NVIDIA · unified x86

NVIDIA DGX Spark (GB10)

NVIDIA DGX Spark (GB10) has 128 GB of unified memory at 273 GB/s — about 119.04 GiB usable after driver and compositor overhead. 2091 of 2118 indexed models fit at 32K context with q8_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
128 GB
LPDDR5X-8533
Bandwidth
273 GB/s
256-bit bus
Tensor FP16
125 TF
dense
TDP
140 W
$3999 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
vision language 191text 1796audio tts 21image 2audio asr 39video 16embedding 26

What fits at 32K context

largest quantization that fits, per model · 2091 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Qwen3.5-397B-A17BMoEUD-IQ2_XXS403B117.69 GiB0.50 GiB119.04 GiB0.00 GiB12±37%
Mistral-Small-4-119B-2603MoEQ8_0119B117.79 GiB0.37 GiB118.99 GiB0.05 GiB10±37%
command-a-plus-05-2026-bf16MoEQ4_0219B117.42 GiB0.76 GiB118.98 GiB0.06 GiB8±37%
DeepSeek-Coder-V2-Instruct-0724MoEIQ4_XS236B116.94 GiB1.12 GiB118.90 GiB0.14 GiB9±37%
DeepSeek-V2.5MoEIQ4_XS236B116.94 GiB1.12 GiB118.90 GiB0.14 GiB9±37%
DeepSeek-Coder-V2-InstructMoEIQ4_XS236B116.94 GiB1.12 GiB118.90 GiB0.14 GiB9±37%
MiniMax-M2.7MoEIQ4_XS229B114.00 GiB4.12 GiB118.90 GiB0.14 GiB8±37%
Trinity-Large-PreviewMoEIQ2_M399B116.50 GiB1.42 GiB118.75 GiB0.29 GiB11±37%
Trinity-Large-TrueBaseMoEIQ2_M399B116.50 GiB1.42 GiB118.75 GiB0.29 GiB11±37%
MiniMax-M2.1MoEIQ4_XS229B113.78 GiB4.12 GiB118.68 GiB0.36 GiB8±37%
MiniMax-M2MoEIQ4_XS229B113.78 GiB4.12 GiB118.68 GiB0.36 GiB8±37%
MiniMax-M2.5MoEIQ4_XS229B113.53 GiB4.12 GiB118.43 GiB0.61 GiB8±37%
step-3.5-flashQ4_K199B110.56 GiB6.92 GiB118.31 GiB0.73 GiB2±12.9%
Qwen3.5-REAP-262B-A17BMoEQ3_K_M262B116.42 GiB0.50 GiB117.77 GiB1.27 GiB11±37%
Qwen3-235B-A22B-abliteratedMoEI1-Q3_K_L235B113.46 GiB3.12 GiB117.41 GiB1.63 GiB7±37%
Ornith-1.0-397BMoEUD-IQ2_M397B115.83 GiB0.50 GiB117.18 GiB1.86 GiB12±37%
gpt-oss-120b-abliteratedMoEQ8_0117B115.76 GiB0.61 GiB117.17 GiB1.87 GiB10±37%
GLM-4.7-REAP-218B-A32BMoEIQ4_XS218B110.09 GiB6.11 GiB117.04 GiB2.00 GiB5±37%
Step-3.7-FlashQ4_K_S201B109.05 GiB6.92 GiB116.80 GiB2.24 GiB2±12.9%
Llama-4-Maverick-17B-128E-InstructMoEKV unresolvedUD-IQ1_S402B112.48 GiB3.19 GiB116.49 GiB2.55 GiB11±37%
MiMo-V2-FlashMoEKV unresolvedIQ3_XXS310B113.14 GiB1.99 GiB115.98 GiB3.06 GiB9±37%
Solar-Open2-250BMoEQ3_K_M250B111.63 GiB3.19 GiB115.65 GiB3.39 GiB9±37%
GLM-4.6-REAP-268B-A32BMoEQ3_K_S269B108.47 GiB6.11 GiB115.42 GiB3.62 GiB6±37%
grok-2MoEQ3_K_S270B109.94 GiB4.25 GiB115.13 GiB3.91 GiB3±37%
MiniMax-M2.7-BF16-ultra-uncensored-hereticMoEQ3_K_L229B110.22 GiB4.12 GiB115.13 GiB3.91 GiB8±37%
GLM-4.5MoEUD-IQ2_XXS358B107.95 GiB6.11 GiB114.90 GiB4.14 GiB6±37%
GLM-4.7MoEUD-IQ2_XXS358B107.95 GiB6.11 GiB114.90 GiB4.14 GiB6±37%
GLM-4.6MoEUD-IQ2_XXS357B107.48 GiB6.11 GiB114.43 GiB4.61 GiB6±37%
Trinity-Large-ThinkingMoEIQ2_S399B112.03 GiB1.42 GiB114.27 GiB4.77 GiB12±37%
GLM-4.6-Derestricted-v3MoEIQ2_M357B107.14 GiB6.11 GiB114.09 GiB4.95 GiB6±37%
Nex-N2-ProMoEIQ2_S397B112.23 GiB0.50 GiB113.58 GiB5.46 GiB12±37%
GLM-4.5-AirMoEQ8_0110B109.39 GiB3.05 GiB113.27 GiB5.77 GiB7±37%
GLM-4.5-Air-DerestrictedMoEQ8_0110B109.39 GiB3.05 GiB113.27 GiB5.77 GiB7±37%
Qwen3-Coder-REAP-363B-A35BMoEUD-IQ1_M363B107.85 GiB4.12 GiB112.80 GiB6.24 GiB7±37%
dots.llm1.instMoEQ5_K_S143B95.08 GiB16.47 GiB112.38 GiB6.66 GiB4±37%
Mixtral-8x22B-v0.1MoEQ6_K141B107.60 GiB3.72 GiB112.17 GiB6.87 GiB3±37%
MiniMax-M3MoEIQ2_XXS427B108.60 GiB1.99 GiB111.41 GiB7.63 GiB10±37%
MiniMax-M2.1-REAP-139B-A10BMoEI1-Q6_K139B106.40 GiB4.12 GiB111.31 GiB7.73 GiB7±37%
m51Lab-MiniMax-M2.7-REAP-139B-A10BMoEQ6_K139B106.40 GiB4.12 GiB111.31 GiB7.73 GiB7±37%
Llama-4-Scout-17B-16E-InstructMoEKV unresolvedQ8_0109B106.67 GiB3.19 GiB110.68 GiB8.36 GiB7±37%
DeepSeek-V4-FlashMoEUD-IQ3_S291B109.25 GiB0.03 GiB110.14 GiB8.90 GiB13±37%
MiMo-V2.5MoEKV unresolvedUD-IQ3_S311B106.98 GiB1.99 GiB109.82 GiB9.22 GiB10±37%
GLM-4.6VMoEQ8_0108B105.81 GiB3.05 GiB109.69 GiB9.35 GiB7±37%
Hermes-4-405BIQ2_XXS406B99.91 GiB8.37 GiB109.35 GiB9.69 GiB2±12.9%
Hermes-3-Llama-3.1-405BIQ2_XXS406B99.91 GiB8.37 GiB109.35 GiB9.69 GiB2±12.9%
NVIDIA-Nemotron-3-Super-120B-A12B-BF16MoEUD-Q6_K124B106.87 GiB1.46 GiB109.13 GiB9.91 GiB9±37%
DeepSeek-V4-Flash-0731MoEUD-IQ3_S304B108.10 GiB0.03 GiB108.98 GiB10.06 GiB13±37%
Qwen3-VL-235B-A22B-ThinkingMoEQ3_K_M236B104.72 GiB3.12 GiB108.68 GiB10.36 GiB7±37%
Qwen3-VL-235B-A22B-InstructMoEQ3_K_M236B104.72 GiB3.12 GiB108.68 GiB10.36 GiB7±37%
Qwen3-235B-A22BMoEQ3_K_M235B104.72 GiB3.12 GiB108.68 GiB10.36 GiB7±37%
Qwen3-235B-A22B-Thinking-2507MoEQ3_K_M235B104.72 GiB3.12 GiB108.68 GiB10.36 GiB7±37%
Qwen3-235B-A22B-Instruct-2507MoEQ3_K_M235B104.72 GiB3.12 GiB108.68 GiB10.36 GiB7±37%
c4ai-command-r-plus-08-2024Q8_0104B102.74 GiB4.25 GiB107.97 GiB11.07 GiB2±12.9%
Mistral-Medium-3.5-128BQ6_K_L128B101.13 GiB5.84 GiB107.93 GiB11.11 GiB2±12.9%
Hy3MoEQ2_K299B101.28 GiB5.31 GiB107.43 GiB11.61 GiB7±37%
Gemma-4-Dark-Gemistry-31BQ8_032.7B102.98 GiB3.28 GiB107.14 GiB11.90 GiB2±12.9%
Qwen3.5-REAP-212B-A17BMoEIQ4_XS212B105.39 GiB0.50 GiB106.74 GiB12.30 GiB11±37%
ERNIE-4.5-300B-A47B-PTQ2_K_L300B101.76 GiB3.59 GiB106.27 GiB12.77 GiB2±12.9%
gemma-2-27b-itF3227.2B101.43 GiB3.48 GiB105.85 GiB13.19 GiB2±12.9%
gpt-oss-20b-hereticMoEQ8_020.9B102.72 GiB0.41 GiB103.92 GiB15.12 GiB6±37%
●From the file◐Predictedwhat these mean

Speed is modeled, not measured: decode is memory-bandwidth bound, so tokens per second is bytes read per token against achievable bandwidth. Mixture-of-experts models carry a wider band because only the routed experts are read each step, and few have been measured publicly.

Questions people ask

What AI models can a NVIDIA DGX Spark (GB10) run?
2091 of 2118 indexed open-weight models fit a NVIDIA DGX Spark (GB10) at 32,768 context with q8_0 KV cache, the largest being Qwen3.5-397B-A17B at UD-IQ2_XXS. That covers text, vision-language, image, video and speech models.
How much usable memory does a NVIDIA DGX Spark (GB10) actually have?
Its nameplate is 128 GB, but about 119.04 GiB is available to a model once driver and compositor overhead is accounted for.
Is a NVIDIA DGX Spark (GB10) fast for local AI?
Its memory bandwidth is 273 GB/s, and that figure — not teraflops — is what governs token generation speed. Capacity decides what you can run; bandwidth decides how fast it runs.