So this is an entirely hypothetical idea at this point.
So, sometimes I sit and ponder about "crazy" and/or "useless" things but I recently had one that I really like and did't know how exactly to try implement it.Bare with me.
- 3rd gen Ryzen is coming out in the (near) future and has been confirmed to support PCIe 4.0
- PCIe 4.0 ~ 2 * PCIe 3.0 ---> PCIe 4.0 x 8 ~ PCIe 3.0 Bandwidth
- PCIe 3.0 IIRC won't scale up to PCIe 4.0, it will work in the slot but not run a better than in a 3.0 slot
- Currently x3/470 chipsets support PCIe Bifurcation (x16 or x8/x8) out of the cpu, some b series board support it to (its a BIOS thing)
- A PCIe 3.0 x16 GPU in a PCIe 4.0 x8 slot will perform exactly as if it was in a 3.0 x8 slot, because there are still only 8 physical lanes connected and the GPU cannot utilize the higher per lane bandwidth.
What I would like to do is the same thing but with PCIe 4.0 to 3.0. Just think about it, if the IMO likely parts of the Zen 2 rumors are true, you could have an mITX board with 2 PCIe 3.0 x16 GPUs connected to it at ~full bandwidth and a rumored 16c/32t processor.
You could do a ridiculous water cooled work station build with the rumored R9 3850X (16c/32t 4.3/5.1 GHz), and two 2080 tis (or if Navi comes out swinging), in something like Jayztwocents' Loque ghost build (GPUs would be single slot because WC). I don't know how 2 240 rads would do though. Perf/L would be crazy.
Circuit design is not my strong suit, I'd love some input on the idea.