Which config would do better, efficiency-wise: 5090 or a pair of 4080 Supers?

[CaptainHalon]

Usually a single larger gpu is more power efficient, but I'm not sure on this one, especially with the power slider variable. Assuming you had a motherboard that could do x8/x8, and you could tweak power with afterburner, which setup do you think would be superior in terms of balancing performance and efficiency?

[muziqaz]

2x 4080 is safer bet since you will have better utilisation wth majority of projects.

[CaptainHalon]

2x 4080 is safer bet since you will have better utilisation wth majority of projects.

Makes sense. Probably safer for the 12VHPR too.

[arisu]

The 5090 uses 575W and can compute about 105 TFLOPS. The 4080S uses 320W and can compute about 52 TFLOPS. So as far as raw compute power goes, two 4080S GPUs is almost exactly as fast as one 5090, but would use more power, at 640W total. That's an extra 65W.

If you are on Linux, you can use MPS (viewtopic.php?t=43130) to fully utilize a single 5090 by running multiple WUs on it at once. It will do far more work than running a single project on a 5090. You can be gentle on the power cables by reducing power from 575W to 500W.

Points in favor of 2x 4080S:

• You get more total L2 cache (64 MB in two cards instead of 96 MB in one card)
• Less stress on the 12VHPWR
• Lower species level means you'll be less affected by shortages (although this is an artificial problem)
• Better cooling for higher boost clocks (6 slots total of to cool 640W total vs 2 slots to cool 575W)

Points in favor of 1x 5090:

• Faster and more power-efficient VRAM (GDDR7 vs the 4080S's horribly power-hungry GDDR6X)
• If using MPS, better ability to scale usage, e.g. possibility to run 3 projects at once if it benefits, not just two
• Fewer PCIe lanes are used up on your motherboard
• Less expensive (2500 USD x1 vs 1400 USD x2 = 2800 USD)

The internal hardware of the 4080S and 5090 that gets used by FAH is basically the same. Same generation SMs, same CUDA cores, same scheduler. The only real difference is the 5090 has more CUDA cores packed into it and a faster, more efficient generation of VRAM.

You'll be more power-efficient if you get the 5090, and you'll save some money while you're at it. If you use MPS, you'll be able to fully-utilize the GPU. However, GPU WU shortages may affect you more often (this may change in the future). I'd personally lean slightly on the side of the 5090 if you don't mind a bit of tinkering with power settings and MPS to optimize things, otherwise I'd lean slightly on the side of 2x 4080S.

[CaptainHalon]

I decided to try out a 5090 on Windows. I set the power level as low as I could, which was 66%, so it runs around 400W. Seems to consistently get WU's, but sometimes the PPD will say 40-45M, where other times it's more like 20-25M, depending on the WU. I guess that's where Linux could come in handy running multiple WU's simultaneously, but I don't really trust being able to set power levels properly in Linux. Never had a good experience with that in the past as it would generally just ignore everything I tried.