Building a dedicated rig with 3 GPUs

[rickoic]

I suspect FAHclient has the gpu's mixed up and that's really the 960. And the difference between the 1070's is due to the x16 x8 slot they are in.
After moving folding to my M.2 card yesterday here are some new numbers. Plus it's been in use for a couple of years and I've noticed the rear fan having problems, so probably ready to be replaced.
CPU 12 P14631 52,798
2080 x16 slot P16600 1,871,921
1070 x16 slot p16600 759,026
1070 x8 slot p14467 187,074
960 x8 slot P14459 118,023

Est PPD 3,001,139

[HaloJones]

There shouldn't be that much difference just from x16 to x8. I've got a 1070 in an x4 Gen 3 slot and it's currently at >1m on a P16600. It is typically only around 10% slower than the TitanX in an x16 above it.

[rickoic]

Looks like that 1070 is going south on me and will probably be quitting on me soon.

[Neil-B]

… would it be worth switching the two 1070s around in your rig to check it is a difference in the cards rather than a difference in the rig slots? … though I guess you might well have already done that

[markdotgooley]

There shouldn't be that much difference just from x16 to x8. I've got a 1070 in an x4 Gen 3 slot and it's currently at >1m on a P16600. It is typically only around 10% slower than the TitanX in an x16 above it.

For a while I had a WiFi card on an x1-wide PCIe slot. For this motherboard, that lowers all but the principal PCIe 3.0 x16 slot to x1. Then I got a USB dongle for WiFi. I changed BIOS settings so that one RTX 2060 is on PCIe 3.0 x8 rather than x16 and the other is on PCIe 2.0 x4 (the motherboard can do no better).

Going from PCIe 3.0 x16 to PCIe 3.0 x8 makes almost no difference. Speeding up the other card from PCIe 2.0 x1 to PCIe 2.0 x4 helps a lot but it’s still a bit throttled at some times.

Perhaps a faster card would suffer from x8 rather than x16. I don’t know. Ethereum miners and such claim that x1 connections don’t slow the GPU cards down but in my experience they certainly do slow down folding...

[HaloJones]

I've seen no speed impact from gen3x4 upwards. I'e got another board which delivers 2 gen2x16 slots via a mux (Gigabyte G1 Sniper) and both cards perform the same.

x1 I can see as a problem but >x4 and it won't suffer.

[MeeLee]

There shouldn't be that much difference just from x16 to x8. I've got a 1070 in an x4 Gen 3 slot and it's currently at >1m on a P16600. It is typically only around 10% slower than the TitanX in an x16 above it.

If it's a Gen 2 slot x8 it could be under Windows.
Should be OK for Linux though.

The M.2 slots (if PCIE 2.0) should be even slower, and probably not recommended to fold on, as they're only x4.
If they're PCIE 3.0, you could still use them for most GTX GPUs in Linux.
The RTX 2060 is already seeing a serious drop in performance in Linux, let alone Windows.

[rickoic]

Moved my gpu's around and now have this:
5,167,541 PPD
CPU 12 P14630 46,531 PPD
2080 X16 Slot P13414 1,723,074 PPD
2070S X16 Slot P13414 1,660,455 PPD
1080 X8 Slot P13414 947,044 PPD
1070 X8 Slot P14457 783,701 PPD

Moved suspected bad 1070 and 960 to a computer by themselves and getting this:
1070 P11760 785,916 PPD big improvement
960 P16600 214,656 PPD

[HaloJones]

More computers adds redundancy