#2668 Community benchmark: RTX 4090 48GB and 64GB system RAM. unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q4_K_XL

Description

# Community benchmark on NVIDIA GeForce RTX 4090 48GB and 64GB system RAM
_(Custom RTX 4090 with doubled VRAM. Should be on par with RTX 6000 Ada)_

Measured on 2026-10-15 with `tools/strata_bench.py` (suite 1, 3 runs per test, medians with ranges). Speeds are the engine's own timings; first-token times are measured by the client over HTTP. Speed only: no correctness check was run.

## Hardware and software

- GPU and VRAM: NVIDIA GeForce RTX 4090, 48 GiB VRAM (Water Cooled)
- CPU: Intel(R) Xeon(R) CPU E3-1220 v6 @ 3.00GHz (4 cores, 4 threads); 
- installed RAM: 60.7 GiB (DDR4-2666); 
- storage: ADATA XPG MARS 980 (2000000 IOPS)
- OS: Linux 7.2.6-070206-generic; GPU driver: 615.71.09; CUDA: 13.3
- Strata: project 0.1.44.1, commit not a git checkout; engine 0.1.44.1
- Background workloads and power limits: 4090 at 300W limit

## Model and configuration

- Model and quantization: qwen3.8-flash-next-unsloth-ud-q4_k_xl
- Context: 262144 tokens; parallel slots: 1
- Settings: the install's own configuration, unchanged; thinking off for the suite; temperature 0

```text
strata --pack unsloth-ud-q4_k_xl --native Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf --expert-profile expert-profile.bin --expert-cache auto --prefill auto --spec 4 --mtp rt --max-context 262144 --kv int8 --kv-resident 32768 --resident-budget-gib 33 --pcie-frac 0.00 --spec-min-p 0.70 --pool-workers 2
```

Environment of the config: `STRATA_HC_Q8=1`

## Method

Run by `python tools/strata_bench.py` against the HTTP API of a Strata server (started by the benchmark from the install configuration); model loading is not timed. One warm-up chat and a 1K prompt are discarded first. Every prompt starts with a unique tag, so the conversation cache cannot help the fresh-prompt tests.

- short chat: the three questions of `tools/ab_engine.py`, up to 100 new tokens, 3 runs
- 4K / 16K prompts: generated word lists sized by a calibration request, 24 new tokens, 3 fresh prompts each
- agent session: system prompt and four tools, then 5 turns that each add about 1500 words of a project file and continue from the model's own earlier answers (the way agent clients resend a conversation), 64 new tokens per turn, 3 sessions; the cache numbers come from the answer's usage (`cached_tokens`) and timings (`prompt_n`)
- concurrency (when run): N requests sent at the same moment, total tokens over the wall time of the burst, latency over all requests

## Results

| Configuration | Actual prompt tokens | Reused tokens | Generated tokens | Runs | Prompt tok/s median and range | Decode tok/s median and range | TTFT seconds median and range |
| --- | ---: | ---: | ---: | ---: | --- | --- | --- |
| short chat | 46 (43-51) | 0 | 100 (100-100) | 9 | not measured | 72.4 (31.8-91.2) | 0.36 (0.27-0.77) |
| 4K prompt, fresh | 4019 (4019-4019) | 0 (0-0) | 24 cap | 3 | 1103.8 (1066.4-1105.4) | not measured | 3.66 (3.66-3.79) |
| 16K prompt, fresh | 16058 (16058-16058) | 0 (0-0) | 24 cap | 3 | 2061.9 (1971.2-2072.0) | not measured | 7.82 (7.78-8.18) |
| agent session, follow-up turns | 14544 (14516-14596) at the last turn | see below | up to 64 | 3 | see below | 85.8 (46.5-134.3) | 3.64 (3.40-3.86) |

Ranges are the lowest and highest of the runs; `Runs` counts requests that returned a token.

### Agent session (conversation cache)

Cache hit over the follow-up turns: 68.5 % (68.3-68.7); earlier-turn tokens read again per session: 74 (4-214) (earlier turns that the cache did not hold; the 13729 (13728-13730) new tokens of the files are read once); first turn (nothing cached) first token: 1.80 s (1.80-1.82).

| Turn (run 1) | Prompt tokens | Reused | Read | Prompt tok/s | First token (s) | Decode tok/s |
| ---: | ---: | ---: | ---: | ---: | ---: | ---: |
| 0 | 609 | 0 | 609 | 342.2 | 1.803 | 45.8 |
| 1 | 3744 | 621 | 3123 | 815.0 | 3.858 | 46.5 |
| 2 | 5277 | 3808 | 1469 | 425.6 | 3.476 | 58.8 |
| 3 | 8721 | 5340 | 3381 | 889.7 | 3.828 | 66.6 |
| 4 | 11600 | 8738 | 2862 | 774.4 | 3.726 | 57.7 |
| 5 | 14516 | 11617 | 2899 | 785.2 | 3.723 | 46.6 |

Per-request data: `results.json`. Engine log lines from the start: `engine-log-excerpt.txt` (when this run started the server).

## Correctness and limitations

Not measured: this is a speed suite. The conversation uses the model's own replies, so token counts differ a little between runs. Machine details come from the server, `nvidia-smi`/`rocm-smi` and the OS; background load and power limits are not recorded. Total time of the suite: 135 s.

## Peak Temperatures
- GPU: 55C
- SSD: 59C