Completed 8x AMD MI50 - 256GB VRAM + 256GB RAM rig for $3k

gpu4.pics
via MLDataScientist
AMD Instinct MI50 32GB
6× AMD Instinct MI50 32GB + 2× AMD Instinct MI60 32GB · 256 GB total VRAM · AMD EPYC 7532 (32-core)
32 GB VRAM AMD 300 W TDP
running gpt-oss 120B· Q8
58
tokens / sec
Generation
58tok/s
higher is better
Prompt / prefill
tok/s
higher is better
Time to first token
lower is better
Composite score
Model size
120B
Runtime
AMD EPYC 7532 (32-core) · 256 GB RAM
gpu4.pics/r/curated-1nhd5ks

All benchmarks on this rig

ModelQuantGenPromptTTFTPowerPerf/W$/MtokScore
gpt-oss 120BQ858
Llama 3.3 70BAWQ25
Qwen3 235BQ4_121