Compare
Compare local AI hardware with workload-aware output.
AMD Instinct MI100 32GB wins for coding in balanced mode
Based on model fit, speed, and quality across top recommendations.
AMD Instinct MI100 32GB
WinnerSQwen 3.6 27B
llama.cppq4-k-mRuns well
21.5 GB / 32.0 GB
32.6 tok/s187K ctx
SQwen 3.5 27B
llama.cppq4-k-mRuns well
23.7 GB / 32.0 GB
52.3 tok/s58K ctx
SNemotron 3 Nano 30B
llama.cppq4-k-mRuns well
24.8 GB / 32.0 GB
46.9 tok/s63K ctx
Quick comparison
| Metric | AMD Instinct MI100 32GB | RTX 4090 24GB |
|---|---|---|
| Models that fit | 3 | 3 |
| Avg decode tok/s | 43.9 | 37.9 |
| Best grade score | 98 | 93 |