Am i processor limited?

[drnickriviera]

Repurposing a mining rig to fold. Has 4 1660 supers on risers w/ pci-e 3.0 x1. Pentium gold 5400T processor 2C/4T 3.1ghz. Sometimes I don't think i'm getting full ppd out of the cards. Proc usage stays pegged at 100%. If I use 3 cards it goes down to 80% and seem to get better ppd, but need more stats. Think i need a cpu with more cores? Would an older i5 with 4C and no threads be any better?

 

[crashtech]

I think you are on the right track, Folding on an Nvidia rig takes one whole thread per GPU to run. More ideally you would have 4 cores for the GPUs plus at least one to handle system overhead. Here's a Coffeelake hex-core for $95: https://www.ebay.com/itm/304705906882
But if you are on a budget then a quad-core would be a step up, and they're cheaper.

 

[StefanR5R]

Folding@home is also one of the few Distributed Computing projects which cause a lot of traffic on the PCIe bus from/to the graphics card. (As far as I know, it is the one project with the most traffic on the bus.) Therefore, the single lane connections which you have will throttle the cards to some degree.

One way to reduce the PCIe traffic somewhat would be to fold on Linux instead of Windows. (That's not because of a difference between the Windows and Linux FahCore implementations, but because of differences of the GPU driver stacks. On Windows, there is more frequent memory copying going on, which is the price to pay for features such as GPU driver crash dumps and GPU driver live updates.) But even on Linux, a GTX 1660S may be limited somewhat by the single lane link, I suspect.

Bigger GPUs are more affected by this though.

 

[StefanR5R]

PS, regarding CPU bottlenecks in F@h:
I saw it on dual-GTX 1080Ti computers with i7-7700K and on a triple-GTX 1080Ti computer with an E5-2696 v4 that F@h performance suffered as soon as I loaded more than 4−2 = 2 logical CPUs on the i7-7700K (4c/8t CPU), or more than 22−3 = 19 logical CPUs on the E5-2696 v4 (22c/44t CPU), with CPU heavy work from other applications. That is,

• each FahCore_22 worker wants a core for itself, if possible;
• HyperThreading doesn't help in this case.

 

[VirtualLarry]

I was folding on a mining rig last Dec., 5x gtx 1660 ti. I originally had a celeron G3900 2C/2T in there, couldn't get more than 1.8Mil PPD. With an i5 4C/4T, I could get 4Mil PPD, with EITHER 4x or 5x cards. Meaning, I think, that I was still CPU limited on that last card.

Mobo was ASRock H110 BTC PRo+, using 1x riser cards.

 

[lane42]

X1 slots are killers on folding. I had cards that should have done 1m ppd on x16 slots doing 700k on x1 slots.

 

[drnickriviera]

Thanks guys. I guess I should have put a little more info in too. I do have the cards frequency locked to reduce power draw. This rig is 100% solar powered, so anything I can do to reduce wattage is good. I need to do more testing, but I don't think the 3.0 x1 slot is hurting me much. I did notice it in 2.0 x1. Wish I could run the same WU over and over to test them on x16 and x1. I'll probably just run 3 or 2 cards for a while until I find a 65w 6core

 

[cellarnoise]

I ask this question of myself every single day. The consistent answer back from myself has been YES, so far.

D.C. is addicting

 

[StefanR5R]

This rig is 100% solar powered, so anything I can do to reduce wattage is good. [...] I'll probably just run 3 or 2 cards for a while until I find a 65w 6core

If you find a 95 W chip for cheap, you could take that one as well and power-limit it. Though I don't know what the most suited BIOS options would be. I was a bit lazy with my own Kaby Lake i7s and simply have switched off Turbo Boost most of the time.