Pull requests / #1225

#1225 bench: add R9 dual RTX 2080 Ti community results

closed · @cezartolvai · 0 commentaires · Sur GitHub

BenchmarksMulti-GPUNVIDIA / CUDAModels & quants

Description

## Summary

Adds a results-only community benchmark report for a Ryzen 9 5950X with 2x RTX 2080 Ti (SM 75), using Flash-Next GSQ-RCO IQ3_XXS at 131K context and an automatic two-GPU layer split.

- 3 fresh-prompt runs at 4K and 32K, each with 256 greedy output tokens
- Raw streamed responses, requests, engine metrics, and aggregate JSON included
- No model files, packs, credentials, or engine changes

## Median results

| Prompt | Prompt tok/s | Decode tok/s |
| ---: | ---: | ---: |
| 4,096 | 411.1 [404.0-413.9] | 52.8 [52.6-54.2] |
| 32,768 | 477.2 [435.0-574.0] | 42.5 [42.5-42.8] |

The report retains the 32K prefill variation and documents limitations: no 128K, recall, vision, thermal, or concurrency runs.

Sur le site

Liens install, modèles, releases.