Message boards : Graphics cards (GPUs) : Problem with 181.20 (on dual-GPU card)
Author | Message |
---|---|
Whatever you enable or disable Sli in control panel with theses new drivers, Boinc see only 1 CUDA device with my 9800gx2. | |
ID: 5404 | Rating: 0 | rate: / Reply Quote | |
Backing up to 180.48 doesn't resolve the problem, even with secon monitor connected, i'm dry of solution, someone here maybe? | |
ID: 5406 | Rating: 0 | rate: / Reply Quote | |
As *I* understand it this is a limitation of the API at the moment ... | |
ID: 5410 | Rating: 0 | rate: / Reply Quote | |
Well, he's got a single card with 2 chips, the 9800GX2.. so I don't think he's going to remove any SLI cables ;) | |
ID: 5446 | Rating: 0 | rate: / Reply Quote | |
Tham you for your interest. | |
ID: 5483 | Rating: 0 | rate: / Reply Quote | |
Backing up to 180.48 doesn't resolve the problem, even with secon monitor connected, i'm dry of solution, someone here maybe? Same (new) card, 9800GX2, ran two WUs sucessful (216739, 216744), no problem on Vista/64 and nVidia 180.48, but uses each CPU (E4300@2,4GHz) to full load, bad. Perhaps the small performance of the PCIe 1.0-board causes this usage, I will change board at weekend, had no time yet. I disabled SLI in the driver settings, connected second monitor, activated the second monitor, the second CUDA-accoint in BOINC-manager 6.4.5 was recognized at once. 78-79ms/step for a 2435 credit-WU should be o.k., running at stock. If someone can recommend newer drivers especially for 9800GX2, please post. | |
ID: 5629 | Rating: 0 | rate: / Reply Quote | |
Backing up to 180.48 doesn't resolve the problem, even with secon monitor connected, i'm dry of solution, someone here maybe? The CPU usage is a problem in windows with the 6.55 application ... though you seem to have higher than many of the rest of us have seen (or are measuring it better?) ... You may see an improvement with a new board ... but it will not drop to the "proper" level of 2-3% until they un-bug the application. For my part I would not want you to buy a new MB and find that it did not change your situation ... if it is something you want to do for other reasons, then fire away ... if only for this, be advised that you may not see significant changes .... ____________ | |
ID: 5630 | Rating: 0 | rate: / Reply Quote | |
For my part I would not want you to buy a new MB and find that it did not change your situation ... if it is something you want to do for other reasons, then fire away ... if only for this, be advised that you may not see significant changes .... Thanks, but I have not to buy a new one, this computer with a P43-Board (PCIe 2.0) should transfer more data over the slot, I hope it will unburden the CPU resp. reduce total performance loss a bit. O.k. if the application is the bottleneck, the reorganisation to another board is just for future performance. The 9800GX2 should profit from 2.0, or is that hope a mistake in scientific projekts like GPUgrid (not games) ? | |
ID: 5631 | Rating: 0 | rate: / Reply Quote | |
For my part I would not want you to buy a new MB and find that it did not change your situation ... if it is something you want to do for other reasons, then fire away ... if only for this, be advised that you may not see significant changes .... You will see an improvement in performance. Sight unseen, limited knowledge, and consulting only my crystal ball, I would guess about 10% gain as the tops ... more likely it will be 1 to 2% total performance improvement. But *YOUR* issue is CPU usage and I don't think that you will see a significant change there ... BUT I COULD BE WRONG ... :) This stuff is new enough that we are all kind of groping ... I just did not want you to spend $400 on a top of teh line MB to find no change ... at times that is not an issue for me in the budget so I might do it just on the off chance ... but, most people are more financially constrained and I was a little worried about you chasing a ghost ... Please report back ... take some more looks at CPU usages for the tasks and post them ... then do the same after the upgrade so we can learn from your experience ... and can then advise others in the future ... Just like we are watching those that are getting the 295 cards to see if they are worth the expense ... for me this is of interest in that I will be likely replacing two old systems with another new one and I will be looking to put some GPUs in it ... heck, I have two open slots as we speak ... of course the PS I have can't support another big card though I could likely put a couple of 9800 GT or similar if I wanted to ... I did see some package deals with a pair of GPU and PS for a grand so that might be something I might do for the i7 later on ... along with the new computer ... won't that be fun ... :) ____________ | |
ID: 5632 | Rating: 0 | rate: / Reply Quote | |
I hope it will unburden the CPU resp. reduce total performance loss a bit. I'd expect much less than 1% difference between PCIe 1.0 and 2.0, as the GPU does all the crunching locally. And I don't expect any difference in CPU usage, but feel free to prove me wrong ;) @Paul: I don't think there was a bug in the 6.56 app which caused the errors on XP64. Look, they're not sending different commands to the driver based on which (windows-)platform the app is run. If most drivers are OK with the code but XP64 isn't, it's likely a bug in that driver (*), not in the new way the new app handles polling the GPU. (*) there's a separate driver for XP64 MrS (edit: spelling) ____________ Scanning for our furry friends since Jan 2002 | |
ID: 5649 | Rating: 0 | rate: / Reply Quote | |
@Paul: I don't think there was a but in the 6.56 app which cuased the errors on XP64. Look, they're not sending different commands to the driver based on which (windows-)platform the app is run. If most drivers are OK with the code but XP64 isn't, it's likely a bug in that driver (*), not in the new way the new app handles polling the GPU. You are probably correct ... However, I have seen cases where the same code running only failed in limited cases pointing to other causes ... when the problem was in the code itself ... USUALLY this is a circumstance where the program does not handle the differences in the way the drivers (or external code) CORRECTLY, but DIFFERENTLY reacts to the application. A classic example might be where the driver returns a handle that is sized for the OS, a correct response, but the application improperly expects a handle of a specific size ... and in this case the problem may only surface in cases where that handle is dereferenced, for example, by the application. So, the error is in the application, not the drivers, but the problem only surfaces on the one platform ... Anyway, it *IS* only speculation, but ... in the long run I suppose I should stop doing it ... ____________ | |
ID: 5652 | Rating: 0 | rate: / Reply Quote | |
You're right, it could also be a real bug in the app. Either way, might be nice if it's fixed ;) | |
ID: 5654 | Rating: 0 | rate: / Reply Quote | |
You're right, it could also be a real bug in the app. Either way, might be nice if it's fixed ;) Concur ____________ | |
ID: 5661 | Rating: 0 | rate: / Reply Quote | |
We're having a similar report here on a GTX 295. Vista and Win7 didn't work, on XP 32 it worked correctly. So I'm quite certain we can point to the NV drivers here. | |
ID: 5715 | Rating: 0 | rate: / Reply Quote | |
Message boards : Graphics cards (GPUs) : Problem with 181.20 (on dual-GPU card)