System hard freeze while GPUs processing with ATI7970 if using computer

[Matt]

Hi Milos,

Yeah I played around for a very long time with CPU/GPU cores and I guess each machine will be slightly different with Motherboard, CPU and GPU combos/cores/bandwidths etc. I have a dual quad core CPU machine and I found the sweetspot for me was 4 cpu threads/cores. Any more or less than that and total performance went down. As you have observed the GPU's need some interaction with the CPU during depth reco even if they are 'turned off'.  I have never had a problem with ram leakage/limitations as always had 256 GB as a minimum

Results from your statue benchmark on high below to illustrate my point on latest beta build:

Stock no overclocking

0 cores

Device 1 performance: 572.199 million samples/sec (GeForce GTX 590)
Device 2 performance: 560.82 million samples/sec (GeForce GTX 590)
Device 3 performance: 792.417 million samples/sec (Tahiti)
Total performance: 1925.44 million samples/sec

2 cores

Device 1 performance: 50.8449 million samples/sec (CPU)
Device 2 performance: 606.999 million samples/sec (GeForce GTX 590)
Device 3 performance: 601.872 million samples/sec (GeForce GTX 590)
Device 4 performance: 876.044 million samples/sec (Tahiti)
Total performance: 2135.76 million samples/sec)

4 cores

Device 1 performance: 83.7371 million samples/sec (CPU)
Device 2 performance: 583.827 million samples/sec (GeForce GTX 590)
Device 3 performance: 598.242 million samples/sec (GeForce GTX 590)
Device 4 performance: 909.663 million samples/sec (Tahiti)
Total performance: 2175.47 million samples/sec

6 cores

Device 1 performance: 122.919 million samples/sec (CPU)
Device 2 performance: 589.601 million samples/sec (GeForce GTX 590)
Device 3 performance: 556.646 million samples/sec (GeForce GTX 590)
Device 4 performance: 784.322 million samples/sec (Tahiti)
Total performance: 2053.49 million samples/se

Moderate GPU Overclock

4 core

Device 1 performance: 81.7631 million samples/sec (CPU)
Device 2 performance: 636.155 million samples/sec (GeForce GTX 590)
Device 3 performance: 612.912 million samples/sec (GeForce GTX 590)
Device 4 performance: 1016.49 million samples/sec (Tahiti)
Total performance: 2347.32 million samples/sec

Matt

[Wishgranter]

heheh. THANX. for my was best to disable all Cores ( 2,7 Ghz ) and my CPU use is jumping from 0 to approx 40 % when reconstructing....

[Matt]

Yep my CPU's (3.3 ghz) run at between 65-90% when CPU is 'disabled'

[kris3d]

set as follows:
because this card does not reach even the factory settings.
it just only AMD ....

[Wishgranter]

try use MSI afterburner for overclocking....

[Matt]

Hi kris,

I can get my 7970 to run at 1250/1850 set in catalyst/afterburner/evga precision but in photoscan the GPU still does not use its full resources (fluctuates between 30-80%). It is more a photoscan implementation issue as it runs at 97% constantly for very long periods when folding proteins for Stanford FAH.

[David Cockey]

I recall reading that PhotoScan varies GPU usage to maximize overall speed, and the amount of GPU usage depends on GPU speed, CPU speed and memory speed. But that's only a recollection.

[spacepainter]

In the manual they say; run one core less if you are using GPU's. If I understood well.
I guess you have tried all that.