I have an issue very similar to the previously closed issue #166
I'm running a basic Wan 2.2 T2V workflow and if i set CLIPLoaderMultiGPU to cuda:0 everything runs and I get the desired video result. If i set CLIPLoaderMultiGPU to cuda:1 everything runs, I see the rise in cuda:1 vram usage except I get a blank black video. Unlike the previous issue, I don't have a memory overflow issue, I think my 16gb 5060ti is definitely enough to cover the 6.7gb text encoder.
I tried using the suggested CLIPLoaderDisTorch2MultiGPU and I get similar results. If I set donor device to cuda:1 and device to cuda:0 everything works, if I set device to cuda:1 I get blank black video. I have also tried the VAELoaderMultiGPU and VAELoaderDisTorch2MultiGPU and both works as expected.
My comfy-ui-multigpu version is 2.6.3 on a a new installation of ComfyUI v 0.18.2 (pytorch version: 2.12.0+cu130, Python 3.12.3),
I have an issue very similar to the previously closed issue #166
I'm running a basic Wan 2.2 T2V workflow and if i set CLIPLoaderMultiGPU to cuda:0 everything runs and I get the desired video result. If i set CLIPLoaderMultiGPU to cuda:1 everything runs, I see the rise in cuda:1 vram usage except I get a blank black video. Unlike the previous issue, I don't have a memory overflow issue, I think my 16gb 5060ti is definitely enough to cover the 6.7gb text encoder.
I tried using the suggested CLIPLoaderDisTorch2MultiGPU and I get similar results. If I set donor device to cuda:1 and device to cuda:0 everything works, if I set device to cuda:1 I get blank black video. I have also tried the VAELoaderMultiGPU and VAELoaderDisTorch2MultiGPU and both works as expected.
My comfy-ui-multigpu version is 2.6.3 on a a new installation of ComfyUI v 0.18.2 (pytorch version: 2.12.0+cu130, Python 3.12.3),