Measured performance · local AI
Benchmarks
Loading benchmark evidence…
Measured performance · local AI
Loading benchmark evidence…
Optional user-local diagnostics
Run an optional diagnostic on your own PC. These results are separate from UOS Benchmark Intelligence, which is built from globally sourced, validated measured evidence.
Public GPU and CPU rankings use reviewed external benchmark evidence; they do not require installing or running a local agent. This diagnostics page is optional and shows only imported local runs.
| Model | GPU | Context | Prompt | Generation | Evidence |
|---|---|---|---|---|---|
| gguf:bartowski/Qwen3.8-27B-GGUF@f7f782271ce686458a3702c5d9f30d9c1354087c/Qwen3.8-27B-Q6_K_S.gguf | AMD Radeon RX 7900 XTX | 64K | 54.2 tok/s | 5.8 tok/s | Measured · reviewed |
| gguf:bartowski/Qwen3.8-27B-GGUF@f7f782271ce686458a3702c5d9f30d9c1354087c/Qwen3.8-27B-Q6_K_S.gguf | AMD Radeon RX 7900 XTX | 32K | 82.4 tok/s | 25 tok/s | Measured · reviewed |
| gguf:bartowski/Qwen3.8-27B-GGUF@f7f782271ce686458a3702c5d9f30d9c1354087c/Qwen3.8-27B-Q6_K_S.gguf | AMD Radeon RX 7900 XTX | 8K | 602.4 tok/s | 31 tok/s | Measured · reviewed |
| Qwen3.8-27B-Q6_K_S.gguf | Radeon RX 7900 XTX | 8K | 599.6 tok/s | 31 tok/s | Measured · review pending |
| Qwen3.8-27B-Q6_K_S.gguf | Radeon RX 7900 XTX | 8K | 596.8 tok/s | 31 tok/s | Measured · review pending |
Full run console for power users: hardware identity, runtime settings, variance, and review state for every run.
AMD Radeon RX 7900 XTX · Qwen3.8-27B-Q6_K_S.gguf
Run 22cd5b02 · Sep 25, 2026, 1:14 AM UTC · 8K context · Throughput benchmark and live inference both completed.
Optional user-local test; separate from UOS Benchmark Intelligence.
Against run e152e8f9, using strict comparability rules.
Already stored as local evidence. Accounts are not available yet.
The local agent supports user-run diagnostics. It does not supply or qualify UOS Benchmark Intelligence indices.
Local agent · this PC only
Detected hardware and run controls appear here once the agent is connected.
Needs a reviewed model identity before it can become shared benchmark evidence.
Your system
Local model
The exact file is pinned by its SHA-256 fingerprint, but it is not a reviewed artifact or model. Results stay local evidence until it is.
Configuration
8K
This run
32K
Not measured
64K
Not measured
Measured results
Prompt processing
596.8tok/s
llama-bench, 512-token prompt at 8192 context depth
Token generation
30.99tok/s
±0.03 tok/s · llama-bench, 128 generated tokens at 8192 context depth
Time to first token
364 ms
llama-server, short prompt
Server ready (load)
10.6 s
llama-server start until ready (includes model load)
Live generation
30.36tok/s
llama-server, 8 generated tokens
Repetitions
History
3runs
At 8K context depth
Hover a value for how it was measured. “Unknown” means not measured — never zero.
Charts
Context scaling needs at least two measured context depths. This run: 8K measured.
Evidence
| 4a06cc98 |
| AMD Radeon RX 7900 XTX |
| Unknown · Q6_K_S |
| 8K |
| 602.4 tok/s |
| 31.03 tok/s |
| Completed · reviewed |
| Sep 25, 2026, 7:46 PM UTC |
| e63fde06 | AMD Radeon RX 7900 XTX | Qwen3.8-27B-Q6_K_S.gguf · Q6_K_S | 8K | 599.6 tok/s | 31.03 tok/s | Completed · pending | Sep 25, 2026, 1:47 AM UTC |
| 22cd5b02 | AMD Radeon RX 7900 XTX | Qwen3.8-27B-Q6_K_S.gguf · Q6_K_S | 8K | 596.8 tok/s | 30.99 tok/s | Completed · pending | Sep 25, 2026, 1:14 AM UTC |
Select two or more runs. The first selected is the baseline.