TPF is WAY long -- P7641

[Blasko9]

I installed the v7 (universal client) last night and was very pleased with new interface, but the TPF on my gtx550 ti went from 5-6min to 1 hour, 23min!
My GT520 stayed at 23-25min.

I benchmarked, checked loads and temps and reduced threads on the SMP core but nothing has helped. The load on the 550 is usually pegged at 99% but now fluctuates between 93-98%. I have read about TPF issues and clunker WUs but have never heard mention of anthing this bad. The gt520 is running 7641 and the gtx550 is on 7640.

Any ideas on what's wrong? Suggestions on how to fix?

[patonb]

Have you checked if it is runnimng 3d clocks?

Get something like gpu-z and check the monitoring tab, and ensure the cards running full clocks.

How hot is it? If its not 60C or more its in 2d.

[Grandpa_01]

Ummm sorry but you got the clunkers but the 550 should be doing better around 15 min I think. Somewhere around 80+% GPU usage depending on GPU and very low PPD compared to other GPU WU's. I would look at the clocks as patonb suggested then start looking at some have suggested that the PCIe lanes are congested with the throughput needed with these, no one has really figured out a fix for them yet.

[Blasko9]

The temp is 71C which is normal, the clocks are maxed according to EVGA precision (is that what you mean by 3d clocks?). I did reboot also. Besides the 1 hour TPF the other thing that is unusual is that load meter is not pegged at 99%.

If the PCIe bus is congested that may work double against me. I have a PCIe 1.0 slot, half the bandwidth of a 2.0 slot right?

The GT 520 is in a regular PCI slot and has a TPF of 18 minutes which is normal. The 550 averages about 3x the work of the 520 so this is real strange.

[bruce]

Some Projects do a lot of computing on a small amount of data. PCIe congestion or speed won't matter on WUs with very little data since the GPU's compute engine is the limiting factor.

Some Projects need to do computations on a large amount of data. PCIe speed or congestion could be the limiting factor rather than the GPU if there is a large amount of data.

People have observed that PCIe bus speed doesn't matter, but those observations were make when GPU projects had 500 atoms. That's probably not true when you're talking about projects with over 20000 atoms. You may be the first to have reported the results of (1) a powerful GPU and (2) a slow PCIe bus and (3) a WU with lots of atoms. If you have a choice, put the fastest GPU in the machine with the fastest PCIe bus.

[Blasko9]

"If you have a choice, put the fastest GPU in the machine with the fastest PCIe bus."

Yep, that's what I have now. The GT 520 is in the PCI (as in preAGP) slot and the GTX550ti is in the PCIe slot.

I have been folding for a few months and have never seen the 520 have a TPF anywhere near the 550. Over 3x faster right now.

[k1wi]

Would be interesting to see whether you could drop the multiplier on the PCIe bus and see whether PPD drops.

[Blasko9]

Ok, I feel a little dumb but I still have a problem. The v7 interface is telling me that the 520 is in slot 00 and the 550ti is in 01. That is not correct. I paused the 520 and in EVGA precision the 550 load meter dropped to zero.

This makes sense. I still have a clunker WU, instead of having a TPF of 20-25min it is 1 hour but the 520 got stuck with it not the 550.

I will report the label issue in the appropritate section.

[Ripper36]

Solution for problem of two different card types:

viewtopic.php?f=67&t=19989&p=199379#p199379

Worked for me!