SDXL INT8 Basic Workflow

For Nvidia use stock ComfyUI with --use-pytorch-cross-attention.

python main.py \
  --use-pytorch-cross-attention 

Minimum start up parameters for ROCm\FlashAttention + AMD Triton:

IMPORTANT: Do not install flashattention offered by pip get official ROCm\flashattention from github.

export FLASH_ATTENTION_TRITON_AMD_ENABLE=TRUE

python main.py \
  --use-flash-attention \
  --disable-xformers 

Benchmarks

You can find raw benchmarks here.