Vulkan version doesn't work on RTX 4070 ?

[olivbrau]

Hi everybody
I'm still trying sd on my new computer which has a RTX 4070 GPU (12 GB VRAM)
I have a problem with CUDA (cf. another issue) so I try Vulkan.
It worked well on my old computer (with a RTX A1000 4GB RAM)
But it doesn't work on the RTX 4070 : it blocks on the 2d "Compiling shaders"
Here is the log :

C:\Users\obrau\Desktop\sd-master-9578fdc-bin-win-avx2-x64>"cuda\vulkan\sd.exe" --diffusion-model ".\Flux.1 Q4 F16\Dev\flux1-dev-q4_0.gguf" --clip_l ".\Flux.1 Q4 F16\clip_l.safetensors" --vae ".\Flux.1 Q4 F16\ae.safetensors" --t5xxl ".\Flux.1 Q4 F16\t5xxl_fp16.safetensors" -p "a cute cat" --sampling-method euler --steps 10 -W 512 -H 512 -s 42 -t 16
ggml_vulkan: Found 2 Vulkan devices:
ggml_vulkan: 0 = NVIDIA GeForce RTX 4070 SUPER (NVIDIA) | uma: 0 | fp16: 1 | warp size: 32
ggml_vulkan: 1 = Microsoft Direct3D12 (NVIDIA GeForce RTX 4070 SUPER) (Dozen) | uma: 0 | fp16: 1 | warp size: 32
ggml_vulkan: Compiling shaders..............................Done!
ggml_vulkan: Compiling shaders.....

And it is blocked here ...
Does anyone has an idea ?
Thanks a lot in advance
Olivier

[stduhpf]

That's weird. Have you tried running llama.cpp Vulkan with that same card to see if it works?

[jeffbolznv]

ggml_vulkan: 1 = Microsoft Direct3D12 (NVIDIA GeForce RTX 4070 SUPER) (Dozen) | uma: 0 | fp16: 1 | warp size: 32

This is a Vulkan-on-D3D12 layer that gets installed on some systems. It's probably having some problem with the shaders. There's a post here with instructions about how to remove it: https://steamcommunity.com/app/2198150/discussions/0/4330853251136279833/?ctp=12#c4844274022643931439