DGX Spark GB10

Model configurations

Each card is a model and hardware setup, including any documented host RAM, storage or CPU requirements. Select a recipe to update its speeds, context and exact command.

Search configurations

Sort byModelHardware setupTotal memory (GB)RecipeContextDecode (tok/s)Prefill (tok/s)Evidence↑ A–Z

36 configurations · 46 recipes

DeepSeek-V4-Flash

DGX Spark GB10128 GB1 card per machine · 1 machine

vLLMEXL3 3 bpw

45.5decode tok/s

625prefill tok/s

375Kcontext

Reported by MiaAI-Lab · Aug 24, 2026

CommandFull recipe ›

DeepSeek-V4-Flash

2 × DGX Spark GB10256 GB total1 card per machine · 2 machines

Exact recipevllm · FP8 · 1024K contextvllm · 320K contextvllm · 977K context

56decode tok/s

—prefill tok/s

1024Kcontext

Reported by MiaAI-Lab · Sep 16, 2026

CommandFull recipe ›

DeepSeek-V4-Flash

3 × DGX Spark GB10384 GB total1 card per machine · 3 machines

vLLMFP8

41.9decode tok/s

—prefill tok/s

1024Kcontext

Reported by MiaAI-Lab · Sep 16, 2026

CommandFull recipe ›

DeepSeek-V4.1-Flash

2 × DGX Spark GB10256 GB total1 card per machine · 2 machines

Exact recipevllm · 256K contextvllm · EXL3 2.9 bpw · 586K context

22.5decode tok/s

1,939prefill tok/s

256Kcontext

Tested on this card · Oct 3, 2026

CommandFull recipe ›

DeepSeek-V4.1-Flash

3 × DGX Spark GB10384 GB total1 card per machine · 3 machines

SGLangFormat not recorded

51decode tok/s

2,000prefill tok/s

256Kcontext

Reported by MiaAI-Lab · Sep 25, 2026

CommandFull recipe ›

DeepSeek-V4.1-Flash

4 × DGX Spark GB10512 GB total1 card per machine · 4 machines

SGLangFormat not recorded

87.7decode tok/s

5,925prefill tok/s

1024Kcontext

Reported by MiaAI-Lab · Sep 25, 2026

CommandFull recipe ›

DiffusionGemma 26B A4B

DGX Spark GB10128 GB1 card per machine · 1 machine

vLLMFormat not recorded

55.8decode tok/s

463prefill tok/s

256Kcontext

Reported by toncao · Sep 30, 2026

CommandFull recipe ›

diffusiongemma-26B-A4B-it-AWQ-INT4

DGX Spark GB10128 GB1 card per machine · 1 machine

vLLMFormat not recorded

101.4decode tok/s

462prefill tok/s

256Kcontext

Reported by toncao · Sep 30, 2026

CommandFull recipe ›

Gemma 4 26B A4B

DGX Spark GB10128 GB1 card per machine · 1 machine

vLLMNVFP4

29.8decode tok/s

—prefill tok/s

256Kcontext

Reported by MiaAI-Lab · Jul 5, 2026

CommandFull recipe ›

Gemma 4 31B

DGX Spark GB10128 GB1 card per machine · 1 machine

vLLMNVFP4

—decode tok/s

—prefill tok/s

256Kcontext

Reported by MiaAI-Lab · Jul 16, 2026

CommandFull recipe ›

gemma-4-12B-it-AWQ-INT4

DGX Spark GB10128 GB1 card per machine · 1 machine

vLLMFormat not recorded

43.1decode tok/s

843prefill tok/s

128Kcontext

Reported by toncao · Sep 30, 2026

CommandFull recipe ›

gemma-4-26B-A4B-it-AWQ-4bit

DGX Spark GB10128 GB1 card per machine · 1 machine

vLLMFormat not recorded

63.2decode tok/s

412prefill tok/s

256Kcontext

Reported by toncao · Sep 30, 2026

CommandFull recipe ›

gemma-4-26B-A4B-it-AWQ-8bit

DGX Spark GB10128 GB1 card per machine · 1 machine

vLLMFormat not recorded

49decode tok/s

401prefill tok/s

256Kcontext

Reported by toncao · Sep 30, 2026

CommandFull recipe ›

gemma-4-31B-it-AWQ-4bit

DGX Spark GB10128 GB1 card per machine · 1 machine

vLLMFormat not recorded

22.2decode tok/s

98prefill tok/s

256Kcontext

Reported by toncao · Sep 30, 2026

CommandFull recipe ›

GLM-5.2

3 × DGX Spark GB10384 GB total1 card per machine · 3 machines

Exact recipevllm · NVFP4 · 230K contextvllm · NVFP4 · 340K contextvllm · NVFP4 · 372K context

25.5decode tok/s

—prefill tok/s

230Kcontext

Reported by MiaAI-Lab · Aug 3, 2026

CommandFull recipe ›

GLM-5.3-Flash

2 × DGX Spark GB10256 GB total1 card per machine · 2 machines

Exact recipesglang · NVFP4 · 256K contextvllm · NVFP4 · 256K context

30.7decode tok/s

—prefill tok/s

256Kcontext

Earlier check · Date not recorded

CommandFull recipe ›

GLM-5.3-Flash

4 × DGX Spark GB10512 GB total1 card per machine · 4 machines

Exact recipesglang · NVFP4 · 1024K contexttensorfold · EXL3 4 bpw · 1024K contextvllm · NVFP4 · 256K context

59.1decode tok/s

—prefill tok/s

1024Kcontext

Earlier check · Date not recorded

CommandFull recipe ›

Hy3-295B

2 × DGX Spark GB10256 GB total1 card per machine · 2 machines

vLLMNVFP4

—decode tok/s

—prefill tok/s

128Kcontext

Reported by MiaAI-Lab · Jul 7, 2026

CommandFull recipe ›

Inkling-Small

2 × DGX Spark GB10256 GB total1 card per machine · 2 machines

SGLangNVFP4

33.9decode tok/s

—prefill tok/s

1024Kcontext

Reported by MiaAI-Lab · Aug 2, 2026

CommandFull recipe ›

Laguna-S-2.1

DGX Spark GB10128 GB1 card per machine · 1 machine

vLLMNVFP4

76.7decode tok/s

—prefill tok/s

256Kcontext

Reported by MiaAI-Lab · Jul 22, 2026

CommandFull recipe ›

Leanstral-1.5-119B-A6B

2 × DGX Spark GB10256 GB total1 card per machine · 2 machines

vLLMFP8

21.1decode tok/s

—prefill tok/s

256Kcontext

Reported by MiaAI-Lab · Jul 4, 2026

CommandFull recipe ›

Ling-3.0-Flash

DGX Spark GB10128 GB1 card per machine · 1 machine

SGLangFormat not recorded

83decode tok/s

—prefill tok/s

256Kcontext

Reported by MiaAI-Lab · Aug 25, 2026

CommandFull recipe ›

MiMo-V2.5

2 × DGX Spark GB10256 GB total1 card per machine · 2 machines

vLLMNVFP4

30decode tok/s

—prefill tok/s

977Kcontext

Reported by MiaAI-Lab · Jul 15, 2026

CommandFull recipe ›

MiMo-V2.6-Flash

2 × DGX Spark GB10256 GB total1 card per machine · 2 machines

SGLangFormat not recorded

69.8decode tok/s

—prefill tok/s

1024Kcontext

Reported by MiaAI-Lab · Sep 22, 2026

CommandFull recipe ›

Muse Glimmer 30B

DGX Spark GB10128 GB1 card per machine · 1 machine

vLLMNVFP4

35decode tok/s

—prefill tok/s

256Kcontext

Reported by MiaAI-Lab · Aug 11, 2026

CommandFull recipe ›

Nemotron 3.5 Lightning 30B-A3B

DGX Spark GB10128 GB1 card per machine · 1 machine

SGLangNVFP4

—decode tok/s

—prefill tok/s

1024Kcontext

Reported by MiaAI-Lab · Aug 11, 2026

CommandFull recipe ›

Nemotron-Labs-3-Puzzle-75B-A9B

DGX Spark GB10128 GB1 card per machine · 1 machine

vLLMNVFP4

—decode tok/s

—prefill tok/s

256Kcontext

Reported by MiaAI-Lab · Jul 8, 2026

CommandFull recipe ›

Ornith-1.5-35B-A3B

DGX Spark GB10128 GB1 card per machine · 1 machine

vLLMNVFP4

86.3decode tok/s

—prefill tok/s

256Kcontext

Reported by MiaAI-Lab · Aug 19, 2026

CommandFull recipe ›

Qwen3.6-27B

DGX Spark GB10128 GB1 card per machine · 1 machine

vLLMNVFP4

45.6decode tok/s

1,083prefill tok/s

256Kcontext

Reported by MiaAI-Lab · Jul 31, 2026

CommandFull recipe ›

Qwen3.6-35B-A3B

DGX Spark GB10128 GB1 card per machine · 1 machine

vLLMNVFP4

95.1decode tok/s

—prefill tok/s

256Kcontext

Reported by MiaAI-Lab · Jul 29, 2026

CommandFull recipe ›

Qwen3.8-27B

DGX Spark GB10128 GB1 card per machine · 1 machine

Exact recipeexllamav3 · EXL3 3.5 bpw · 256K contextsglang · NVFP4 · 256K contextvllm · NVFP4 · 977K context

47.5decode tok/s

—prefill tok/s

256Kcontext

Reported by MiaAI-Lab · Sep 1, 2026

CommandFull recipe ›

Qwen3.8-27B-AWQ-BF16-INT4

DGX Spark GB10128 GB1 card per machine · 1 machine

vLLMFormat not recorded

34.5decode tok/s

582prefill tok/s

256Kcontext

Reported by toncao · Sep 30, 2026

CommandFull recipe ›

Qwen3.8-27B-AWQ-FP8

DGX Spark GB10128 GB1 card per machine · 1 machine

vLLMFP8

29.7decode tok/s

562prefill tok/s

256Kcontext

Reported by toncao · Sep 30, 2026

CommandFull recipe ›

Qwen3.8-Flash-Next

2 × DGX Spark GB10256 GB total1 card per machine · 2 machines

vLLMNVFP4

54.4decode tok/s

2,962.2prefill tok/s

256Kcontext

Reported by MiaAI-Lab · Sep 16, 2026

CommandFull recipe ›

Step-3.7-Flash

2 × DGX Spark GB10256 GB total1 card per machine · 2 machines

vLLMNVFP4

31.5decode tok/s

—prefill tok/s

256Kcontext

Reported by MiaAI-Lab · Jun 20, 2026

CommandFull recipe ›

Step-5-Preview

4 × DGX Spark GB10512 GB total1 card per machine · 4 machines

vLLMFormat not recorded

21.3decode tok/s

888prefill tok/s

256Kcontext

Tested on this card · Oct 6, 2026

CommandFull recipe ›

Speeds are the selected recipe’s recorded measurements, not estimates; “—” means unmeasured. Evidence distinguishes checks on this card, sibling-card checks, earlier checks and publisher reports.

Sources: NVIDIA, NVIDIA (2)