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 8K 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
text 1796vision language 191audio tts 21image 2audio asr 39video 16embedding 26

What fits at 8K context

largest quantization that fits, per model · 2091 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
MiMo-V2-FlashMoEKV unresolvedIQ3_XS310B117.68 GiB0.50 GiB119.03 GiB0.01 GiB10±37%
Qwen3-235B-A22B-Instruct-2507MoEIQ4_XS235B117.24 GiB0.78 GiB118.85 GiB0.19 GiB8±37%
Qwen3-235B-A22B-Thinking-2507MoEIQ4_XS235B117.24 GiB0.78 GiB118.85 GiB0.19 GiB8±37%
Qwen3.5-397B-A17BMoEUD-IQ2_M403B117.81 GiB0.12 GiB118.78 GiB0.26 GiB12±37%
Mistral-Small-4-119B-2603MoEQ8_0119B117.79 GiB0.09 GiB118.71 GiB0.33 GiB11±37%
MiMo-V2.5MoEKV unresolvedUD-IQ3_XXS311B117.27 GiB0.50 GiB118.62 GiB0.42 GiB10±37%
command-a-plus-05-2026-bf16MoEQ4_0219B117.42 GiB0.36 GiB118.58 GiB0.46 GiB8±37%
GLM-4.7-REAP-218B-A32BMoEIQ4_NL218B116.20 GiB1.53 GiB118.56 GiB0.48 GiB6±37%
Qwen3-235B-A22BMoEIQ4_XS235B116.89 GiB0.78 GiB118.50 GiB0.54 GiB8±37%
ERNIE-4.5-300B-A47B-PTUD-IQ3_XXS300B116.56 GiB0.90 GiB118.38 GiB0.66 GiB2±12.9%
Qwen3-VL-235B-A22B-ThinkingMoEIQ4_XS236B116.70 GiB0.78 GiB118.31 GiB0.73 GiB8±37%
Qwen3-VL-235B-A22B-InstructMoEIQ4_XS236B116.70 GiB0.78 GiB118.31 GiB0.73 GiB8±37%
Qwen3-235B-A22B-abliteratedMoEI1-IQ4_XS235B116.68 GiB0.78 GiB118.29 GiB0.75 GiB8±37%
step-3.5-flashQ4_1199B115.15 GiB2.14 GiB118.12 GiB0.92 GiB2±12.9%
DeepSeek-Coder-V2-Instruct-0724MoEIQ4_XS236B116.94 GiB0.28 GiB118.06 GiB0.98 GiB10±37%
DeepSeek-V2.5MoEIQ4_XS236B116.94 GiB0.28 GiB118.06 GiB0.98 GiB10±37%
DeepSeek-Coder-V2-InstructMoEIQ4_XS236B116.94 GiB0.28 GiB118.06 GiB0.98 GiB10±37%
Trinity-Large-PreviewMoEIQ2_M399B116.50 GiB0.67 GiB118.00 GiB1.04 GiB12±37%
Trinity-Large-TrueBaseMoEIQ2_M399B116.50 GiB0.67 GiB118.00 GiB1.04 GiB12±37%
Qwen3.5-REAP-262B-A17BMoEQ3_K_M262B116.42 GiB0.12 GiB117.39 GiB1.65 GiB11±37%
grok-2MoEIQ3_M270B115.25 GiB1.06 GiB117.25 GiB1.79 GiB3±37%
Ornith-1.0-397BMoEUD-IQ2_M397B115.83 GiB0.12 GiB116.81 GiB2.23 GiB12±37%
gpt-oss-120b-abliteratedMoEQ8_0117B115.76 GiB0.16 GiB116.72 GiB2.32 GiB11±37%
Step-3.7-FlashUD-Q4_K_M201B113.71 GiB2.14 GiB116.68 GiB2.36 GiB2±12.9%
GLM-4.5MoEUD-IQ2_M358B114.03 GiB1.53 GiB116.40 GiB2.64 GiB8±37%
GLM-4.7MoEUD-IQ2_M358B114.03 GiB1.53 GiB116.40 GiB2.64 GiB8±37%
GLM-4.6MoEUD-IQ2_M357B113.56 GiB1.53 GiB115.92 GiB3.12 GiB8±37%
MiniMax-M2.7MoEIQ4_XS229B114.00 GiB1.03 GiB115.81 GiB3.23 GiB10±37%
MiniMax-M2.1MoEIQ4_XS229B113.78 GiB1.03 GiB115.59 GiB3.45 GiB10±37%
MiniMax-M2MoEIQ4_XS229B113.78 GiB1.03 GiB115.59 GiB3.45 GiB10±37%
MiniMax-M2.5MoEIQ4_XS229B113.53 GiB1.03 GiB115.34 GiB3.70 GiB10±37%
Hermes-4-405BIQ2_XS406B111.15 GiB2.09 GiB114.33 GiB4.71 GiB2±12.9%
Hermes-3-Llama-3.1-405BIQ2_XS406B111.15 GiB2.09 GiB114.33 GiB4.71 GiB2±12.9%
Llama-4-Maverick-17B-128E-InstructMoEKV unresolvedUD-IQ1_S402B112.48 GiB0.80 GiB114.10 GiB4.94 GiB14±37%
Trinity-Large-ThinkingMoEIQ2_S399B112.03 GiB0.67 GiB113.53 GiB5.51 GiB13±37%
Solar-Open2-250BMoEQ3_K_M250B111.63 GiB0.80 GiB113.26 GiB5.78 GiB11±37%
Nex-N2-ProMoEIQ2_S397B112.23 GiB0.12 GiB113.21 GiB5.83 GiB13±37%
MiniMax-M2.7-BF16-ultra-uncensored-hereticMoEQ3_K_L229B110.22 GiB1.03 GiB112.04 GiB7.00 GiB10±37%
GLM-4.5-AirMoEQ8_0110B109.39 GiB0.76 GiB110.98 GiB8.06 GiB8±37%
GLM-4.5-Air-DerestrictedMoEQ8_0110B109.39 GiB0.76 GiB110.98 GiB8.06 GiB8±37%
GLM-4.6-REAP-268B-A32BMoEQ3_K_S269B108.47 GiB1.53 GiB110.84 GiB8.20 GiB8±37%
DeepSeek-V4-FlashMoEUD-IQ3_S291B109.25 GiB0.03 GiB110.14 GiB8.90 GiB13±37%
MiniMax-M3MoEIQ2_XXS427B108.60 GiB0.50 GiB109.92 GiB9.12 GiB11±37%
Qwen3-Coder-REAP-363B-A35BMoEUD-IQ1_M363B107.85 GiB1.03 GiB109.71 GiB9.33 GiB8±37%
GLM-4.6-Derestricted-v3MoEIQ2_M357B107.14 GiB1.53 GiB109.51 GiB9.53 GiB9±37%
Mixtral-8x22B-v0.1MoEQ6_K141B107.60 GiB0.93 GiB109.39 GiB9.65 GiB3±37%
DeepSeek-V4-Flash-0731MoEUD-IQ3_S304B108.10 GiB0.03 GiB108.98 GiB10.06 GiB13±37%
Llama-4-Scout-17B-16E-InstructMoEKV unresolvedQ8_0109B106.67 GiB0.80 GiB108.29 GiB10.75 GiB8±37%
MiniMax-M2.1-REAP-139B-A10BMoEI1-Q6_K139B106.40 GiB1.03 GiB108.22 GiB10.82 GiB9±37%
m51Lab-MiniMax-M2.7-REAP-139B-A10BMoEQ6_K139B106.40 GiB1.03 GiB108.22 GiB10.82 GiB9±37%
NVIDIA-Nemotron-3-Super-120B-A12B-BF16MoEUD-Q6_K124B106.87 GiB0.37 GiB108.03 GiB11.01 GiB10±37%
Llama-3_1-Nemotron-51B-InstructF1651.5B95.94 GiB10.63 GiB107.50 GiB11.54 GiB2±12.9%
GLM-4.6VMoEQ8_0108B105.81 GiB0.76 GiB107.40 GiB11.64 GiB9±37%
dots.llm1.instMoEQ5_K_M143B101.84 GiB4.12 GiB106.79 GiB12.25 GiB8±37%
Qwen3.5-REAP-212B-A17BMoEIQ4_XS212B105.39 GiB0.12 GiB106.37 GiB12.67 GiB11±37%
Gemma-4-Dark-Gemistry-31BQ8_032.7B102.98 GiB1.29 GiB105.15 GiB13.89 GiB2±12.9%
c4ai-command-r-plus-08-2024Q8_0104B102.74 GiB1.06 GiB104.78 GiB14.26 GiB2±12.9%
Llama-3_3-Nemotron-Super-49B-v1_5BF1649.9B92.89 GiB10.63 GiB104.46 GiB14.58 GiB2±12.9%
Valkyrie-49B-v2.1BF1649.9B92.89 GiB10.63 GiB104.46 GiB14.58 GiB2±12.9%
Llama-3_3-Nemotron-Super-49B-v1BF1649.9B92.89 GiB10.63 GiB104.46 GiB14.58 GiB2±12.9%
●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 8,192 context with q8_0 KV cache, the largest being MiMo-V2-Flash at IQ3_XS. 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.