Last edit, all my latest tasks are called anonymous platform on my results page now. Is that normal? I don't think they were last night.
Yes, that's normal. By using an app_info.xml file, you tell the server that you are no longer using the project-supplied applications for the known platforms but an independent (anonymous) one.
That's also the drawback with going AP: you won't automatically get the newest applications but have to find and download them before incorporating them in your app_info.xml by hand.
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
Here's a complete app_info for windows with all the latest apps in it, just change the count value to 0.5, 0.33, or 0.25 depending if you want to run 2, 3 or 4 Cuda Wu's at once:
Claggy
Hi,
why not create a project preference for that. Would make it a lot easier and less error prone.
Just a little update, running 4 WU's on the GPU now. RAM usage is fluctuating between 1200-1300 MB. But the core downclock bug kicked in with all the restarting of tasks so I'll restart now to see how stable it runs overnight.
Restarted now. Running 6 CPU tasks along with the 4 GPU WUs. Overall CPU usage is at ~93% so the GPU isnt being starved of CPU cycles. GPU usage fluctuates around 73% (it was almost 100% when the core was downclocked), GPU memory use is between 1200-1300 MB.
if you can only reach about 75% gpu-load, the tasks are starving. don't care about overall CPU-usage - try to free another cpu-core and see if it improves..
Just a little update, running 4 WU's on the GPU now. RAM usage is fluctuating between 1200-1300 MB. But the core downclock bug kicked in with all the restarting of tasks so I'll restart now to see how stable it runs overnight.
Restarted now. Running 6 CPU tasks along with the 4 GPU WUs. Overall CPU usage is at ~93% so the GPU isnt being starved of CPU cycles. GPU usage fluctuates around 73% (it was almost 100% when the core was downclocked), GPU memory use is between 1200-1300 MB.
if you can only reach about 75% gpu-load, the tasks are starving. don't care about overall CPU-usage - try to free another cpu-core and see if it improves..
True, but even without running any cpu tasks the load only goes to 80%. I think it's because of win7. Got the same thing too on GPUGRID, it just maxes out at a set percentage (differs a bit per task there).
I think the best performance I get is when I leave 4 threads free for the GPU. Any more and it doesn't improve gpu usage. Even then each app only uses 5% of the CPU. Could that be changed by changing
True, but even without running any cpu tasks the load only goes to 80%. I think it's because of win7. Got the same thing too on GPUGRID, it just maxes out at a set percentage (differs a bit per task there).
you got those nvidia-physics drivers installed? maybe those are slowing them down..
Quote:
I think the best performance I get is when I leave 4 threads free for the GPU. Any more and it doesn't improve gpu usage. Even then each app only uses 5% of the CPU. Could that be changed by changing
0.200000
to something like?
0.400000
nope - this is only informative for boinc and does not change anything in the apps behaviour.
Maybe, but I also use this system for gaming and I'd rather keep physx for that. I haven't installed the 3D vision drivers though.
well at the very least you should try uninstalling the PhysX driver temporarily to see if in fact that is the reason your DC projects can't max out your GPU usage. at least then we'll know if PhysX had anything to do with it, and you can go right back to using PhysX. and on the off chance it turns out PhysX isn't the culprit, it would still be worth your while to explore why your GPU isn't reaching full usage under that kind of load...
...besides, inquiring minds would like to know :-)
With 3 the load is lower. And doesn't PCIE x16 give plenty of bandwidth? It's not like I'm running 2 GPU's and the PCIE drops to 8x. I've been told on GPUGRID that win7 is about 11% slower than winXP and linux with their CUDA tasks because of the way win7 handles it. I think it's likely that's the cause here.
1 WU is about 35%
2 WU is about 60%
3 WU is about 70%
4 WU is about 80%
It depends a bit on what else I'm doing with the PC or the kind of CPU tasks that Im running. But if I leave it alone 4 WU's running together will complete in about 90-100 minutes. I'm more than happy with this and it runs stable, what it doesn't seem to like is suspending all 4 cuda tasks at once though, that can sometimes cause a display driver crash and recovery (I'm thinking this may be because my GPU is overclocked by about 10%). Read somewhere it could be caused by a lack of power but my PSU is 850 watt which should be plenty for my system with just 1 GPU, and besides the GPU regularly gets stressed more while gaming (especially the witcher 2 but damn it looks awesome!).
I'm hoping to get a new CPU cooler to replace the stock cooler in the next few weeks. I'll be able to overclock the CPU a bit then and maybe that will help the CUDA tasks out too.
edit
I also want to thank everyone for their feedback and advice, much appreciated!
RE: Last edit, all my
)
Yes, that's normal. By using an app_info.xml file, you tell the server that you are no longer using the project-supplied applications for the known platforms but an independent (anonymous) one.
That's also the drawback with going AP: you won't automatically get the newest applications but have to find and download them before incorporating them in your app_info.xml by hand.
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
RE: Here's a complete
)
Hi,
why not create a project preference for that. Would make it a lot easier and less error prone.
Michael
Team Linux Users Everywhere
RE: Just a little update,
)
if you can only reach about 75% gpu-load, the tasks are starving. don't care about overall CPU-usage - try to free another cpu-core and see if it improves..
RE: RE: Just a little
)
True, but even without running any cpu tasks the load only goes to 80%. I think it's because of win7. Got the same thing too on GPUGRID, it just maxes out at a set percentage (differs a bit per task there).
I think the best performance I get is when I leave 4 threads free for the GPU. Any more and it doesn't improve gpu usage. Even then each app only uses 5% of the CPU. Could that be changed by changing
0.200000to something like?
0.400000RE: True, but even without
)
you got those nvidia-physics drivers installed? maybe those are slowing them down..
nope - this is only informative for boinc and does not change anything in the apps behaviour.
Maybe, but I also use this
)
Maybe, but I also use this system for gaming and I'd rather keep physx for that. I haven't installed the 3D vision drivers though.
RE: Maybe, but I also use
)
well at the very least you should try uninstalling the PhysX driver temporarily to see if in fact that is the reason your DC projects can't max out your GPU usage. at least then we'll know if PhysX had anything to do with it, and you can go right back to using PhysX. and on the off chance it turns out PhysX isn't the culprit, it would still be worth your while to explore why your GPU isn't reaching full usage under that kind of load...
...besides, inquiring minds would like to know :-)
Well, it does reach full load
)
Well, it does reach full load in projects like milkyway and primegrid so I don't know.
RE: Well, it does reach
)
maybe it's just a bandwidth limit on the bus. you may try to scale down to 3 tasks and see if it improves..
With 3 the load is lower. And
)
With 3 the load is lower. And doesn't PCIE x16 give plenty of bandwidth? It's not like I'm running 2 GPU's and the PCIE drops to 8x. I've been told on GPUGRID that win7 is about 11% slower than winXP and linux with their CUDA tasks because of the way win7 handles it. I think it's likely that's the cause here.
1 WU is about 35%
2 WU is about 60%
3 WU is about 70%
4 WU is about 80%
It depends a bit on what else I'm doing with the PC or the kind of CPU tasks that Im running. But if I leave it alone 4 WU's running together will complete in about 90-100 minutes. I'm more than happy with this and it runs stable, what it doesn't seem to like is suspending all 4 cuda tasks at once though, that can sometimes cause a display driver crash and recovery (I'm thinking this may be because my GPU is overclocked by about 10%). Read somewhere it could be caused by a lack of power but my PSU is 850 watt which should be plenty for my system with just 1 GPU, and besides the GPU regularly gets stressed more while gaming (especially the witcher 2 but damn it looks awesome!).
I'm hoping to get a new CPU cooler to replace the stock cooler in the next few weeks. I'll be able to overclock the CPU a bit then and maybe that will help the CUDA tasks out too.
edit
I also want to thank everyone for their feedback and advice, much appreciated!