Questions and Answers :
GPU applications :
GTX 260 to GTX 295 and only got 2,000 more RAC?
Message board moderation
Author | Message |
---|---|
GTP Send message Joined: 5 Jul 99 Posts: 67 Credit: 137,504,906 RAC: 0 |
I have been running this new card for 2 weeks now and only see 2k more RAC! What a waste of money! Can this be correct? I went from 7,000 RAC to 9,000 RAC! Bummed, Aaron Lephart www.smartcar451.com |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13736 Credit: 208,696,464 RAC: 304 |
When upgrading hardware, or installing an optimised application, it can take a month to 6 weeks to show up in RAC values. And that's when the system isn't having issues coping with all the work being uploaded. If the number of Work Units processed per hour increased by x%, then eventually that's roughly how much your RAC will increase (with it's usual ups & downs). Grant Darwin NT |
Fred W Send message Joined: 13 Jun 99 Posts: 2524 Credit: 11,954,210 RAC: 0 |
I have been running this new card for 2 weeks now and only see 2k more RAC! What a waste of money! Can this be correct? When things are stable my GTX295 is good for about 8000 RAC on its own to add to the 8000 RAC of my overclocked quaddie. Even when things are stable, it takes more than 2 weeks for RAC changes to settle down (check how your Pendings have jumped when we can see them again). F. |
GTP Send message Joined: 5 Jul 99 Posts: 67 Credit: 137,504,906 RAC: 0 |
So was this estimate of 14-14k RAC wrong? http://setiathome.berkeley.edu/forum_thread.php?id=54386 All the best, Aaron Lephart www.smartcar451.com |
Fred W Send message Joined: 13 Jun 99 Posts: 2524 Credit: 11,954,210 RAC: 0 |
So was this estimate of 14-14k RAC wrong? Well mine has never got anywhere near that. It is running at stock clock and gives the figures I posted. F. |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
When things are stable my GTX295 is good for about 8000 RAC ... Really? You let run the stock_CUDA_app or Raistmer's_CUDA_VLARkill_app? One of my 4 OCed GTX260-216 make a ~ 8,000 RAC with Raistmer's latest app. |
zpm Send message Joined: 25 Apr 08 Posts: 284 Credit: 1,659,024 RAC: 0 |
pending credit page is still disable which is giving people false impressions about how much credit they are getting for work that is pending which they can't see. give it time and Patience!!!! |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
I would recommend following manual for GPU CUDA user. I would let run per GPU* one CPU-Core idle. Because of the BOINC client [boinc.exe] which have sometimes big CPU peaks. (depend of your WU cache size) Because of this all other under 'normal' are involved/disturbed - CPU and GPU tasks. It could be that the CPU support of the GPUs would be stop and then the GPU calculation make a break. That's why I don't crunch on the CPU of my GPU cruncher. 1 CPU-Core for boinc.exe and 1 CPU-Core for 'System' (what ever this is in TaskManager). And if two CUDA tasks stop similar, the new CUDA tasks can be prepared simultaneously on the CPU without disturbing the other two GPUs. If 4 CUDA WUs finish simultaneously.. two GPUs idle to the time the other two CUDA WUs are prepared on the CPU. The CPU preparation on my GPU cruncher is ~ 12 sec./CUDA WU. (AMD Phenom II X4 940 BE @ 4 x 3.0** GHz with 4 x OCed GTX260-216) [* with the nVIDIA GeForce GTX2xx series] [** for now stock speed, without OC] |
Fred W Send message Joined: 13 Jun 99 Posts: 2524 Credit: 11,954,210 RAC: 0 |
When things are stable my GTX295 is good for about 8000 RAC ... I don't hold with VLARkill so I rebrand them running the stock CUDA app. Can't overclock the GTX295 because of heat (at the moment it is 20% UNDERclocked but the figures were from when it was running at stock speed). Initial load of a CUDA WU to the GPU takes about 25 sec of CPU time but the rest of the time that CPU core is crunching a 603 (I don't see any point in leaving it idle for the rest of the ~10 minutes on a 0.4x AR CUDA WU). F. |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
I don't hold with VLARkill so I rebrand them running the stock CUDA app. Can't overclock the GTX295 because of heat (at the moment it is 20% UNDERclocked but the figures were from when it was running at stock speed). Initial load of a CUDA WU to the GPU takes about 25 sec of CPU time but the rest of the time that CPU core is crunching a 603 (I don't see any point in leaving it idle for the rest of the ~10 minutes on a 0.4x AR CUDA WU). Yes, O.K., it depend which kind of CPU and GPU. [ EDIT: And how much of it.. ;-) ] One of my 4 OCed GTX260-216 make an AR=0.44x WU in 6:45 [m:s] and shorties in 2:30. And I have big (25 % CPU, 100 % Core) and very long (~ 5 sec. every ~ 5 sec.) boinc.exe peaks because of small cache of ~ 4 days.. ~ 3,500 WUs.. ;-) I had let run 4 CPU WUs and 4 GPU WUs.. and the performance of the GPUs were very very bad and the RAC would be much much less than like now. So for my GPU cruncher (system) I need to let idle the CPU. |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
BTW. If you don't want to let run CUDA_VLARkill_app. Since short time the new CUDA_V12_app is available with and without VLARkill. So your CPU preparation time would be ~ 50 % faster/shorter. Also the GPU calculation would get a little booster. In the Installer is the VLARkill_app. But the nonVLARkill_app is available alone. http://lunatics.kwsn.net/index.php?module=Downloads;catd=9 But you need also the CUDA_V2.2 (3 x .dll's) from the Installer. So you could install in a temporary folder and take what you need. I done it like this.. ;-) |
Fred W Send message Joined: 13 Jun 99 Posts: 2524 Credit: 11,954,210 RAC: 0 |
Thanks, Sutaru. I am already running the CUDA v2.2 dll's so I will likely grab the updated nonVLARkill and see what that does. F. |
Sterling_Aug Send message Joined: 27 Sep 02 Posts: 54 Credit: 14,105,725 RAC: 0 |
http://setiathome.berkeley.edu/show_host_detail.php?hostid=4578997 My dual quad core Xeon workstation with a GTX295 CUDA card is getting almost 14,500 RAC. Before I had a CUDA enabled card, I was getting around 5,000 RAC. |
Sterling_Aug Send message Joined: 27 Sep 02 Posts: 54 Credit: 14,105,725 RAC: 0 |
That same PC is now up to almost 15,400 RAC and still not showing any signs of leveling off any time soon. |
Sterling_Aug Send message Joined: 27 Sep 02 Posts: 54 Credit: 14,105,725 RAC: 0 |
That same PC is now up to almost 15,400 RAC and still not showing any signs of leveling off any time soon. It is now almost 16,750 and moving up the list of top computers. #11 and about to move into the top 10. I was only getting about 5,000 RAC prior to adding the GTX295 CUDA card. |
Sterling_Aug Send message Joined: 27 Sep 02 Posts: 54 Credit: 14,105,725 RAC: 0 |
20,800 RAC and still climbing. |
Fred J. Verster Send message Joined: 21 Apr 04 Posts: 3252 Credit: 31,903,643 RAC: 0 |
Hi, CUDA enabled cards have 'extreme' computation way's (SSSE3) and very little 'other' things to do, except putting a few pixels on your screen . They have enormous calculatin capabilies. But they are 'simple', like OpenGL and DIRECT X9/10. In whitch, f.i. memory-faults, only show up, like a 'pixel' is NOT in place, not really something to worry about. But when it's used to calculate certain calculations, it's possible that a fault can make an WU INvalidate. Probably ONLY TESLA_CARDS/SYSTEMS are safe on the long run! I don't mean to scare off people, who are running CUDA -Devices. I think they can add a lot (of necessary) processing to the project. |
Westsail and *Pyxey* Send message Joined: 26 Jul 99 Posts: 338 Credit: 20,544,999 RAC: 0 |
Could you please elaborate what has led you to draw this conclusion? Also, what exactly is different about the Tesla cards vs. Geforce cards? P.S. On a side note, glanced at your hosts and wanted to ask if your X9650 was OC'ed? Thanks, Keep Crunching! "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 |
Sterling_Aug Send message Joined: 27 Sep 02 Posts: 54 Credit: 14,105,725 RAC: 0 |
Bouncing between 23,500 and 23,800 RAC now with the GTX295. |
Fred J. Verster Send message Joined: 21 Apr 04 Posts: 3252 Credit: 31,903,643 RAC: 0 |
Probably ONLY TESLA_CARDS/SYSTEMS are safe on the long run! Hi, @ Westsail & *Pyxey* , yes, I have the QX9650 running @ 3,5GHz, now. (Backed off during the heat, from 3,65GHz). My 'worries', since I've blown a few cards, cheap ones, are the temps on some cards and the MTBF (Mean Time Between Failures). Geforce cards are made for graphics and there is little known about them running at full speed, all the time and doing computations. TESLA cards are made for computing. {Don't know how 'much' is in a name :)} IMHO, I do think, only time (& OC ), can tell. BTW, sorry for my late reply. |
©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.