Strata benchmarks
Strata is a local inference runtime for Qwen3.8-Flash-Next.
101 community reports
Strata results by GPU
Hardware with reported Strata results, with each report's model, quant and speed, newest first. Runs on several cards, with speculative decoding, with batched requests or with part of the model in system RAM say so, since each describes a different setup. Prompt processing is how fast the input is read, not how long you wait for the first word. How to measure time to first token.
Strata on NVIDIA RTX 309025 reports
- Qwen3.8 125B · 6B active · Q438.0 t/s(part in system RAM)
- Qwen3.8 125B · 6B active · IQ3_S130.0 t/s(2 cards)
- Edge0 145B · q3k48.6 t/s, 1,481 t/s prompt(speculative, part in system RAM)
- Qwen3.8 125B · 6B active · UD IQ4_XS80.4 t/s, 1,036 t/s prompt(2 cards)
- Qwen3.8 125B · 6B active · IQ441.0 t/s(part in system RAM)
Strata on NVIDIA RTX 509014 reports
- Qwen3.8 125B · 6B active · IQ3_S6,710 t/s prompt(speculative, part in system RAM)
- Qwen3.8 125B · 6B active · IQ3_S236.0 t/s, 6,272 t/s prompt(speculative, part in system RAM)
- DeepSeek V4.1 Flash 552B · 16B active · IQ2_XXS35.0 t/s(part in system RAM)
- Qwen3.8 125B · 6B active · IQ3_S164.6 t/s(part in system RAM)
- DeepSeek V4.1 Flash 552B · 16B active10.0 t/s(part in system RAM)
Strata on NVIDIA RTX 3060 12GB9 reports
- Qwen3.8 125B · 6B active · IQ2_XS33.0 t/s, 109 t/s prompt(part in system RAM)
- Qwen3.6 35B · 3B active · Q2_K_XL118.0 t/s(part in system RAM)
- Qwen3.6 35B · 3B active · Q2_K_XL106.0 t/s(part in system RAM)
- Qwen3.6 35B · 3B active · UD-Q2_K_XL71.5 t/s, 1,539 t/s prompt(part in system RAM)
- Qwen3.8 125B · 6B active · IQ3_S(GSQ-RCO)33.0 t/s, 440 t/s prompt(part in system RAM)
Strata on NVIDIA RTX 40908 reports
- Model not named · Q4_K_XL25.0 t/s(2 cards)
- Qwen3.8 125B · 6B active · IQ2_XS122.0 t/s(part in system RAM)
- Qwen3.8 125B · 6B active · IQ3_S2,076 t/s prompt(speculative, part in system RAM)
- Qwen3.8 27B · Q3S95.0 t/s(2 cards)
- Qwen3.8 125B · 6B active · UD-IQ4_XS72.0 t/s, 2,000 t/s prompt(speculative, part in system RAM)
Strata on NVIDIA RTX 5070 Ti8 reports
- Qwen3.8 125B · 6B active · IQ3_S89.2 t/s(part in system RAM)
- Qwen3.8 125B · 6B active · Q390.0 t/s(part in system RAM)
- Qwen3.8 125B · 6B active · iq3s99.2 t/s, 839 t/s prompt(speculative, part in system RAM)
- DeepSeek V4.1 Flash 552B · 16B active · 1.75 bpw28.8 t/s(part in system RAM)
- Qwen3.8 125B · 6B active · IQ3_S195.0 t/s, 3,800 t/s prompt(2 cards, part in system RAM)
Strata on NVIDIA RTX 50807 reports
- Qwen3.8 125B · 6B active · UD-IQ4_XS50.0 t/s(part in system RAM)
- Model not named · IQ3100.0 t/s
- Qwen3.8 125B · 6B active75.0 t/s(part in system RAM)
- Qwen3.8 125B · 6B active · IQ3_S5,500 t/s prompt(part in system RAM)
- Qwen3.8 125B · 6B active · IQ3_XXS150.0 t/s, 4,396 t/s prompt(speculative, part in system RAM)
Strata on AMD Radeon AI PRO R9700 32GB5 reports
- DeepSeek V4.1 Flash 763B · IQ2_XXS25.6 t/s(2 cards, part in system RAM)
- Qwen3.8 125B · 6B active · IQ3_S86.8 t/s, 1,207 t/s prompt(2 cards)
- Model not named · IQ2_XXS15.0 t/s(part in system RAM)
- Qwen3.8 125B · 6B active · Q360.0 t/s(2 cards)
- Qwen3.8 125B · 6B active · IQ3_S90.0 t/s(part in system RAM)
Strata on AMD Strix Halo 128GB5 reports
- Qwen3.8 125B · 6B active · IQ3_S27.6 t/s
- Qwen3.8 125B · 6B active · UD-IQ4_XS770 t/s prompt(speculative)
- Qwen3.8 125B · 6B active · UD-IQ4_XS42.7 t/s, 1,294 t/s prompt(speculative)
- Qwen3.8 125B · 6B active · UD-IQ4_XS42.0 t/s, 887 t/s prompt(speculative)
- Qwen3.8 125B · 6B active · UD-IQ4_XS53.8 t/s, 1,293 t/s prompt
Strata on NVIDIA RTX 5060 Ti 16GB5 reports
- Qwen3.8 27B · Q4_K_XL20.0 t/s(2 cards, part in system RAM)
- Qwen3.8 125B · 6B active · IQ3_S GSQ RCO1,600 t/s prompt(part in system RAM)
- Qwen3.8 125B · 6B active · IQ3_S81.5 t/s(2 cards, part in system RAM)
- Qwen3.8 125B · 6B active · IQ3_S60.0 t/s(part in system RAM)
- Qwen3.8 · IQ1_M55.0 t/s, 1,500 t/s prompt(part in system RAM)
Strata on AMD RX 7900 XTX4 reports
- Qwen3.8 125B · 6B active · IQ3_XXS92.0 t/s, 2,619 t/s prompt(2 cards, speculative, part in system RAM)
- Qwen3.8 125B · 6B active · IQ3_XXS800 t/s prompt(part in system RAM)
- Qwen3.8 125B · 6B active · IQ3_S86.0 t/s, 3,100 t/s prompt(speculative)
- Qwen3.8 125B · 6B active · Q3_K_S1,300 t/s prompt(2 cards)
Strata on NVIDIA RTX Pro 6000 Blackwell4 reports
- Kimi K3.1 2800B · 2.81/2.56/1.44 bpw 3-tier MoE (~1.92 bpw avg)10.8 t/s(part in system RAM)
- Qwen3.8 125B · 6B active · 6bit150.0 t/s(3 cards)
- Qwen3.8 125B · 6B active · IQ3_S205.3 t/s(speculative)
- Qwen3.8 125B · 6B active · UD-Q4_K_XL3,319 t/s prompt(speculative)
Strata on NVIDIA RTX 4070 Ti Super3 reports
- Qwen3.8 125B · 6B active · UD-Q4_K_XL36.1 t/s(speculative, part in system RAM)
- Qwen3.8 · Q2197.0 t/s, 3,300 t/s prompt(part in system RAM)
- Qwen3.8 125B · 6B active · IQ3_S72.0 t/s, 1,700 t/s prompt(part in system RAM)
Strata on NVIDIA RTX 50703 reports
- DeepSeek V4.1 Flash 552B · 16B active · IQ2_XS28.0 t/s, 519 t/s prompt(speculative, part in system RAM)
- Qwen3.8 125B · 6B active · IQ3_S53.0 t/s, 1,620 t/s prompt(speculative, part in system RAM)
- Qwen3.8 125B · 6B active · IQ3_XXS44.8 t/s, 414 t/s prompt(part in system RAM)
Strata on AMD RX 6900 XT 16GB2 reports
- Qwen3.8 125B · 6B active · Q2_041.5 t/s(part in system RAM)
- Qwen3.8 125B · 6B active · GSQ-RCO25.0 t/s(part in system RAM)
Strata on AMD RX 7900 XT2 reports
- Model not named · UD-IQ1_M53.3 t/s(speculative, part in system RAM)
- Qwen3.8 125B · 6B active · UD-IQ4_XS54.0 t/s, 1,100 t/s prompt(part in system RAM)
Strata on AMD RX 9070 XT 16GB2 reports
- DeepSeek V4.1 Flash 552B · 16B active · IST-DAS 1.75 bpw7.0 t/s(part in system RAM)
- Qwen3.8 125B · 6B active · UD-IQ4_XS40.0 t/s(part in system RAM)
Strata on NVIDIA L40S2 reports
- Model not named · IQ2_XS2,200 t/s prompt(part in system RAM)
- DeepSeek V4 Flash 284B · 13B active · IQ1_M73.7 t/s, 1,975 t/s prompt(speculative, part in system RAM)
Strata on NVIDIA RTX 3060 Ti 8GB2 reports
- Qwen3.8 125B · 6B active · IQ2_XS18.0 t/s(speculative, part in system RAM)
- Qwen3.8 125B · 6B active · IQ3_XXS27.0 t/s, 475 t/s prompt(part in system RAM)
Strata on NVIDIA RTX 30702 reports
- Model not named40.0 t/s(2 cards, speculative)
- Qwen3.8 125B · 6B active · IQ3_S40.0 t/s, 1,000 t/s prompt(speculative, part in system RAM)
Strata on NVIDIA RTX 3080 20GB2 reports
- DeepSeek V4.1 Flash 552B · 16B active · IQ2-XXS37.0 t/s(4 cards, part in system RAM)
- Qwen3.8 125B · 6B active · IQ3_XXS105.0 t/s, 5,000 t/s prompt(4 cards)
Strata on NVIDIA RTX 3090 Ti2 reports
- Qwen3.8 125B · 6B active · IQ3_S148.9 t/s, 1,887 t/s prompt(2 cards, speculative)
- Qwen3.8 125B · 6B active · IQ3-XXS105.0 t/s, 2,000 t/s prompt(2 cards, speculative)
Strata on NVIDIA RTX 40702 reports
- Qwen3.8 125B · 6B active · IQ3_S40.0 t/s, 1,000 t/s prompt(part in system RAM)
- Qwen3.8 27B · Q4_K_M13.0 t/s(2 cards)
Strata on NVIDIA RTX 4070 Super2 reports
- Qwen3.8 125B · 6B active · GGUF33.2 t/s, 947 t/s prompt(part in system RAM)
- Qwen3.8 125B · 6B active40.8 t/s, 879 t/s prompt(part in system RAM)
Strata on NVIDIA RTX 40802 reports
- Qwen3.8 125B · 6B active · GSQ RCO IQ3_XXS95.0 t/s, 3,000 t/s prompt(part in system RAM)
- Qwen3.8 125B · 6B active · Q355.0 t/s, 590 t/s prompt(part in system RAM)
Strata on NVIDIA RTX 4080 Super2 reports
- Qwen3.8 125B · 6B active · q3_xss1,000 t/s prompt(part in system RAM)
- Qwen3.8 125B · 6B active · IQ3_S40.0 t/s(part in system RAM)
Strata on NVIDIA V100 16GB2 reports
- Qwen3.8 125B · 6B active · Q2_052.0 t/s, 399 t/s prompt(speculative, part in system RAM)
- Qwen3.8 125B · 6B active · UD-Q4_K_XL7,090 t/s prompt(4 cards, speculative, part in system RAM)
Strata on NVIDIA V100 32GB2 reports
- DeepSeek V4.1 Flash 552B · 16B active · IQ2_XXS23.8 t/s, 127 t/s prompt(2 cards, part in system RAM)
- Qwen3.8 125B · 6B active · IQ3_XXS16.7 t/s, 62 t/s prompt(speculative, part in system RAM)
Strata on AMD MI50 32GB1 report
- DeepSeek V4.1 Flash 552B · 16B active · IQ1_M/IQ2_XS/IQ2_XXS (IST-DAS 2.00bpw)23.0 t/s, 170 t/s prompt(speculative, part in system RAM)
Strata on AMD RX 6750 XT1 report
- Qwen3.8 125B · 6B active · 4-bit25.0 t/s, 150 t/s prompt(part in system RAM)
Strata on AMD RX 6800 XT 16GB1 report
- Qwen3.8 125B · 6B active · TQ2_025.0 t/s, 700 t/s prompt(part in system RAM)
Strata on AMD RX 7600 8GB1 report
- Qwen3.8 125B · 6B active · IQ2_XS24.0 t/s, 68 t/s prompt(part in system RAM)
Strata on AMD RX 7800 XT 16GB1 report
- Qwen3.8 125B · 6B active · IQ3_XXS64.9 t/s, 986 t/s prompt(speculative, part in system RAM)
Strata on AMD RX 9060 XT 16GB1 report
- Qwen3.8 125B · 6B active · Q2_045.5 t/s, 162 t/s prompt(speculative, part in system RAM)
Strata on AMD RX 90701 report
- Qwen3.8 125B · 6B active · IQ260.0 t/s(part in system RAM)
Strata on Intel Arc B580 12GB1 report
- Qwen3.8 125B · 6B active · IQ2_XS21.0 t/s(speculative, part in system RAM)
Strata on Intel Arc Pro B701 report
- Qwen3.8 125B · 6B active · IQ3_XXS51.1 t/s, 1,456 t/s prompt(part in system RAM)
Strata on M1 Max 64GB1 report
- Qwen3.8 125B · 6B active24.0 t/s(part in system RAM)
Strata on M2 Pro 16GB1 report
- Qwen3.8 125B · 6B active · iQ3_xxs5.5 t/s, 77 t/s prompt(part in system RAM)
Strata on M4 24GB1 report
- Qwen3.8 125B · 6B active · Q2_08.1 t/s(part in system RAM)
Strata on M5 Max 128GB1 report
- Qwen3.8 125B · 6B active74.0 t/s
Strata on M5 Pro 48GB1 report
- Qwen3.8 125B · 6B active · IQ3_S32.0 t/s(speculative, part in system RAM)
Strata on M6 32GB1 report
- Qwen3.8 125B · 6B active · Q2_010.0 t/s, 200 t/s prompt
Strata on NVIDIA CMP 170HX 64GB (unlocked)1 report
- Qwen3.8 125B · 6B active · Q4_K_XL100.0 t/s(2 cards, part in system RAM)
Strata on NVIDIA CMP 170HX 8GB1 report
- Qwen3.8 125B · 6B active · IQ3_XXS121.0 t/s, 630 t/s prompt(2 cards, speculative, part in system RAM)
Strata on NVIDIA H100 80GB1 report
- Qwen3.6 35B · 3B active · Q8_0290.0 t/s, 11,486 t/s prompt
Strata on NVIDIA RTX 2080 Ti 22GB (modded)1 report
- Qwen3.8 125B · 6B active · IQ3_S60.1 t/s, 28 t/s prompt(2 cards, part in system RAM)
Strata on NVIDIA RTX 4060 Ti 16GB1 report
- GLM-5.3 320B · 18B active · UD-IQ3_XXS13.0 t/s(2 cards, part in system RAM)
Strata on NVIDIA RTX 5070 Ti Laptop 12GB1 report
- Qwen3.8 125B · 6B active · IQ3_XXS51.0 t/s, 1,500 t/s prompt(part in system RAM)
Strata on NVIDIA RTX Pro 4500 Blackwell 32GB1 report
- Qwen3.8 125B · 6B active · NVFP41,680 t/s prompt(speculative, part in system RAM)
Strata on NVIDIA T4 16GB1 report
- Qwen3.8 125B · 6B active · UD-Q4_K_XL38.7 t/s, 1,160 t/s prompt(2 cards, part in system RAM)
Strata on NVIDIA Tesla P100 16GB1 report
- Qwen3.8 27B30.0 t/s(part in system RAM)
Strata on NVIDIA Tesla P40 24GB1 report
- Qwen3.8 125B · 6B active · IQ2_XS34.8 t/s, 501 t/s prompt(2 cards)
Strata, GPU not identified30 reports
- Qwen3.8 125B · 6B active61.0 t/s
- Qwen3.8 125B · 6B active13.0 t/s(part in system RAM)
- Qwen3.8 125B · 6B active · UD-Q4_K_XL50.0 t/s
- Qwen3.8 27B34.7 t/s(3 cards, speculative)
- Qwen3.8 125B · 6B active · IQ2_XS50.0 t/s, 1,000 t/s prompt
What is Strata?
Strata is a local inference runtime for Qwen3.8-Flash-Next. It spreads work across the GPU, system RAM and storage, so available VRAM alone does not describe the setup. Its upstream project documents Windows and Linux installation, a chat interface and APIs for local applications. Check the repository and build in each report before comparing results.
Start with the upstream project for its supported configurations. Community ports can have different installation requirements and hardware-specific optimizations.
Source: Upstream Strata
Strata-V100 is a community fork
Strata-V100 maintains a Volta-focused version of Strata and periodically incorporates upstream changes. Its documentation requires CUDA 12.x and a build containing sm_70 support. CUDA 13 cannot compile that target. Features added for newer GPU architectures do not automatically improve V100 performance.
Record whether a result used this fork, upstream Strata or another build. Keep the V100 memory variant, card count, PCIe layout, system RAM and expert placement beside the result. The V100 32GB hardware page provides additional reported configurations.
Source: Strata-V100 documentation
Strix Halo needs its own evidence
Upstream Strata lists Strix Halo among its experimental community paths, built from source on Linux. Use the Strix Halo reports to inspect the model, operating system, runtime build and speculation settings actually tested. A result from an NVIDIA fork cannot establish performance on the AMD system.
Source: Upstream platform documentation
Read the workload before the speed
Compare actual input and output counts, prefix reuse, quantization, MTP settings and request concurrency. Configured context capacity does not tell you how much text a benchmark processed.
The V100 fork preserves a historical long-context sweep that recorded thermal slowdown during its 128K and 256K prompts. Those measurements belong to the documented build and cooling conditions. They are not measurements of every later revision.
Use the benchmark-reading guide before comparing engines. Keep prefill, generation and application latency separate, and treat missing settings as unknown.
Source: V100 measurement details
Source-check footer: 13 October 2026. Community-reported evidence, not tests performed by LlamaPerf. Recheck changing platform requirements before publishing.
Compare GPUs running Strata
Every card is compared on one model, Qwen3.6 35B · 3B active at 8-bit, the size and bit class that most cards here have plain runs of. A card with no run of that model shows none rather than a figure from a different model.
| GPU | VRAM | Reports | Median speed, Qwen3.6 35B · 3B active 8-bit |
|---|---|---|---|
| NVIDIA RTX 3090nvidia | 24GB | 33 | no Qwen3.6 35B · 3B active 8-bit run reported |
| NVIDIA RTX 5090nvidia | 32GB | 16 | no Qwen3.6 35B · 3B active 8-bit run reported |
| NVIDIA RTX 4090nvidia | 24GB | 11 | no Qwen3.6 35B · 3B active 8-bit run reported |
| NVIDIA RTX 3060 12GBnvidia | 12GB | 10 | no Qwen3.6 35B · 3B active 8-bit run reported |
| AMD RX 7900 XTXamd | 24GB | 9 | no Qwen3.6 35B · 3B active 8-bit run reported |
| NVIDIA RTX 5070 Tinvidia | 16GB | 9 | no Qwen3.6 35B · 3B active 8-bit run reported |
| NVIDIA RTX 5080nvidia | 16GB | 7 | no Qwen3.6 35B · 3B active 8-bit run reported |
| AMD Radeon AI PRO R9700 32GBamd | 32GB | 6 | no Qwen3.6 35B · 3B active 8-bit run reported |
Strata against other engines
Pairs of plain runs on the same card, of the same model size at the same bit class: one device, one request, no speculative decoding, the whole model in memory. Context length and build still differ between the two sides, and each side shows its own.
No matched pair yet. No card has plain runs of one model size at one bit class on Strata and on another engine, so llamaperf can't say how it compares on speed. Add a run.
Top models on Strata
Frequently asked
Is Strata faster than other engines?
llamaperf has no matched comparison for Strata yet: no card has plain runs of the same model size at the same bit class on Strata and on another engine. Speed claims about engines need that pairing, so this page doesn't make one.