Message boards :
Number crunching :
AK V8 + CUDA MB team work mod
Message board moderation
Previous · 1 . . . 5 · 6 · 7 · 8 · 9 · 10 · 11 . . . 12 · Next
Author | Message |
---|---|
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
It is true for CPU-only units. But I was talking about the CPU-GPU units, where the difference gives an indication about GPU time. Total elapsed time gives almost the same indication. Cause it includes time when GPU awaits data (~ CPU time), OS overhead and GPU-only time. Difference between elapsed time and CPU time will not give pure GPU time anyway (there is OS overhead). Actually, pure GPU time doesn't matter. Elapsed time is matter cause exactly it defines how many tasks you can complete in 24 hours (we all live by wall clock time, not by CPU or GPU time ;) ) |
SoNic Send message Joined: 24 Dec 00 Posts: 140 Credit: 2,963,627 RAC: 0 |
Agree, but the website doesn't show the wall time. It shows the CPU processing time. I was thinking that it would be nice to have there the GPU processing time also. |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14656 Credit: 200,643,578 RAC: 874 |
Agree, but the website doesn't show the wall time. It shows the CPU processing time. I was thinking that it would be nice to have there the GPU processing time also. Agreed, but that would require changes to the BOINC framework, not just SETI - more and more projects will be persuaded to develop CUDA applications in the futre, and they should all show their GPU processing times consistently. But for the moment, BOINC are concentrating on just getting the darn thing to work at all.... |
RuthlessRufus Send message Joined: 18 Oct 07 Posts: 11 Credit: 70,386,101 RAC: 28 |
Any chance we could get an SSSE4 optimized version for us 45nm quad people? |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Any chance we could get an SSSE4 optimized version for us 45nm quad people? No chance. I have 45nm quad and benched SSE4.1 vs SSSE3 AK_v8 in close to real life conditions - SSSE3 wins. So until some more SSE4.1-specific optimizations will be added I see no sense to do separate SSE4.1 builds. And SSSE3 version provided. |
RuthlessRufus Send message Joined: 18 Oct 07 Posts: 11 Credit: 70,386,101 RAC: 28 |
So even on a CPU-only cruncher the SSSE3 version will be best? How much RAC are we talking here? |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
On 45nM quads, you'd typically see only about 1% performance difference, with the favour going to the SSSE3x build overall. Note that the build here, I believe, is a very slightly newer AKv8b build which has shifted angle range characteristics to start with, so direct comparison with any AKv8 (non-b) build won't make a lot of sense either. Also I would assume in these setups that the RAC component would be dominated by the GPU throughput, for the most part, reducing any impact even further of AKv8 build differences. v8b has not achieved significant performance improvements for most machines, to yield enough justification releasing it, as that would be kindof 'Fixing something that ain't broke', cloud the current situation rather needlessly, and the focus has been on AstroPulse development for some time (and now Cuda). With dual core 45nM chips, things tends to go the other way by only 1% favouring the SSE4.1 build, due to the less pressure on the bus (half the cores = half contention), and longer stride in the SSE4.1 build is more 'greedy'. ( but need to compare across angles ranges due to shifted logic. v8b SSSE3 here will have an advantage as Low angle ranges improvement was detected in recent tweaks. ) In other words, It's not worth worrying about for this particular package, given that these 'specials' are only really intended to bridge the gap until Boinc & project work out how to get the applications working together happily. Its' all an indication, IMO, that the specific microarchitectural optimisations (such as SSE4.1) have too little payoff at current stages of development, to justify so many builds. But things may change in the future as more uses for the different instructions and chip specific tweaks are found. That would likely be built with internal dispatch, rather than the traditional separate builds though, as having so many builds has become a maintenance problem already. (less is more, and KISS principles need to be applied in full) Jason "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
Voyager Send message Joined: 2 Nov 99 Posts: 602 Credit: 3,264,813 RAC: 0 |
It's hard to keep up with the builds. I'm running AKV8b on ssse3. Is this the latest? |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
yes. |
Dweezil Send message Joined: 29 Dec 99 Posts: 16 Credit: 31,083,717 RAC: 0 |
Hi Raistmer Installed the mod and finally got 3 tasks running. It was running before but stopped doing it this week for some reason. I tried the cc_config.xml fix but that didn't fix anything. Using 6.4.5, e6700 + 8800gts, XP Pro with more memory than I can use. It seems that everything but the astropulse keeps getting a computation error. Both the MB and AK computations seem to fail and the stderr.txt file isn't updating so I've no idea what to even try. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Hi Raistmer For CUDA app (from stderr of one of your results): Cuda error 'cudaMemcpyToSymbol(cudaAcc_GaussFit_settings, (void*) &settings, sizeof(settings))' in file 'd:/BTR/seticuda/seti_boinc/client/cuda/cudaAcc_gaussfit.cu' in line 454 : invalid device symbol. Try to upgrade driver version to at least 180.xx AK_V8 will try to start CUDA app until there is another working CUDA app exists already. As CUDA app fails constantly (wrong drivers) you will never get AK_v8 work on CPU too. You can use un-modded AK_v8 w/o CUDA app or downgrade to stock 6.08 and leave CPU for AstroPulse or other projects - all this in case you can't upgrade video driver. |
Dweezil Send message Joined: 29 Dec 99 Posts: 16 Credit: 31,083,717 RAC: 0 |
Video drivers updated to the latest Nvidia had. 181.22. As soon as I manage to get some WU's downloaded I'll let you know if there are still problems. Thanks :) |
Paul D Harris Send message Joined: 1 Dec 99 Posts: 1122 Credit: 33,600,005 RAC: 0 |
This App rocks! With my i7 oc and xmp memory and my cuda I did over 15000 credits yesterday! |
Dweezil Send message Joined: 29 Dec 99 Posts: 16 Credit: 31,083,717 RAC: 0 |
Well now I keep getting errors of a different nature. <core_client_version>6.4.5</core_client_version> <![CDATA[ <message> CreateProcess() failed - The system cannot find the file specified. (0x2) </message> ]]> And there is this weirdness. From the BOINC client. 2/12/2009 1:42:17 AM|SETI@home|Output file 21ja09ac.7515.18886.5.8.30_0_0 for task 21ja09ac.7515.18886.5.8.30_0 absent Thoughts? |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14656 Credit: 200,643,578 RAC: 874 |
Two separate issues. First, the system can't find a vital file. Since you're posting in a specialist "anonymous platform" thread, we can assume that you're using an app_info.xml file. Have a look in that file, pick out all the filenames, and make sure that every file mentioned is actually in your project folder. If there's one missing (and it looks as if there is), copy it again from the package you downloaded. Secondly, there's a file missing. So SETI can't do any work. So there aren't any results. So the file that should contain the results is missing, too (because the application that should have created it didn't run). Not weird, logical when you think about it. |
Dweezil Send message Joined: 29 Dec 99 Posts: 16 Credit: 31,083,717 RAC: 0 |
Now that is just freaky odd. The missing file (thanks Richard for pointing me in the right direction) was there prior to updating the video drivers. How AK_v8b_win_SSSE3x_GPU_CPU_team.exe managed to vanish I've no idea. |
Diamond Send message Joined: 29 Jul 99 Posts: 12 Credit: 13,474,762 RAC: 0 |
Don't mean to be lame but, i am an advanced user. Could someone go through an install proceedure to help me make sure i got this right? then maybe i could make an installation wrapper for this. I've missed something. On one machine i have 2 cudas (on 2 8800 GTXs), and one cuda on the other machines (single card) but not CPU activity on either. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Don't mean to be lame but, i am an advanced user. Could someone go through an install proceedure to help me make sure i got this right? then maybe i could make an installation wrapper for this. I've missed something. On one machine i have 2 cudas (on 2 8800 GTXs), and one cuda on the other machines (single card) but not CPU activity on either. V7 for dual GPU host, V8 for single GPU host, all relevant instructions in head post of this thread and http://lunatics.kwsn.net/gpu-crunching/modified-seti-mb-cuda-opt-ap-package-for-full-gpu-utilize.msg13941.html;topicseen#msg13941 |
Andrew Tapp Send message Joined: 26 Jan 02 Posts: 11 Credit: 27,364,018 RAC: 0 |
Hi, this is my first post here. I have been running S@H exclusively for the last few years and specifically this mod for a number of weeks. Despite appearing to run well i have not had any CUDA work units. Does anyone have any idea why this might be? Am running BOINC 6.4.5 on dual Zeon SuperMicro Board with Vista Ultimate SP1 64-bit and latest nVidia 181.22 drivers. Graphics card is a 8800 Ultra. cc_config.xml has been populated with 9 cpus and placed in the C:\ProgramData\BOINC folder. Your help would be appreciated. |
Westsail and *Pyxey* Send message Joined: 26 Jul 99 Posts: 338 Credit: 20,544,999 RAC: 0 |
Try for check you have selected to receive MB Cuda tasks in you web portal system preferences for Seti. Also, try to compare with stock application if it can work. Good luck and welcome! "The most exciting phrase to hear in science, the one that heralds new discoveries, is not Eureka! (I found it!) but rather, 'hmm... that's funny...'" -- Isaac Asimov |
©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.