Q4 and q5: Suitable for most users, with moderate memory requirements and good generation quality. q8: Suitable for high-end graphics card users, the generation quality is close to the original version. Using flux gguf version in comfyui? comfyui uses: Deepseek 32b requires at least 21gb of video memory, and the qwq 32b q8 quantized version requires 34gb of video memory. At the same time, the requirements are much higher than the previous models. I have tried 2080ti 22gb * 2, radeonvii 16gb * 2, and 4090 48gb.
⁴ᴷ MTA Bus 2023 Volvo Group Nova Bus LFS on the Q8 (JFK Depot) YouTube