Message boards :
Number crunching :
Lunatics Windows Installer v0.40 release notes
Message board moderation
Previous · 1 . . . 9 · 10 · 11 · 12 · 13 · 14 · Next
Author | Message |
---|---|
Jord Send message Joined: 9 Jun 99 Posts: 15184 Credit: 4,362,181 RAC: 3 |
ATM my average for them is around 5.75hrs Mine is with the 331_AVX app, on average 5.55 hours. The person running with the default app and 59,600 seconds, has a locked 2500. Not sure if that makes a difference, with Turbo Boost and all. This was my last AP on ATI for the moment. 2381564519 5967479 2 Apr 2012 | 20:15:57 UTC 5 Apr 2012 | 0:37:23 UTC Completed and validated 121,396.53 119,551.80 695.44 AstroPulse v6 v6.01 2381564520 6065655 2 Apr 2012 | 20:16:00 UTC 5 Apr 2012 | 0:34:22 UTC Completed and validated 60,787.31 295.47 695.44 AstroPulse v6 Anonymous platform (ATI GPU) OK, so maybe it was a long one (although, it was paired against an OCed i3-540), but still... I have the sneaking suspicion that CUDA does this better. (Nothing against the Builders of the Apps!) :-) |
Wiggo Send message Joined: 24 Jan 00 Posts: 34744 Credit: 261,360,520 RAC: 489 |
ATM my average for them is around 5.75hrs My average would likely be lower but most of mine have been interrupted by MB's going into high priority mode though I have had a few finish in that 5.5hr time (I've also had them finish a lot sooner too but I'm happy with the average that I have with them seeing as v505 AP's use to average 6.25hrs). I believe that nVIDIA cards also use OpenCL to process AP's, not CUDA, from what I've read so things could be an entirely different story there. Cheers. |
cliff Send message Joined: 16 Dec 07 Posts: 625 Credit: 3,590,440 RAC: 0 |
I'm wondering when the wisdom.dat file is created in the projects folder and what creates it? Due to a receent change of motherboard and a re-install of Win7 to be able to install SP1, I had the boinc data dir backed up. Now it includes a wisdom.dat file and I dont know if its updated at all or if once genearated it is static. If its static once generated can it be deleted in order to create another uptodate file, or can an update be forced? Regards, Cliff, Been there, Done that, Still no damm T shirt! |
Mike Send message Joined: 17 Feb 01 Posts: 34255 Credit: 79,922,639 RAC: 80 |
You can just delete it. The app will create a new one. With each crime and every kindness we birth our future. |
cliff Send message Joined: 16 Dec 07 Posts: 625 Credit: 3,590,440 RAC: 0 |
You can just delete it. Thanks Mike. Cheers, Cliff, Been there, Done that, Still no damm T shirt! |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
I have the sneaking suspicion that CUDA does this better. (Nothing against the Builders of the Apps!) :-) Can something be better if it just doesn't exits ;) There is no CUDA AstroPulse at all. There will be one eventually, but currently only OpenCL NV and OpenCL ATi vrsions. About your issue with ATI AP task - please, report blanking % also cause completion times strongly depend on this param. Also, you could met ATi GPU low usage problem we see time to time. I have opinion that it's driver synching issue cause there is correlation between CPU time used and elapsed time. Tasks with extremely long elapsed time often show considerably less (!) used CPU time than usual ones. But it's only correlation, not dependance. There are extremely long tasks w/o decreased CPU time as well. In general, this issue can be solved by suspending/restarting GPU processing. GPU-Z and similar programs will show something like ~68% or even less GPU usage even running 3 A/MB tasks simultaneously. After restart usage can return to ~100% again. |
Jord Send message Joined: 9 Jun 99 Posts: 15184 Credit: 4,362,181 RAC: 3 |
[Can something be better if it just doesn't exits ;) It does in a way, over at Einstein/Albert. There the Binary Radio Pulsar Arecibo) (BRP) searches run way better on CUDA than on OpenCL. About your issue with ATI AP task - please, report blanking % also cause completion times strongly depend on this param. The one running 78k seconds was: single pulses: 2 repetitive pulses: 1 percent blanked: 2.39 The one running 60k+ seconds was: single pulses: 4 repetitive pulses: 1 percent blanked: 4.01 Also, you could met ATi GPU low usage problem we see time to time. Yes, I saw that throughout the last task. At most it used 4% GPU (from GPU-Z), but mostly it was zero. In general, this issue can be solved by suspending/restarting GPU processing. As can be seen from its stderr.txt, I restarted the 78k+ multiple times. That didn't fix anything. But what may have fixed it actually eventually restarting and finishing, was me uninstalling the Intel HD 2000 graphics drivers that had sneakingly managed to install themselves on my system, I suspect with one of the chipset driver changes I made the past days. I noticed that as GPU-Z all of a sudden showed Intel HD graphics as something it would check. GPU-Z and similar programs will show something like ~68% or even less GPU usage even running 3 A/MB tasks simultaneously. After restart usage can return to ~100% again. Yeah well, I'll run one task at the same time on the GPU. Never tried if multiple ones were even possible on my ATI, although I have the memory for it (2GB). And besides, how many months will they then take to finish, huh? ;-) I'll run MBs on my ATI for now and APs only on the CPU. |
Cosmic_Ocean Send message Joined: 23 Dec 00 Posts: 3027 Credit: 13,516,867 RAC: 13 |
I finally got my first 10 <10% blanked non-30/30 tasks to validate. It took 44 tasks to get those 10. Since reaching 10, I have not received any new APs due to no fresh tapes. Along the way though, the ETA has adjusted itself from the original 209 or 219 hours down to about 94, so a 10-day cache now consists of the three that are running plus 4-5 "ready to start." I should start getting tons of APs once some new tapes get loaded and start splitting though. Linux laptop: record uptime: 1511d 20h 19m (ended due to the power brick giving-up) |
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
I finally got my first 10 <10% blanked non-30/30 tasks to validate. It took 44 tasks to get those 10. Since reaching 10, I have not received any new APs due to no fresh tapes. The new tasks should come in with ETA around 9.4 hours since the APR limiting is still at ten times whatever <flops> the host is sending. You might end up with about 25 tasks per core with a 10 day cache setting if the supply is good enough. But then the first one completed will boost DCF to compensate. The ten times limit is probably going to be raised to fifty times next Tuesday, at which point new tasks will have ETAs of roughly 6.5 hours, the ~45 GFLOPS indicated by APR will actually be used. The cache could possibly get up to more than 25 days of actual run time before one of those completes and boosts DCF again. Probably not, and for your host the limit of 50 tasks per CPU may be slightly less than 25 days runtime anyhow. Still, if it were my system I'd shorten the cache settings somewhat at least until everything finally stabilizes. Joe |
Cosmic_Ocean Send message Joined: 23 Dec 00 Posts: 3027 Credit: 13,516,867 RAC: 13 |
I finally got my first 10 <10% blanked non-30/30 tasks to validate. It took 44 tasks to get those 10. Since reaching 10, I have not received any new APs due to no fresh tapes. If nothing else, when we went to creditnew and the ETAs fell through the floor, the one thing that saved me from getting too much work on a 10-day cache was the 1gb limit that I set years ago, thinking I would never hit it but at least it was a generous safeguard. I ended up with ~22 days of APs on the old 2P Opteron machine. The single-core Sempron didn't fair as well though. It was doing MB at the time and ended up with something like 170 days worth of those, since shorties were estimated at something like 34 seconds instead of 2.5 hours. I think it would be pretty cool though to finally.. for the first time ever, get the "you have reached the limit" message. I have never gotten one of those. Linux laptop: record uptime: 1511d 20h 19m (ended due to the power brick giving-up) |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
[Can something be better if it just doesn't exits ;) It's completely different app.
So it's low GPU usage bug indeed. Interesting that it could be fixed with unistall of Intel's drivers. I see low GPU usage bug on my host as well time to time... and my motherboard has onboard Intel graphics too. Maybe it could be fixed in the same way? Will try. @Others who noticed this bug - what motherboards do you use? Do they have embedded/onboard GPu chipsets ?
As statistics for my HD6950 shows, RAC-wise performance of AP ATi outperforms MB ATi in few (!) times. Doing AP on beta already gave RAC >25k while max for that host on main was ~12k So you could try ;) Also, running few AP tasks on same GPU should give better performance boost than running few MB tasks on same GPU (because of blanking done on CPU). GPU with 2GB memory definitely can do few AP at once. It takes from 3 to 1,5h to complete 3 tasks on my HD6950. [EDIT: CPU/GPU times -wise AP performance is better than for MB too. Perhaps it's because of better CPU MB overall optimisation from one side and few hard places (i.e., low workitems number available) in algorithm on GPU MB from another side. AP algorithm (especially with big unrolls) almost always have enough workitems to keep GPU busy. |
kittyman Send message Joined: 9 Jul 00 Posts: 51468 Credit: 1,018,363,574 RAC: 1,004 |
The kitties are waiting to see the change actually happen. They are gonna run it out by the fur on their britches. No flops, no BS. Boinc is gonna have to handle it, on both the server and the host end. Most of my rigs have proven in the past that they can handle whatever foibles are tossed their way. The kitties trust in DA......LOL. Actually, ROFLMAO. "Freedom is just Chaos, with better lighting." Alan Dean Foster |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
Asus P8Z68-V PRO with i7-2600K and Intel HD Graphics 3000, the Intel drivers are installed, but there is no monitor connected to its port, and the Display adaptor doesn't show up in device manager, Edit: for later reference its version: 8.15.10.2321 that i have installed. Edit 2: Intel Graphics drivers now uninstalled. Claggy |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Hm... I have nothing attached to onboard GPU too. And only HD6900 driver listed in device manager. Maybe it was just coincidence? And not driver uninstall helped but just reboot in process of uninstallation? ... @Ageless Could you give AP tasks more attention again and look if after Intel driver removal there is no GPU low usage bug on your host indeed? |
Jord Send message Joined: 9 Jun 99 Posts: 15184 Credit: 4,362,181 RAC: 3 |
@Ageless Sure, I'll re-enable AP on ATI. For the record, I don't have any monitor attached to the VGA outlet on the motherboard either. Only to the ATI card. I wasn't using the Intel HD graphics, but apparently that doesn't matter for installing the drivers or not. I have now disabled it in Device Manager, as extra security measure. |
Cosmic_Ocean Send message Joined: 23 Dec 00 Posts: 3027 Credit: 13,516,867 RAC: 13 |
The new tasks should come in with ETA around 9.4 hours since the APR limiting is still at ten times whatever <flops> the host is sending. You might end up with about 25 tasks per core with a 10 day cache setting if the supply is good enough. But then the first one completed will boost DCF to compensate. Follow-up. Got a new task this morning. ETA: 06:02:15. The last task I had before I got that one was 77:38:41. Because of the way DCF and estimates work, it took about 20 to go from a little over 200 down to about 175, 15 more to drop down to about 125, and about five to get down to about 75. I imagine it would have only taken about 5 or so more for the ETA to be pretty close anyway. Linux laptop: record uptime: 1511d 20h 19m (ended due to the power brick giving-up) |
Cosmic_Ocean Send message Joined: 23 Dec 00 Posts: 3027 Credit: 13,516,867 RAC: 13 |
And some more tasks showed up with an ETA of 04:41:47. Linux laptop: record uptime: 1511d 20h 19m (ended due to the power brick giving-up) |
Wiggo Send message Joined: 24 Jan 00 Posts: 34744 Credit: 261,360,520 RAC: 489 |
I personally don't think that they'll ever get those estimates fixed. I've got 1 file batch telling me 8hrs then newer file batch telling me 77hrs then I just completed another from a different batch again in 4.5hrs. :( Cheers. |
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
Cosmic_Ocean wrote: And some more tasks showed up with an ETA of 04:41:47. Definitely in the territory where you may be able to get the "you have reached the limit" message. Joe |
Guvner Send message Joined: 20 May 99 Posts: 2 Credit: 380,494 RAC: 0 |
NVM - Posted question in a new thread... |
©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.