$2800 Rig with 8x Radeon Pro V620 Achieves High-Speed Qwen3.8-Flash-Next Inference.

Core summary

One sentence to understand this update

A $2800 rig featuring 8x Radeon Pro V620 GPUs (256GB VRAM) and a custom vLLM fork can run Qwen3.8-Flash-Next at 60-100 t/s decode and over 3000 t/s prefill.

Impact & opportunity

What this could mean

Builders and researchers can learn from this cost-effective hardware configuration to achieve high-performance local large model inference deployments on a limited budget.