#2120 Community benchmark: GeForce RTX 5090 + NVIDIA RTX 4000 Ada

Descrição

# Community benchmark on NVIDIA GeForce RTX 5090 + NVIDIA RTX 4000 Ada Generation

Measured on 2026-10-12 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 5090 + NVIDIA RTX 4000 Ada Generation, 52 GiB VRAM
- CPU: AMD Ryzen Threadripper 3970X 32-Core Processor (32 cores, 64 threads); installed RAM: 125.2 GiB (KF3200C16D4/32GX); storage: Samsung SSD 990 EVO Plus 4TB
- OS: Linux 6.12.0-211.53.1.el10_2.x86_64; GPU driver: 615.71.09; CUDA: not read
- Strata: project 0.1.43, commit 7695859a; engine 0.1.43
- Background workloads and power limits: not recorded
- Motherboard: PRIME-TRX40-PRO-S-10DE

## Model and configuration

- Model and quantization: qwen3.8-flash-next-iq3_s
- Context: 262144 tokens; parallel slots: 3
- Settings: the install's own configuration, unchanged; thinking off for the suite; temperature 0

```text
strata --pack iq3_s --native Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf --ple-gguf Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf --expert-profile expert-profile.bin --expert-cache auto --prefill auto --spec 4 --mtp rt --max-context 262144 --kv int8 --kv-resident 32768 --vision --vram-reserve-mib 700 --control-vector-scaled Qwen3.8-Flash-Next-experimental-speed-projection.gguf:1.0 --control-vector-layer-range 4 44 --cvec-mode project --cvec-dir per-layer --spec-min-p 0.5 --batch-groups 2 --trim-stage-weights --conversation-cache-mib 8192 --conversation-cache-slots 6 --conversation-cache-min-free-mib 4096
```

## Method

Run by `python tools/strata_bench.py` against the HTTP API of a Strata server (already running); 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 | 135.7 (113.6-141.9) | 0.25 (0.22-0.30) |
| 4K prompt, fresh | 4019 (4019-4019) | 0 (0-0) | 24 cap | 3 | 2810.5 (2800.9-2816.6) | not measured | 1.45 (1.44-1.45) |
| 16K prompt, fresh | 16058 (16058-16058) | 0 (0-0) | 24 cap | 3 | 2673.5 (2638.1-2674.6) | not measured | 6.03 (6.03-6.11) |
| agent session, follow-up turns | 14423 (14422-14432) at the last turn | see below | up to 64 | 3 | see below | 167.5 (125.0-235.8) | 1.42 (1.30-1.48) |

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: 68.7 % (68.6-68.8); earlier-turn tokens read again per session: 3 (3-73) (earlier turns that the cache did not hold; the 13633 (13633-13635) new tokens of the files are read once); first turn (nothing cached) first token: 0.99 s (0.96-1.03).

| Turn (run 1) | Prompt tokens | Reused | Read | Prompt tok/s | First token (s) | Decode tok/s |
| ---: | ---: | ---: | ---: | ---: | ---: | ---: |
| 0 | 609 | 0 | 609 | 645.9 | 0.957 | 100.9 |
| 1 | 3709 | 629 | 3080 | 2186.3 | 1.426 | 128.9 |
| 2 | 5242 | 3772 | 1470 | 1144.9 | 1.302 | 136.5 |
| 3 | 8686 | 5306 | 3380 | 2324.1 | 1.476 | 142.5 |
| 4 | 11571 | 8711 | 2860 | 2047.2 | 1.42 | 167.5 |
| 5 | 14423 | 11577 | 2846 | 2040.7 | 1.423 | 133.7 |

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: 64 s.


[results.json](https://github.com/user-attachments/files/33342072/results.json)