I can not run an application on a graphics card with 256M of memory?
In describing the application client_state.xml 220200960.000000 . but in reality the application uses all the 256M (tested on gpu with 512 MB). Below the screen is attached to 8600gt256.
[/img]
Copyright © 2024 Einstein@Home. All rights reserved.
BRP3cuda+256M
)
On my GTX260 the CUDA-app uses up to 300MB of RAM. The average is around 270 MB.
Too bad, i wanted to test my
)
Too bad, i wanted to test my Nvidia 8400GS in my Laptop with einstein. It has 256mb in theory. I know its damn smlow, but i was curious to see if in this hybrid mode it might be worth a shot. Maybe that will be looked into in the later versions, so far the cuda app really does a lot on my 8800GT so maybe even the 8400GS might help out a bit. On all other projects it really is too slow to be worth it, but with the hybrid mode it could be interesting i guess.
RE: I can not run an
)
Does it run when the screen is off? How about when you are not using the pc?
RE: Too bad, i wanted to
)
What do you mean by 'hybrid mode'? If you mean adding system ram to make the gpu think it has more memory, no it will not work. The whole point of using a gpu is too get the whole app into its memory, if you are trying to use system memory you are not making it faster, but ALOT slower instead. The programmers set up the gpu apps to only see the gpu memory, not any additional system memory.
Also an 8400GS has 16 cuda cores and a core clock speed of 400mhz while the 8800GT has 112 cuda cores and a core clock speed of 600mhz, that is ALOT of cuda core difference!!
RE: What do you mean by
)
Well on the 8800GT it runs rougly around 6000 Seconds and uses around 2000 Seconds of CPU time. And uses 0.20 CPU. I think the cuda app is not run pure on the GPU but also using the CPU for some tasks. I have no clue how it exactly works, but it seems to be different from Milkyway and other GPU tasks.
I know. But from my results i was a bit baffled that the 8800GT often is as fast as a lot newer cards and only half as fast as a GTX 580. If i am not wong that card has 512 Cuda cores at 772mhz. Thats also a big difference to the 8800GT wich still at least in my results only is twice as fast.
RE: RE: What do you mean
)
Yes you are correct, it does use both the cpu and gpu to crunch, that is what the 0.20 cpu means. It is a percentage of the cpu being used as it crunches and is why people with several gpu's should consider leaving one cpu just for feeding their gpu's instead of using all of them for crunching. Some projects use more of the cpu than others.
Again you are correct, the 580 does have 512 cores and runs at 772mhz, and only being twice as fast as an 8800GT means that the gpu app is not as efficient, or optimized, as much as it could be. Over at Aqua they pulled the gpu app because the cpu app was faster!