2026-10-08 · America/Los_Angeles · 社区动态 · #14

$2800 rig with 8x Radeon Pro V620 (256 GB VRAM) + custom vLLM fork = Qwen3.8-Flash-Next at 60 to 100 t/s decode and 3000+ t/s prefill

Post title is slightly misleading, I don't think you can get these for $350 each anymore but they're still pretty cheap all things considered. They're Radeon Pro V620's which are older RDNA2 enterprise cloud gaming cards with 32 GB VRAM. (Ignore the RTX 4090 on the side, it's just used for stuff like image/video gen models, no LLMs) But I bought these cards a couple months ago as a gamble to see if I could build a big VRAM rig with usable speed for relative peanuts. I was struggling with llama.cpp for a long time, but the prefill was pretty bad (around 350-450 t/s average with this same…

热度 47.4 / 100；排名与评分保留该期记录。