Message boards :
Number crunching :
Not enough time to complete by deadline.
Message board moderation
| Author | Message |
|---|---|
|
Send message Joined: 7 Jul 07 Posts: 8 Credit: 960,646 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]()
|
Hi, I have tried several WUs and the same happens on deadline. Here's last WU http://www.gpugrid.net/result.php?resultid=945503 Since this card is listed color netural, I've been testing it. It seems to be able to run the WUs, but not in time. Is there a problem making the deadline longer (e.g. 7 days instead of 5)? I haven't had any problems with the duration on 9600GT. Just the 8500GT. Thanks Bill |
rebel9Send message Joined: 9 Mar 09 Posts: 1 Credit: 65,021,535 RAC: 0 Level ![]() Scientific publications ![]() ![]()
|
Yes, I agree. I, and I expect most of us, run the GPU apps when I'm not using the machine, i.e. "Use GPU while computer is in use" is disabled. I also have a low-end card, for BOINC CUDA purposes. It has taken me months, on and off, to get to the point where a WU will complete without crashing this thing or aborting, through numerous crashes and BOINC and driver upgrades and finally having to install TThrottle. I finally finished one yesterday after some huge number of hours. I ran that one past the return date just to see if it would finish. Most WUs I get allow me around 3 or 4 days to complete. No-one who's running GPU WUs this way is likely to be able to do that. You really need to give us a lot more time to get these things through. I received no credit for my almost 200 hours of work. Equally, no-one wants to have to abort a WU that's consumed over a hundred hours because the deadline can't be met. I'm not a credit junkie so I'm more interested that the science is useful but it is pretty annoying and will deter a lot of potential crunchers. Cheers, Gary. |
|
Send message Joined: 18 Aug 08 Posts: 121 Credit: 59,836,411 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
If you have "weak" GPU i recomended Seti@Home or starting Collatz Conjecture. Gues from GPUGRID want quick back WU. POLISH NATIONAL TEAM - Join! Crunch! Win! |
|
Send message Joined: 7 Jul 07 Posts: 8 Credit: 960,646 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]()
|
Ht TomaszPawel, Thanks, been running all the GPU projects since they came out with WUs. This is the only projects having trouble making deadline. The reason I was wonder if the deadline could be changed was if I run for 5 days and it is resent back out then it could take up to 10 days to return the result for that WU. Seven is less than 10. I'm trying 1 more WU before I detach. Thanks Again Bill |
|
Send message Joined: 17 Aug 08 Posts: 2705 Credit: 1,311,122,549 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
The card you're trying to use has 16 shaders. Officially recommended are 50+. I wouldn't even dream about using such a card for GPU-Grid. Well, not if it's not a nightmare ;) MrS Scanning for our furry friends since Jan 2002 |
GDFSend message Joined: 14 Mar 07 Posts: 1958 Credit: 629,356 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() |
The reason why we are not using cards with 16 shaders is that a normal cpu is probably faster. So, it is not really possible for our heavy applications. gdf |
|
Send message Joined: 5 Jan 09 Posts: 670 Credit: 2,498,095,550 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
Hi, There is no way you are going to complete in time with that card. I have just completed a WU with a 8600GT twice as many shaders and cores, plus it's 3 times faster than a 8500GT and it's over clocked and it took over 55hrs of processing time and 3 days and 6 hours in computer time because it can't run when I am at the computer. In addition, this computer is on 24/7 Radio Caroline, the world's most famous offshore pirate radio station. Great music since April 1964. Support Radio Caroline Team - Radio Caroline |
|
Send message Joined: 20 Jul 09 Posts: 2 Credit: 0 RAC: 0 Level ![]() Scientific publications
|
hi ! Every WU is 5 days ??? with my : CUDA device: Quadro NVS 160M (driver version 18603, compute capability 1.1, 256MB, est. 4GFLOPS) i need a minimum of 10 days to finish the WUs sent... and only if i let it work 24/24 7/7... the post http://www.gpugrid.net/forum_thread.php?id=867#7753 should be posted directly in the home page of the project ! the server should not send work for computers that doesn't meet minimum requirement... --RonanKER |
|
Send message Joined: 17 Aug 08 Posts: 2705 Credit: 1,311,122,549 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
the post should be posted directly in the home page of the project ! It's right here, the 2nd point of joining. the server should not send work for computers that doesn't meet minimum requirement... GDF once said something like "too bad we can't avoid sending work to hosts which are too slow". Which was not received well - not because he can't, but because he'd even think about something like that. MrS Scanning for our furry friends since Jan 2002 |
|
Send message Joined: 7 Jul 07 Posts: 8 Credit: 960,646 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]()
|
Hi, I know. That's what I said. It took 5 1/2 days. WUs only have 5 day deadline. It finished but was too late by 10 hours. |
|
Send message Joined: 16 Apr 09 Posts: 163 Credit: 921,733,849 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
It looks like at least some of the new round of workunits are twice the size (timewise) as in the past. On a 9800GT running with an AMD940 CPU on Windows XP, that means instead of 16 hours, I'm seeing 32 hours. That is OK on a 9800GT (assuming the awarded credits change in proportion to the run time), and even works on a 9600GT. However, in the past, I could just barely complete a workunit in 5 days with a 9400GT. For those configurations, it looks like the change means no more GPUGrid for those workstations -- I know they are considered marginal in the GPUGrid afficianado circles, but it is disappointing. Those workstations will probably migrate over to Aqua. What is also disappointing is the ongoing non-support of ATI GPU's -- which provide more processing power per dollar (and support double precision calculations). It seems a tad inconsistent to push toward 9600 and up nvidia while excluding 38xx and 47xx and 48xx ATI cards from the fun. |
|
Send message Joined: 17 Aug 08 Posts: 2705 Credit: 1,311,122,549 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
What is also disappointing is the ongoing non-support of ATI GPU's -- which provide more processing power per dollar (and support double precision calculations). It seems a tad inconsistent to push toward 9600 and up nvidia while excluding 38xx and 47xx and 48xx ATI cards from the fun. Double precision is no bonus, as GPU-Grid gets by with single precision (and the ATIs are also much faster in single). BOINC support is still.. quasi-nonexistent. But what I think is the killer argument: they'd have to rewrite the client thoroughly.. maybe to the point where they'd have to maintain a 3rd different client besides PS3 and CUDA. Developing under ATIs stream SDK is different from CUDA and the underlying architectures are different: with nVidia you execute one instruction on 32 data in parallel, whereas on ATI you execute 5 instructions on one data at a time. That's why I think you can hardly write one client and have it run fast on both architectures. MrS Scanning for our furry friends since Jan 2002 |
©2025 Universitat Pompeu Fabra