Compare
Compare local AI hardware with workload-aware output.
Operating mode: Balanced. Balanced for general local use. Keeps the ranking neutral across personal and serving workflows.
RTX 5090 32GB wins for coding in balanced mode
Based on model fit, speed, and quality across top recommendations.
AMD Instinct MI60 32GB
SQwen 3.6 27B
llama.cppQ4_K_MRuns well
21.5 GB / 32.0 GB
20.5 tok/s187K ctx
SDevstral Small 2 24B Instruct
llama.cppQ4_K_MRuns well
21.2 GB / 32.0 GB
36.8 tok/s87K ctx
SQwen 3.5 27B
llama.cppQ4_K_MRuns well
23.7 GB / 32.0 GB
32.9 tok/s58K ctx
RTX 5090 32GB
WinnerSQwen 3.6 27B
llama.cppQ4_K_MRuns well
21.5 GB / 32.0 GB
35.1 tok/s187K ctx
SDevstral Small 2 24B Instruct
llama.cppQ4_K_MRuns well
21.2 GB / 32.0 GB
62.0 tok/s87K ctx
SQwen3-Coder 30B A3B Instruct
llama.cppQ4_K_MRuns well
24.2 GB / 32.0 GB
130.7 tok/s102K ctx
Quick comparison
| Metric | AMD Instinct MI60 32GB | RTX 5090 32GB |
|---|---|---|
| Models that fit | 3 | 3 |
| Avg decode tok/s | 30.1 | 75.9 |
| Best grade score | 96 | 100 |