Spark Bench

Top 5 models by benchmark

Each ranking uses the best measured configuration per model family. Spark count and serving runtime can vary across entries.

Decode scaling

Highest aggregate throughput at any measured concurrency using ~256 input / 512 output tokens per request. Higher concurrency can include queueing.

View decode scaling chart
  1. Qwen3.6-35B-A3B NVFP4 (Unsloth)

    1× DGX Spark · vLLM

    Tool eval completeSafety gate failed

    429.21 output tok/s

    Concurrency 16

  2. Qwen3.8-Flash-Next NVFP4

    2× DGX Spark · TensorFold

    Tool eval incompleteSafety gate passed

    417.63 output tok/s

    Concurrency 16

  3. NVIDIA Nemotron 3.5 Bolt 74B-A9B

    1× DGX Spark · TensorFold

    Tool eval completeSafety gate failed

    234.45 output tok/s

    Concurrency 16

  4. Thinking Machines Inkling 2 Mini

    1× DGX Spark · vLLM + DSpark7

    Tool eval completeSafety gate failed

    229.47 output tok/s

    Concurrency 16

  5. Qwen3.8 27B

    1× DGX Spark · TensorFold

    Tool eval incompleteSafety gate failed

    201.49 output tok/s

    Concurrency 16

Long-context prefill

Lowest p95 time to first token on the 245,760-token passkey workload at concurrency 1.

View long-context prefill chart
  1. Qwen3.8-Flash-Next NVFP4

    2× DGX Spark · TensorFold

    Tool eval incompleteSafety gate passed

    98.98 s

  2. DeepSeek V4.1 Flash 1017 EXL3-EAQ

    2× DGX Spark · TensorFold

    Tool eval incompleteSafety gate passed

    120.40 s

  3. NVIDIA Nemotron 3.5 Super VL 120B-A12B

    1× DGX Spark · vLLM

    Tool eval completeSafety gate failed

    155.41 s

  4. Thinking Machines Inkling 2 Mini

    1× DGX Spark · vLLM + DSpark7

    Tool eval completeSafety gate failed

    155.99 s

  5. GLM-5.3 Flash

    2× DGX Spark · TensorFold

    Tool eval unavailableSafety gate unavailable

    163.89 s

Sustained load

Highest aggregate throughput in the 20-minute test at concurrency 8, with zero request errors.

View sustained load chart
  1. Qwen3.8-Flash-Next NVFP4

    2× DGX Spark · TensorFold

    Tool eval incompleteSafety gate passed

    432.90 output tok/s

  2. Qwen3.6-35B-A3B NVFP4 (Unsloth)

    1× DGX Spark · vLLM

    Tool eval completeSafety gate failed

    286.66 output tok/s

  3. Qwen3.8 27B

    1× DGX Spark · TensorFold

    Tool eval incompleteSafety gate failed

    160.59 output tok/s

  4. NVIDIA Nemotron 3.5 Bolt 74B-A9B

    1× DGX Spark · TensorFold

    Tool eval completeSafety gate failed

    145.52 output tok/s

  5. Thinking Machines Inkling 2 Mini

    1× DGX Spark · vLLM + DSpark7

    Tool eval completeSafety gate failed

    145.04 output tok/s

Install model

Copy this prompt and paste it into an AI assistant with access to your DGX Spark.