gpt-oss 120B › NVIDIA DGX Spark (GB10)
gpt-oss 120B on the NVIDIA DGX Spark (GB10)
Best measured: 62.4 tokens/s generation and 2,444 tokens/s prompt processing (Q4_K_M, CUDA, FA on).
For comparison, our size-scaling estimate for this combination is 43.4 tok/s.
Every run
| Quant | Backend | Generate tok/s | Prompt tok/s | Test | Build / by / notes | Source |
|---|---|---|---|---|---|---|
| Q4_K_M | CUDA FA on | 62.4 | 2,444 | tg32 | am17an llama-bench model column: gpt-oss 120B Q4_K - Medium | llama.cpp discussion #16578 |
| MXFP4 | CUDA FA on | 58.7 | 2,444 | tg32 | ggml-org maintainers Benchmarked repo: https://huggingface.co/ggml-org/gpt-oss-120b-GGUF | llama.cpp benches/ (maintainer runs) |
| MXFP4 | CUDA FA on | 57.2 | 1,816 | tg128 | phoyd llama-bench model column: gpt-oss 120B MXFP4 MoE | llama.cpp discussion #15396 |
tg128 = generating 128 tokens; pp512 = reading a 512-token prompt (llama-bench). LocalScore averages several prompt and output lengths, so its numbers run lower than tg128.
No card yet?
Try it on a rented GPU (RunPod)Affiliate link: if you sign up, RigBench may earn a referral credit. It never affects rankings or numbers.
All results for the NVIDIA DGX Spark (GB10) · gpt-oss 120B on other hardware