1. Lift is algorithm specific (some coins have a 512bit state and some have a 256bit state). Each Acorn can provide a MAX of around 30MH of lift to algorithms with a 512bit state, and 60MH for algorithms with a 256 bit state. That is irrespective of the number of GPUs. Power consumption of Acorn at peak is 15W. The limit is PCIe bandwidth, which for this generation of Acorn is PCIe 2.0 x4 = 20Gbps nominal, which yields an effective rate of 16Gbps after accounting for overhead. That math will then get you to the 33MH @ 512bit state, less 10% for safety and underpromising.
2. The number of GPUs to Acorn is extremely GPU and algorithm dependent. X16R is good with 1:1, or even 1:2 on high end cards if your GPU is in x8. ETH is typically more GPUs per Acorn.
3. The important thing is your PCIe bandwidth has to add up. You cant use an Acorn in an M.2 slot @ 20Gbps to effectively accelerate one GPU in a PCIe 1.0 riser at x1 (2.5Gbps).
Edit: I also want to clarify that while Tom (@senseless) and I have collaborated on the 1525 and are generally doing some work together regarding FPGAs, he and AllMine have no involvement in the Acorn project, which is entirely a SQRL (Squirrels Research Labs) project. Ive noticed some confusion around that.
Can you clarify the algo(s) you are talking about when you mention "MH" in general? I know you specify bit state but personally I have no clue which algo uses which bit state...
Also, some real-life practical examples for people with existing rigs would be very useful - most of us got multi-GPU rigs chugging along - typically with x1 risers - and it'd be good to know what we could expect if we plugged an Acorn into the M.2 slot.
Finally, a maybe a "how-to" guide to extract the max out of an Acorn for an existing rig, considering all the parameters you mentioned, eg. "should we change PCI gen settings in BIOS", or "should we replace one of the 1x-16x riser with a 16x-16x riser on one of the GPUs to get the lift" etc.
Very good points. I would also like to add that there seems to have been more emphasis on, shall I say, smaller coins (and newer algos) but I would welcome some clear numbers for, lets say, ethash and other algorithms like it. This is an issue for those who are deciding if GPU mining is still viable or we should all just give up and buy ASIC miners. For the record - I am very happy with Z9 mini so far.