Flint-40B-A2.5B-UD-ctx16k on RTX 3090 (24GB)
MeasuredOn 2026-10-22, Flint-40B-A2.5B-UD-ctx16k Q3_K_XL ran at 128.9 tok/s (chat-short) on an RTX 3090 (24GB) with llama.cpp. Run r_v-sdiljpm7i.
Workload results
| Workload | Backend | Model | decode tok/s | prefill tok/s | TTFT | p50 | p95 |
|---|---|---|---|---|---|---|---|
| chat-short | llama.cpp@0.7.0-dev (build 11803, commit a37cc4610) | Flint-40B-A2.5B-UD-ctx16kQ3_K_XL | 128.9tok/s | 371.7tok/s | 264ms | 7.7ms | 8.0ms |
| chat-long | llama.cpp@0.7.0-dev (build 11803, commit a37cc4610) | Flint-40B-A2.5B-UD-ctx16kQ3_K_XL | 123.0tok/s | 2,739.8tok/s | 1,100ms | 8.1ms | 8.2ms |
| concurrent-decode | llama.cpp@0.7.0-dev (build 11803, commit a37cc4610) | Flint-40B-A2.5B-UD-ctx16kQ3_K_XL | 128.7tok/s | — | — | 7.6ms | 8.0ms |
| agent-trace | llama.cpp@0.7.0-dev (build 11803, commit a37cc4610) | Flint-40B-A2.5B-UD-ctx16kQ3_K_XL | 124.4tok/s | 6,080.5tok/s | 330ms | 7.9ms | 8.2ms |
Reproduce on your machine
Start with this model and workload selection. This command does not capture the original artifact, backend version or server settings:
Match quantization, context capacity, caching and runtime settings for a useful comparison. Duration varies by workload and hardware; the published client may differ from this run's version. How it's measured.
Share this run
Copy the short link, a Reddit-ready summary, or a README badge. Every number is the measured result on this page, and each badge links back to it.
https://llm-speed.com/r/r_v-sdiljpm7iRelated benchmarks
- More Flint-40B-A2.5B-UD-ctx16k benchmarks — every backend and rig that has run this model.
- More RTX 3090 (24GB) LLM benchmarks — every model measured on this hardware.
Provenance
- Run ID
- r_v-sdiljpm7i
- Fingerprint hash
- Public key
- 29R0GbTYySfLmQ8UCn6kqoGvp0LUuOvt5yuFiDmvjj8=
- Received
- 2026-10-22 01:23:04