A100-SXM4-40GB · 40 GB
All tasks · Qualified results
No qualified results yet · All tasks · A100-SXM4-40GB · 40 GB
6 original reports are kept as evidence but do not qualify for comparison; each report states why. This does not show that the model or GPU cannot run it.
Measured alternatives on other GPUs
Measured on different GPUs with the same or smaller listed capacity. These are references for choosing hardware, not performance or compatibility evidence for your selected GPU.
Qwen3.8-27B-NVFP4
RadixArk · 2 tested configurations · 1 GPU
- RTX 5090 · 32 GBNVFP4 · 8K → 2K tokens
- Output throughput · includes prefill · 1 request
- 70.8tok/s
- Observed GPU occupancy
- 27.3GiB
- First token
- 633 ms
Test details
Total output tokens divided by the full request window, including prefill and queueing.
Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.
Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.
- Engine
- vllm
- Protocol
- wbench-fixed-http-v1
- Dataset
- synthetic-tokenizer-v1:repeats=2
- Workload
- 8192in-2048out
Concurrency Mean throughput (tok/s) P95 TTFT Observed occupancy (GiB) GPU power (W) Reports combined View report C1 70.82 29.01 s 633 ms 27.29 405.5 1 bad294c0 ↗ 1 other test configurations on this GPU
- NVFP4 · 32K → 2K tokens
- Output throughput · includes prefill · 1 request
- 62.5tok/s
- Observed GPU occupancy
- 27.3GiB
- First token
- 3.6 s
Test details
Total output tokens divided by the full request window, including prefill and queueing.
Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.
Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.
- Engine
- vllm
- Protocol
- wbench-fixed-http-v1
- Dataset
- synthetic-tokenizer-v1:repeats=2
- Workload
- 32768in-2048out
Concurrency Mean throughput (tok/s) P95 TTFT Observed occupancy (GiB) GPU power (W) Reports combined View report C1 62.48 32.79 s 3.6 s 27.29 433.3 1 746bbc39 ↗
clef-flash
Cloudflare · 4 tested configurations · 1 GPU
- vGPU-32GB · 32 GBbf16 · Text → Structured data · typed-invoices-v1 · Test duration 1 s
- Throughput · Batch 1
- 10.3decisions/sec
- Best measured · Batch 4
- 10.9decisions/sec
- Observed GPU occupancy
- 18.2GiB
- P95 latency
- 104 ms
Test details · 2 batch sizes
Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.
Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.
- Engine
- clef-native
- Protocol
- native-forward-v1
- Dataset
- typed-invoices-v1
- Workload
- decision
Batch size Mean throughput (decisions/sec) P95 TTFT Observed occupancy (GiB) GPU power (W) Reports combined View report B1 10.26 104 ms — 18.17 67.9 1 1d1661ad ↗ B4 10.89 389 ms — 18.49 156 1 1d1661ad ↗ 3 other test configurations on this GPU
- bf16 · Text → Structured data · typed-invoices-v1 · Test duration 6 s
- Throughput · Batch 1
- 10.6decisions/sec
- Best measured · Batch 4
- 11decisions/sec
- Observed GPU occupancy
- 18.2GiB
- P95 latency
- 99 ms
Test details · 2 batch sizes
Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.
Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.
- Engine
- clef-native · PyTorch 2.13.0+cu130
- Protocol
- native-forward-v1
- Dataset
- typed-invoices-v1
- Workload
- decision
Batch size Mean throughput (decisions/sec) P95 TTFT Observed occupancy (GiB) GPU power (W) Reports combined View report B1 10.56 99 ms — 18.17 206.1 2 63f44036 ↗ B4 11.04 402 ms — 18.55 244.4 2 63f44036 ↗ - bf16 · Text → Structured data · typed-invoices-v1 · Test duration 6 s
- Throughput · Batch 1
- 10.6decisions/sec
- Best measured · Batch 4
- 11.1decisions/sec
- Observed GPU occupancy
- 18.2GiB
- P95 latency
- 103 ms
Test details · 2 batch sizes
Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.
Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.
- Engine
- clef-native
- Protocol
- native-forward-v1
- Dataset
- typed-invoices-v1
- Workload
- decision
Batch size Mean throughput (decisions/sec) P95 TTFT Observed occupancy (GiB) GPU power (W) Reports combined View report B1 10.56 103 ms — 18.17 196.8 1 4eb73a4d ↗ B4 11.1 390 ms — 18.56 235.9 1 4eb73a4d ↗ - bf16 · 1K → 1K tokens
- Throughput · Batch 1
- 10.2decisions/sec
- Best measured · Batch 4
- 10.9decisions/sec
- Observed GPU occupancy
- 18.2GiB
- P95 latency
- 104 ms
Test details · 2 batch sizes
Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.
Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.
- Engine
- vLLM
- Protocol
- native-forward-v1
- Dataset
- typed-invoices-v1
- Workload
- decision
Batch size Mean throughput (decisions/sec) P95 TTFT Observed occupancy (GiB) GPU power (W) Reports combined View report B1 10.19 104 ms — 18.17 69.8 1 caef7154 ↗ B4 10.88 388 ms — 18.49 152.8 1 caef7154 ↗
Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.
Measurements without a comparable protocol · 17 rows
These measured rates remain visible, but are excluded from comparison curves because the protocol is unknown.
| Model | Format | Engine | Workload | Concurrency | Throughput | GPU power (W) |
|---|---|---|---|---|---|---|
| Qwen2.5-0.5B ↗ | Q4_K_M | Ollama | 1K | 1 | 73.58 tok/s | 59.47 |
| Qwen2.5-0.5B ↗ | Q4_K_M | Ollama | 1K | 32 | 219.91 tok/s | 89.95 |
| Qwen2.5-0.5B ↗ | Q4_K_M | Ollama | 1K | 8 | 219.04 tok/s | 90.66 |
| Qwen2.5-0.5B ↗ | Q4_K_M | Ollama | 1K | 2 | 224.87 tok/s | 90.68 |
| Qwen2.5-0.5B ↗ | Q4_K_M | Ollama | 1K | 16 | 215.94 tok/s | 91.48 |
| Qwen2.5-0.5B ↗ | Q4_K_M | Ollama | 1K | 4 | 218.84 tok/s | 92.44 |
| Qwen2.5-3B ↗ | Q4_K_M | Ollama | 1K | 1 | 128.03 tok/s | 147.28 |
| Qwen3-0.6B ↗ | BF16 | vLLM | 2K | 8 | 1,314.25 tok/s | 147.52 |
| Qwen2.5-3B ↗ | Q4_K_M | Ollama | 1K | 8 | 132.96 tok/s | 148.65 |
| Qwen2.5-3B ↗ | Q4_K_M | Ollama | 1K | 2 | 132.24 tok/s | 148.93 |
| Qwen2.5-3B ↗ | Q4_K_M | Ollama | 1K | 32 | 126.31 tok/s | 148.97 |
| Qwen2.5-3B ↗ | Q4_K_M | Ollama | 1K | 4 | 133.16 tok/s | 148.97 |
| Qwen2.5-3B ↗ | Q4_K_M | Ollama | 1K | 16 | 131.61 tok/s | 150.13 |
| Qwen3-0.6B ↗ | BF16 | vLLM | 2K | 1 | 363.8 tok/s | 163.83 |
| Qwen3-0.6B ↗ | BF16 | vLLM | 2K | 32 | 4,013.66 tok/s | 251.68 |
| Qwen3-0.6B ↗ | BF16 | vLLM | 2K | 128 | 6,085.75 tok/s | 330.76 |
| Qwen3-0.6B ↗ | BF16 | vLLM | 2K | 512 | 5,960.94 tok/s | 353.35 |
Original reports
| Model | Test task | Format | Engine | Workload | Protocol | Quality score | Test date |
|---|---|---|---|---|---|---|---|
| Qwen2.5-0.5B ↗ | Text generation | Q4_K_M | Ollama | 1K | Method unknown | — | 10/14/2026, 8:37:48 AM |
| qwen2.5-0.5b-instruct-q4_k_m.gguf ↗ | Text generation | Q4_K_M | llama.cpp | 1K | wbench-http | — | 10/14/2026, 9:07:19 AM |
| qwen2.5-1.5b-instruct-q4_k_m.gguf ↗ | Text generation | Q4_K_M | llama.cpp | 1K | wbench-http | — | 10/14/2026, 9:10:06 AM |
| Qwen2.5-3B ↗ | Text generation | Q4_K_M | Ollama | 1K | Method unknown | — | 10/14/2026, 8:44:59 AM |
| qwen2.5-3b-instruct-q4_k_m.gguf ↗ | Text generation | Q4_K_M | llama.cpp | 1K | wbench-http | — | 10/14/2026, 9:12:55 AM |
| Qwen3-0.6B ↗ | Text generation | BF16 | vLLM | 2K | Method unknown | — | 5/29/2026, 3:10:44 AM |
Running benchmarks
0Updates automaticallyNo benchmarks are currently reported as running.
Methodology
Average of error-free runs for the selected recorded configuration. Missing methods and fallback runs are not ranked. Performance index = average / fastest matching hardware × 100. Value index = performance per purchase cost / best comparable value × 100. Quality is separate by dataset and protocol. Prices are dated seller listings, not completed transactions or market averages.
Updated 11/5/2026, 11:07:29 PM