Skip to content
m.recipes
Find a configuration
Sign in
1 models with qualified results

Models

Measured configurations by model. Open a model to compare GPUs on one exact test.

Run this model

Choose your operating system and enter a model repository. Launch support is checked separately from performance results.

Reset

Text generation · All measured GPUs

Speeds compare only within one exact test. Recipe availability is checked on each report against the published recipe executor.

  1. Qwen3-4B-AWQ

    Qwen · 5 tested configurations · 3 GPUs

    • L4 · 24 GBAWQ · 8K → 2K tokens · No cache reuse
      Output throughput · includes prefill · 1 request
      55.3tok/s
      Best measured · 4 concurrent requests
      109tok/s
      Observed GPU occupancy
      19.5GiB
      First token
      1.6 s
      Use this recipe · 4 concurrent requests ↗Evidence report and run availability ↗Same test on other GPUs
      Test details · 2 concurrency levels

      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 · vLLM 0.29.0
      Protocol
      wbench-fixed-http-v3
      Dataset
      migration-analysis-v1:sha256=68590c248e43f32c04dfd8062464fd1e4856917795fe307225e5c6112a54fe94:cache=isolated:repeats=3
      Workload
      8192in-2048out
      ConcurrencyMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
      C155.337.17 s1.6 s19.5471.919301a9ba ↗
      C4109.4378.88 s3.1 s19.5571.919301a9ba ↗
    • 1 other test configurations on this GPU
      • AWQ · 8K → 2K tokens · Prefix cache reuse
        Output throughput · includes prefill · 1 request
        57.4tok/s
        Best measured · 4 concurrent requests
        175tok/s
        Observed GPU occupancy
        19.5GiB
        First token
        60 ms
        Use this recipe · 4 concurrent requests ↗Evidence report and run availability ↗Same test on other GPUs
        Test details · 2 concurrency levels

        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 · vLLM 0.29.0
        Protocol
        wbench-fixed-http-v3
        Dataset
        migration-analysis-v1:sha256=68590c248e43f32c04dfd8062464fd1e4856917795fe307225e5c6112a54fe94:cache=prefix-reuse:repeats=3
        Workload
        8192in-2048out
        ConcurrencyMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
        C157.3935.7 s60 ms19.5571.817e2cbfda ↗
        C4175.3946.75 s114 ms19.5571.917e2cbfda ↗
    • RTX PRO 6000 Blackwell Server Edition · 96 GBAWQ · 8K → 2K tokens · No cache reuse
      Output throughput · includes prefill · 1 request
      215tok/s
      Best measured · 4 concurrent requests
      517tok/s
      Observed GPU occupancy
      86.3GiB
      First token
      298 ms
      Use this recipe · 4 concurrent requests ↗Evidence report and run availability ↗Same test on other GPUs
      Test details · 2 concurrency levels

      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 · vLLM 0.29.0
      Protocol
      wbench-fixed-http-v3
      Dataset
      migration-analysis-v1:sha256=68590c248e43f32c04dfd8062464fd1e4856917795fe307225e5c6112a54fe94:cache=isolated:repeats=3
      Workload
      8192in-2048out
      ConcurrencyMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
      C1214.69.63 s298 ms86.28303.8186d36e4f ↗
      C4516.9416.14 s872 ms86.28339.5186d36e4f ↗
    • 1 other test configurations on this GPU
      • AWQ · 8K → 2K tokens · Prefix cache reuse
        Output throughput · includes prefill · 1 request
        223tok/s
        Best measured · 4 concurrent requests
        742tok/s
        Observed GPU occupancy
        86.3GiB
        First token
        28 ms
        Use this recipe · 4 concurrent requests ↗Evidence report and run availability ↗Same test on other GPUs
        Test details · 2 concurrency levels

        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 · vLLM 0.29.0
        Protocol
        wbench-fixed-http-v3
        Dataset
        migration-analysis-v1:sha256=68590c248e43f32c04dfd8062464fd1e4856917795fe307225e5c6112a54fe94:cache=prefix-reuse:repeats=3
        Workload
        8192in-2048out
        ConcurrencyMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
        C1222.749.19 s28 ms86.28288.81adbfef8a ↗
        C4741.8511.05 s35 ms86.283351adbfef8a ↗
    • T4 · 16 GBAWQ · 8K → 2K tokens · No cache reuse
      Output throughput · includes prefill · 1 request
      15.4tok/s
      Best measured · 4 concurrent requests
      20.3tok/s
      Observed GPU occupancy
      12.7GiB
      First token
      28.5 s
      Use this recipe · 4 concurrent requests ↗Evidence report and run availability ↗Same test on other GPUs
      Test details · 2 concurrency levels

      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 · vLLM 0.29.0
      Protocol
      wbench-fixed-http-v3
      Dataset
      migration-analysis-v1:sha256=68590c248e43f32c04dfd8062464fd1e4856917795fe307225e5c6112a54fe94:cache=isolated:repeats=3
      Workload
      8192in-2048out
      ConcurrencyMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
      C115.39133.36 s28.5 s12.7467.91fcfe2d64 ↗
      C420.31468.13 s53.1 s12.7467.51fcfe2d64 ↗
Sortable tables by exact test condition · 10 rows
Text generation · 8K → 2K tokens · No cache reuse · 1 request · tok/s · 3 rows
ModelHardwareFormatEngineMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
Qwen3-4B-AWQRTX PRO 6000 Blackwell Server Edition · 96 GBAWQvllmvLLM 0.29.0214.69.63 s298 ms86.28303.8186d36e4f ↗
Qwen3-4B-AWQL4 · 24 GBAWQvllmvLLM 0.29.055.337.17 s1.6 s19.5471.919301a9ba ↗
Qwen3-4B-AWQT4 · 16 GBAWQvllmvLLM 0.29.015.39133.36 s28.5 s12.7467.91fcfe2d64 ↗
Text generation · 8K → 2K tokens · No cache reuse · 4 concurrent requests · tok/s · 3 rows
ModelHardwareFormatEngineMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
Qwen3-4B-AWQRTX PRO 6000 Blackwell Server Edition · 96 GBAWQvllmvLLM 0.29.0516.9416.14 s872 ms86.28339.5186d36e4f ↗
Qwen3-4B-AWQL4 · 24 GBAWQvllmvLLM 0.29.0109.4378.88 s3.1 s19.5571.919301a9ba ↗
Qwen3-4B-AWQT4 · 16 GBAWQvllmvLLM 0.29.020.31468.13 s53.1 s12.7467.51fcfe2d64 ↗
Text generation · 8K → 2K tokens · Prefix cache reuse · 1 request · tok/s · 2 rows
ModelHardwareFormatEngineMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
Qwen3-4B-AWQRTX PRO 6000 Blackwell Server Edition · 96 GBAWQvllmvLLM 0.29.0222.749.19 s28 ms86.28288.81adbfef8a ↗
Qwen3-4B-AWQL4 · 24 GBAWQvllmvLLM 0.29.057.3935.7 s60 ms19.5571.817e2cbfda ↗
Text generation · 8K → 2K tokens · Prefix cache reuse · 4 concurrent requests · tok/s · 2 rows
ModelHardwareFormatEngineMean throughput (tok/s)P95TTFTObserved occupancy (GiB)GPU power (W)Reports combinedView report
Qwen3-4B-AWQRTX PRO 6000 Blackwell Server Edition · 96 GBAWQvllmvLLM 0.29.0741.8511.05 s35 ms86.283351adbfef8a ↗
Qwen3-4B-AWQL4 · 24 GBAWQvllmvLLM 0.29.0175.3946.75 s114 ms19.5571.917e2cbfda ↗
Models with reports but no qualified results · 7

Their original reports remain available on each model page. They did not qualify for comparison; each report states why.

Models with reports but no qualified results · 7 rows
ModelTest task
Qwen3-4BQwen/Qwen3-4BText generation
Qwen3-4B-FP8Qwen/Qwen3-4B-FP8Text generation
Qwen3-4B-Instruct-2507-FP8Qwen/Qwen3-4B-Instruct-2507-FP8Text generation
Qwen3-4B-Q4Qwen/Qwen3-4B-GGUF/Qwen3-4B-Q4_K_M.ggufText generation
Qwen3-4B-Q5_K_MQwen/Qwen3-4B-GGUF/Qwen3-4B-Q5_K_M.ggufText generation
Qwen3-4B-Q8_0Qwen/Qwen3-4B-GGUF/Qwen3-4B-Q8_0.ggufText generation
Qwen3-4B-Thinking-2507-FP8Qwen/Qwen3-4B-Thinking-2507-FP8Text generation