>Hint: your device supports --pin-shared-memory for speed improvements if having 16GB or more VRAM. It may cause issues with 12GB VRAM or less.
>Hint: your device supports --cuda-malloc for a bit less vram usage and small speed improvement
>Hint: your device supports --cuda-stream for potential speed improvements but it may cause issues on some GPUs that have 6-8GB VRAM or less.
What is this scam? I have a 3090 with 24GB VRAM and enabling all those made it gen slower, not faster.
Genning a batch count of 5 took 36.1 seconds with default settings, 44.5 seconds with those settings enabled.
Batch size of 5 instead is basically same, settings default 34.2 seconds, settings enabled 36.2 seconds