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 16K context with f16 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 16K 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 GiB0.47 GiB119.01 GiB0.03 GiB12±37%
Mistral-Small-4-119B-2603MoEQ8_0119B117.79 GiB0.35 GiB118.97 GiB0.07 GiB10±37%
DeepSeek-Coder-V2-Instruct-0724MoEIQ4_XS236B116.94 GiB1.05 GiB118.83 GiB0.21 GiB9±37%
DeepSeek-V2.5MoEIQ4_XS236B116.94 GiB1.05 GiB118.83 GiB0.21 GiB9±37%
DeepSeek-Coder-V2-InstructMoEIQ4_XS236B116.94 GiB1.05 GiB118.83 GiB0.21 GiB9±37%
MiniMax-M2.7MoEIQ4_XS229B114.00 GiB3.88 GiB118.66 GiB0.38 GiB8±37%
command-a-plus-05-2026-bf16MoEIQ4_NL219B116.79 GiB0.92 GiB118.52 GiB0.52 GiB8±37%
MiniMax-M2.1MoEIQ4_XS229B113.78 GiB3.88 GiB118.44 GiB0.60 GiB8±37%
MiniMax-M2MoEIQ4_XS229B113.78 GiB3.88 GiB118.44 GiB0.60 GiB8±37%
step-3.5-flashQ4_K199B110.56 GiB7.03 GiB118.42 GiB0.62 GiB2±12.9%
MiniMax-M2.5MoEIQ4_XS229B113.53 GiB3.88 GiB118.19 GiB0.85 GiB8±37%
dots.llm1.instMoEQ5_K_M143B101.84 GiB15.50 GiB118.17 GiB0.87 GiB4±37%
Qwen3.5-REAP-262B-A17BMoEQ3_K_M262B116.42 GiB0.47 GiB117.74 GiB1.30 GiB11±37%
Qwen3-235B-A22B-abliteratedMoEI1-Q3_K_L235B113.46 GiB2.94 GiB117.23 GiB1.81 GiB7±37%
Ornith-1.0-397BMoEUD-IQ2_M397B115.83 GiB0.47 GiB117.15 GiB1.89 GiB12±37%
gpt-oss-120b-abliteratedMoEQ8_0117B115.76 GiB0.59 GiB117.14 GiB1.90 GiB10±37%
Step-3.7-FlashQ4_K_S201B109.05 GiB7.03 GiB116.91 GiB2.13 GiB2±12.9%
GLM-4.7-REAP-218B-A32BMoEIQ4_XS218B110.09 GiB5.75 GiB116.68 GiB2.36 GiB5±37%
Llama-4-Maverick-17B-128E-InstructMoEKV unresolvedUD-IQ1_S402B112.48 GiB3.00 GiB116.30 GiB2.74 GiB11±37%
MiMo-V2-FlashMoEKV unresolvedIQ3_XXS310B113.14 GiB1.88 GiB115.86 GiB3.18 GiB9±37%
Solar-Open2-250BMoEQ3_K_M250B111.63 GiB3.00 GiB115.46 GiB3.58 GiB9±37%
GLM-4.6-REAP-268B-A32BMoEQ3_K_S269B108.47 GiB5.75 GiB115.06 GiB3.98 GiB6±37%
MiniMax-M2.7-BF16-ultra-uncensored-hereticMoEQ3_K_L229B110.22 GiB3.88 GiB114.88 GiB4.16 GiB8±37%
grok-2MoEQ3_K_S270B109.94 GiB4.00 GiB114.88 GiB4.16 GiB3±37%
Trinity-Large-ThinkingMoEIQ2_S399B112.03 GiB1.73 GiB114.59 GiB4.45 GiB11±37%
GLM-4.5MoEUD-IQ2_XXS358B107.95 GiB5.75 GiB114.54 GiB4.50 GiB7±37%
GLM-4.7MoEUD-IQ2_XXS358B107.95 GiB5.75 GiB114.54 GiB4.50 GiB7±37%
GLM-4.6MoEUD-IQ2_XXS357B107.48 GiB5.75 GiB114.07 GiB4.97 GiB7±37%
GLM-4.6-Derestricted-v3MoEIQ2_M357B107.14 GiB5.75 GiB113.73 GiB5.31 GiB7±37%
Nex-N2-ProMoEIQ2_S397B112.23 GiB0.47 GiB113.55 GiB5.49 GiB12±37%
GLM-4.5-AirMoEQ8_0110B109.39 GiB2.88 GiB113.09 GiB5.95 GiB7±37%
GLM-4.5-Air-DerestrictedMoEQ8_0110B109.39 GiB2.88 GiB113.09 GiB5.95 GiB7±37%
Qwen3-Coder-REAP-363B-A35BMoEUD-IQ1_M363B107.85 GiB3.88 GiB112.56 GiB6.48 GiB7±37%
Mixtral-8x22B-v0.1MoEQ6_K141B107.60 GiB3.50 GiB111.96 GiB7.08 GiB3±37%
MiniMax-M3MoEIQ2_XXS427B108.60 GiB1.88 GiB111.29 GiB7.75 GiB10±37%
MiniMax-M2.1-REAP-139B-A10BMoEI1-Q6_K139B106.40 GiB3.88 GiB111.06 GiB7.98 GiB7±37%
m51Lab-MiniMax-M2.7-REAP-139B-A10BMoEQ6_K139B106.40 GiB3.88 GiB111.06 GiB7.98 GiB7±37%
Trinity-Large-TrueBaseMoEI1-IQ2_S399B108.32 GiB1.73 GiB110.88 GiB8.16 GiB12±37%
Llama-4-Scout-17B-16E-InstructMoEKV unresolvedQ8_0109B106.67 GiB3.00 GiB110.49 GiB8.55 GiB7±37%
DeepSeek-V4-FlashMoEUD-IQ3_S291B109.25 GiB0.06 GiB110.16 GiB8.88 GiB13±37%
MiMo-V2.5MoEKV unresolvedUD-IQ3_S311B106.98 GiB1.88 GiB109.70 GiB9.34 GiB10±37%
GLM-4.6VMoEQ8_0108B105.81 GiB2.88 GiB109.51 GiB9.53 GiB7±37%
NVIDIA-Nemotron-3-Super-120B-A12B-BF16MoEUD-Q6_K124B106.87 GiB1.38 GiB109.04 GiB10.00 GiB9±37%
DeepSeek-V4-Flash-0731MoEUD-IQ3_S304B108.10 GiB0.06 GiB109.01 GiB10.03 GiB13±37%
Hermes-4-405BIQ2_XXS406B99.91 GiB7.88 GiB108.86 GiB10.18 GiB2±12.9%
Hermes-3-Llama-3.1-405BIQ2_XXS406B99.91 GiB7.88 GiB108.86 GiB10.18 GiB2±12.9%
Qwen3-VL-235B-A22B-ThinkingMoEQ3_K_M236B104.72 GiB2.94 GiB108.49 GiB10.55 GiB7±37%
Qwen3-VL-235B-A22B-InstructMoEQ3_K_M236B104.72 GiB2.94 GiB108.49 GiB10.55 GiB7±37%
Qwen3-235B-A22BMoEQ3_K_M235B104.72 GiB2.94 GiB108.49 GiB10.55 GiB7±37%
Qwen3-235B-A22B-Thinking-2507MoEQ3_K_M235B104.72 GiB2.94 GiB108.49 GiB10.55 GiB7±37%
Qwen3-235B-A22B-Instruct-2507MoEQ3_K_M235B104.72 GiB2.94 GiB108.49 GiB10.55 GiB7±37%
c4ai-command-r-plus-08-2024Q8_0104B102.74 GiB4.00 GiB107.72 GiB11.32 GiB2±12.9%
Mistral-Medium-3.5-128BQ6_K_L128B101.13 GiB5.50 GiB107.58 GiB11.46 GiB2±12.9%
Gemma-4-Dark-Gemistry-31BQ8_032.7B102.98 GiB3.67 GiB107.53 GiB11.51 GiB2±12.9%
Hy3MoEQ2_K299B101.28 GiB5.00 GiB107.12 GiB11.92 GiB8±37%
Qwen3.5-REAP-212B-A17BMoEIQ4_XS212B105.39 GiB0.47 GiB106.71 GiB12.33 GiB11±37%
ERNIE-4.5-300B-A47B-PTQ2_K_L300B101.76 GiB3.38 GiB106.06 GiB12.98 GiB2±12.9%
gemma-2-27b-itF3227.2B101.43 GiB3.68 GiB106.05 GiB12.99 GiB2±12.9%
Trinity-Large-PreviewMoEIQ2_S399B102.49 GiB1.73 GiB105.05 GiB13.99 GiB12±37%
gpt-oss-20b-hereticMoEQ8_020.9B102.72 GiB0.39 GiB103.90 GiB15.14 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 16,384 context with f16 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.
NVIDIA DGX Spark (GB10) — what AI models can it run locally? — ossmodeldb