Chenyang Oculink SFF-8612 to PCI-E 3.0/4.0 M.2 M-Key to SFF-8611 Host Adapter for eGPU/GPD Win Max2 External Graphics Card & SSD

intel arc a750にて動作しました。 ミニPCのm.2がpcie3.0だったこともありベンチマークスコアとしては直挿し4.0×16より15%〜20%ほど下落しています。 特に不具合もなく相性問題などeGPUではよく話題にあがるものも発生せず安定しています。 ただ私の構成では電源連動はしなかったです。 m.2スロットには付属のカードではなく、別のoculinkメス端子と止めネジがついたものを使用しケースに固定していました。

Product works with my gpd win 4 2024, for people struggling to get it to work here's some tips 1. There's a microscopic on and off switch on the board, make sure it's switched on ( i found it after scratching my head for ages ) 2. If you are trying to connect an nvidia card and get code 43 look up nando4 fix You are welcome

Alles gut verpackt angekommen, angeschlossen und sofort funktioniert, alles bestens

Really happy with this purchase, it comes with the oculink cable and wifi slot to oculink connector but I bought another one with extender through the nvme slot. Much better than the noisy overpriced Aoostar. The build is combined with RTX 5070 and Corsair RM750e modular PSU for a clean build. Zero noise build, the fan only goes up if temperature passes 50 degrees Celsius. The GPU is turn on and off according to the laptop power state. Plug in, Power up, Install latest Nvidia Driver, Gigabyte control center, code 43 script. I write low level CUDA for a living, so here’s some extra info: - nvidia-smi shows 2 GPUs (my original laptop GPU rtx 3050Ti and this one) - 2 GPUs can communicate through PCIe, I tested a NCCL code for vector addition on 2 GPUs and it worked. - CUDA Code can compile with sm_120, and run no problem. Pytorch recognizes 2 GPUs and set the 5070 as GPU 0. Nsight compute works with debugging and uses 5070 if the code is compiled with sm_120.

Klappt super mit einer RTX3060. Ich habe im Laptop eine GTX 1660ti. Hier gab es Probleme. Google hat dann geholfen mit [SCRIPT] nvidia-error43-fixer. Das hilft wenn die eGPU ständig deaktiviert wird. Bei mir muss ich außerdem beim starten das HDMI Kabel entfernen um auch den Laptop Bildschirm nutzen zu können.