RAC: 0 |
The ones Aaron linked are not unmarked and they are not resends. Carola ------- I'm multilingual - I can misunderstand people in several languages! |
arkayn Send message Joined: 14 May 99 Posts: 4438 Credit: 55,006,323 RAC: 0 |
|
dan Send message Joined: 18 Oct 02 Posts: 392 Credit: 25,046,383 RAC: 0 |
oops. not good. anybody else seeing .vlar marked tasks on their GPU? I had a couple vlars show up on cuda. Caught them and changed them over to cpu. Dan |
Viciente Send message Joined: 21 Jan 11 Posts: 228 Credit: 326,384 RAC: 0 |
oops. not good. anybody else seeing .vlar marked tasks on their GPU? everything seems ok here; s@h enhanced 6.03 ~ 1:50:00, astropulse 5.05 ~ 84:00:00, s&h 6.10 cuda ~ 00:30:00 (2x gtx460 sli) .. rgds. |
kittyman Send message Joined: 9 Jul 00 Posts: 51468 Credit: 1,018,363,574 RAC: 1,004 |
I just measured all of my workunits. None of them are longer than any other. They are all exactly 5.25 centimeters, or 2.1 inches. ROFLMAO... "Freedom is just Chaos, with better lighting." Alan Dean Foster |
dan Send message Joined: 18 Oct 02 Posts: 392 Credit: 25,046,383 RAC: 0 |
I just measured all of my workunits. None of them are longer than any other. They are all exactly 5.25 centimeters, or 2.1 inches. Nice one Sten. |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.