Should weak GPU folding?

[TwoFromBLR]

I have this PC:
MSI Z97 Gaming 9 AC;
i7-4790K;
4×8 Gb HyperX;
MSI RX 5700 XT Gaming X;
Straight power 11 1000W.
And i have Palit GTX 650 1Gb. He can folding, but his ETA for each WU is 1.5-2.5 days. This GPU can't finish WU before timeout, but it's enough to finish before expiration. PPD is around 20K.
RX 5700XT installed in first PCI-E slot and running at x16 3.0. He block second slot, so only third slot is available.
CPU PPD with 7 cores is around 30-35K.
So there is 2 ways:
7 CPUs, RX5700XT at x 16, or
6 CPUs, RX5700XT at x8, GTX 650 at x4.
What is better? Price of electric power does not matter.

[HugoNotte]

If I understand correctly, FAH wouldn't use 7 cores for folding anyway. Large primes, of which 7 is the lowest one, don't scale well. Hence FAH client would only use 6 cores anyway. As to your GTX650, if it doesn't manage to return a result before Timeout, the work server will queue another copy of your WU your 650 is busy with to be sent out again. If your 650 manages to return it's result before the 2nd WU's result comes back, your result is being used for science. If not, it might be used for verification purposes.

Some say, your late result is still of a certain value to the project. Others might think, that if your WU would have been assigned to a GPU which would be able to return the result within 8 or 10 hours (a relatively lowly GTX750Ti would be able to that) the project might be able to move on to issue the next generation of WUs depending on your result 24+ hours earlier. Hence your rather slow GPU might be holding the project back by that much time.

[Joe_H]

Usually the GT 630 or 640 cards are about the minimum to complete WUs within the first timeout value by processing 24 hrs a day, but that may depend on exactly which project is being. Also, are you looking at ETA figures right after starting or after 4-5% has been completed? It does take a bit before the estimates settle to a more accurate value.

[markdotgooley]

I have a GT1030 somewhere on a half-height card. Supposedly around 68,000 PPD. I have two RTX 2060 cards working now, 1,000,000 PPD each if they can get the right WUs. I don't think I'll bother with the 1030...

[foldy]

They currently have work units in beta which are very small and cannot fully use any fast GPU but only like 30%. So they plan to restrict these tiny work units to very slow GPUs.

That is the best choice:
6 CPUs, RX5700XT at x8, GTX 650 at x4.

[TwoFromBLR]

Are you looking at ETA figures right after starting or after 4-5% has been completed?

I look at log, not ETA. I know, that right after starting ETA is incorrect, like 24h 00m for every slot.

So they plan to restrict these tiny work units to very slow GPUs.

But before this is better leave 5700XT alone? Then she use 16 PCI lines. Many difference with x8?