Folding Forum

I have had a 2080Ti folding for a while now on a Windows 10 client, powered by a Ryzen 3 2200G. Owing to the integrated GPU this chip only provides 8 PCIe 3.0 lanes to the main PCIe slot.

Around a week ago I replaced the 2200G with a Ryzen 5 3600, which ups the slot to the full 16 lanes. I have reliably observed a roughly 350-400k ppd increase in performance. This isn't just due to some freak WU assignments either - I have a second, higher-clocked Ti folding on 8 lanes and can see the difference when they are both crunching the same projects. Where the faster card was seeing 150-200k ppd more before, it is now behind by about the same amount.

Just thought this might be of interest to anyone considering one, or already folding on one with restricted PCIe lanes.

Your observation is certainly valid for WUs from some, but not all, projects on Windows. Probably the best way to evaluate a project is to look at the size of the upload package. For example, right now my machine is uploading a result that is almost 80 MB and the one that finished just before it was a similar size. The one before that was only 11 MB. The atom count is closely related ... 16000 atoms vs. 2000 ... but it's not necessarily proportional.

Not surprisingly, smaller projects don't saturate the PCIe lanes as much as larger projects.

17:07:50:WU01:FS02:0x21:Folding@home Core Shutdown: FINISHED_UNIT
17:07:52:WU01:FS02:FahCore returned: FINISHED_UNIT (100 = 0x64)
17:07:54:WU01:FS02:Sending unit results: id:01 ....
17:07:55:WU01:FS02:Uploading 79.28MiB to 128.252.203.10

This is expected having the rtx 2080 ti limited by pcie 3.0 x8 by 10-20% on Windows. Linux is less demanding on pcie bandwidth for FAH and also faster in general.

A 2080Ti runs fine on a PCIE 3.0 4x speed slot, in Linux (meaning less than a few percent slower than in a 16x slot).

Like Bruce said, some projects are faster than others. I regularly see my 2080Tis bounce between 2.1 and 3.1 M PPD, depending on the project.
That's nearly 33% more points from slowest to fastest.

They certainly like this beta p14285 WU: