Folding Forum

My experience with two RTX 2060 cards (vanilla at US$350, KO [TU 104] at US$300) on a motherboard is that even with Ubuntu Linux, a GPU of that class needs at least PCIe 2.0 at x4 -- preferably better -- or it can be throttled down to 75% utilization or so. Right now I have the full-width PCIe 3.0 slot at x8 and the full-width PCIe 2.0 at x4. (That's about all this motherboard can manage unless it has only one GPU, in which case PCIe 3.0 x16 is the default.) Giving each card its own core seems to help too, although I keep hearing it at least implied that a thread per GPU is enough.

Mining Ethereum with a mess of cards each at x1 seems to get decent utilization out of GPUs (I don't know from experience -- maybe it doesn't!), but F@H really seems to need more bandwidth.

If I were trying for a low-cost but powerful rig, I'd consider an under-US$200 used workstation and stick an RTX 2060 KO in it -- if I knew it would fit. (Heck, any cheap computer under 10 years old that the card would fit in and that had a suitable power supply might do. Might need an adapter to get power for the GPU.) Cheap used keyboard and mouse and display (though I suppose it can be set up to run without them). Still... US$500 isn't all THAT cheap, but with luck and an adequate supply of WUs that sort of card can produce from 900k to 1.2M points per day. Don't run CPU WUs.

Edit: a lot of used workstations won’t take recent GPUs, though I’ve seen a few YouTube videos showing some that do.

Since I already bought the hardware I will use it. Can anyone confirm that CPU threads are sufficient?
I hope that the PCIe bandwidth is such a big problem given that I already have the hardware.

random_number wrote:Can anyone confirm that CPU threads are sufficient?

Given that this Celeron G3930 is still your CPU it has only 2 cores with 2 threads.
Than it can handle only 2 GPUs, from how I understand it.

If you want a system that runs well on a dual core Celeron CPU, I would probably want to ask you what your budget is.
The two thread celeron is able to run about 2x RTX GPUs; provided you have 2x PCIE 3.0 x4 slots or higher.

But if your budget is low, I would start with a GTX 1660, 1660 Super, or 1660Ti.
If you have the money, and can upgrade to an RTX GPU, do it!
RTX 2060KO is a great GPU, and possibly the best bang for the buck!

If you have slightly more money, a 2080Super. Or a single 2080Ti, which is one of the best GPUs on the market right now.

If power is not an issue, heat might.
However, if you're running an open system, or have a lot of case fans, AMD might be the solution for you!
Their 5700XT is about as fast as an RTX2070, but costs much less.

the CPU processing needed to support a GPU is not uniform. Most of the time, a thread is sufficient. Periodically, the WU validates the WU's status (sort of like a checksum, but much more processing is involved) and then it would be happier with a Core for a few seconds.

random_number wrote:I have now installed all GPUs and the PPD is quite low for 5 GPUS (between 200k and 400k). I think, as some of you have suggested, the CPU has not enough cores to deal with all GPUs. I will buy a new CPU and I was wondering does Folding@home require physical cores or are threads sufficient?

You are going to need a quad core with hyperthreading to feed those gpu's properly, even then you may struggle as 4 of the slots are PCie 2.0 x1. An intel i7 67xx or 77xx is what you need. An i5 may work but you'll likely see some bottlenecking.