Compare
Compare local AI hardware with workload-aware output.
B100 192GB wins for coding in balanced mode
Based on model fit, speed, and quality across top recommendations.
B100 192GB
WinnerSDevstral 2 123B Instruct
llama.cppq6-kRuns well
126.3 GB / 192.0 GB
76.1 tok/s212K ctx
SQwen 3.5 122B A10B
llama.cppq8-0Runs well
153.1 GB / 192.0 GB
169.4 tok/s131K ctx
SGPT-OSS 120B
llama.cppq8-0Runs well
150.2 GB / 192.0 GB
64.2 tok/s131K ctx
Quick comparison
| Metric | B100 192GB | RTX 4090 24GB |
|---|---|---|
| Models that fit | 3 | 3 |
| Avg decode tok/s | 103.2 | 37.9 |
| Best grade score | 99 | 93 |