Folding Forum

My Client Martin_i7-875k_GTX560Ti has received it's first p7640 since it when to "Advanced"
http://foldingforum.org/viewtopic.php?f=24&t=21082

I think that the advanced status for this series of projects is a bit premature.
It is folding very slow because it fails to properly utilise the GPU, using 79%-80% of it.
A little more debugging is required in my opinion.

I am running v6.41r2 and folding p7640-r42-c0-g3 using core 2.22

As the work units get larger and more complex, the GPU may no longer be able to run them at 100% usage if the PCIe bandwidth can't keep up.

See also the Pande Group comment on this project in this already existing topic: viewtopic.php?f=74&t=21180

After my GTX560Ti successfully folded two 7641s, it was next served a 7640 (Run 31, Clone 0, Gen 4). In all of these, my GPU utilization was running around [EDIT] 88%, with correspondingly lower GPU temps [EDIT] 59 deg C. and system wall-plug power draw [EDIT] 324 Watts vs 360 Watts than for the thousands of other GPU WUs in the last year on this card. I thought I'd try to add a little more OC (from 948 to 953) to see what that would do to the TPF.

I woke up this morning to find that there had been four UNSTABLE MACHINE errors (see below) on this WU after 8, 12, 30 and 9 frames. I brought the OC down to 945, and it's stable again (I think) - having completed 55 frames now. The TPF difference is imperceptible between the 953 and 945 clocks, but the stability seems to be either an artifact of the project, or a wall for the GPU at the stock voltage.

mdrun_gpu returned 52
[10:03:06] NANs detected on GPU
[10:03:06]
[10:03:06] Folding@home Core Shutdown: UNSTABLE_MACHINE

[quote="7im"]As the work units get larger and more complex, the GPU may no longer be able to run them at 100% usage if the PCIe bandwidth can't keep up.

See also the Pande Group comment on this project in this already existing topic: viewtopic.php?f=74&t=21180


It was suggested that the above statement was not completely accurate. I apologize if the statement mislead someone to assume I meant TOTAL bandwidth was the gating factor. Eagle, in the other thread, suggests the limitation is the latency of the PCIe bus. And since latency is an integral factor of PCIe bandwidth, I did not feel a need to make that technical distinction.

It also does not change the result. FAH may or may not be able to code around it.

Where did the linked thread go?

edit: Is this the one?

http://foldingforum.org/viewtopic.php?f=74&t=21177

Someone merged the two threads.

sswilson wrote:Where did the linked thread go?

edit: Is this the one?

http://foldingforum.org/viewtopic.php?f=74&t=21177

Either that one or this one: viewtopic.php?f=74&t=21180

There are now so many threads discussing the same issue it is very hard to follow - mjay I suggest we all take it to bruce's thread on statistics ... viewtopic.php?f=74&t=21223&start=0