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 q4_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◐
Hy3MoEIQ3_XXS299B117.43 GiB0.70 GiB118.97 GiB0.07 GiB9±37%
MiMo-V2-FlashMoEKV unresolvedIQ3_XS310B117.68 GiB0.26 GiB118.79 GiB0.25 GiB11±37%
Qwen3.5-397B-A17BMoEUD-IQ2_M403B117.81 GiB0.07 GiB118.72 GiB0.32 GiB12±37%
Mistral-Small-4-119B-2603MoEQ8_0119B117.79 GiB0.05 GiB118.67 GiB0.37 GiB11±37%
Step-3.7-FlashQ4_1201B116.67 GiB1.13 GiB118.64 GiB0.40 GiB2±12.9%
Qwen3-235B-A22B-Instruct-2507MoEIQ4_XS235B117.24 GiB0.41 GiB118.48 GiB0.56 GiB8±37%
Qwen3-235B-A22B-Thinking-2507MoEIQ4_XS235B117.24 GiB0.41 GiB118.48 GiB0.56 GiB8±37%
command-a-plus-05-2026-bf16MoEQ4_0219B117.42 GiB0.19 GiB118.41 GiB0.63 GiB8±37%
MiMo-V2.5MoEKV unresolvedUD-IQ3_XXS311B117.27 GiB0.26 GiB118.38 GiB0.66 GiB11±37%
Qwen3-235B-A22BMoEIQ4_XS235B116.89 GiB0.41 GiB118.13 GiB0.91 GiB8±37%
ERNIE-4.5-300B-A47B-PTUD-IQ3_XXS300B116.56 GiB0.47 GiB117.96 GiB1.08 GiB2±12.9%
Qwen3-VL-235B-A22B-ThinkingMoEIQ4_XS236B116.70 GiB0.41 GiB117.94 GiB1.10 GiB8±37%
Qwen3-VL-235B-A22B-InstructMoEIQ4_XS236B116.70 GiB0.41 GiB117.94 GiB1.10 GiB8±37%
DeepSeek-Coder-V2-Instruct-0724MoEIQ4_XS236B116.94 GiB0.15 GiB117.93 GiB1.11 GiB10±37%
DeepSeek-V2.5MoEIQ4_XS236B116.94 GiB0.15 GiB117.93 GiB1.11 GiB10±37%
DeepSeek-Coder-V2-InstructMoEIQ4_XS236B116.94 GiB0.15 GiB117.93 GiB1.11 GiB10±37%
Qwen3-235B-A22B-abliteratedMoEI1-IQ4_XS235B116.68 GiB0.41 GiB117.93 GiB1.11 GiB8±37%
GLM-4.7-REAP-218B-A32BMoEIQ4_NL218B116.20 GiB0.81 GiB117.84 GiB1.20 GiB7±37%
Trinity-Large-PreviewMoEIQ2_M399B116.50 GiB0.35 GiB117.68 GiB1.36 GiB13±37%
Trinity-Large-TrueBaseMoEIQ2_M399B116.50 GiB0.35 GiB117.68 GiB1.36 GiB13±37%
Qwen3.5-REAP-262B-A17BMoEQ3_K_M262B116.42 GiB0.07 GiB117.33 GiB1.71 GiB11±37%
step-3.5-flashQ4_1199B115.15 GiB1.13 GiB117.11 GiB1.93 GiB2±12.9%
grok-2MoEIQ3_M270B115.25 GiB0.56 GiB116.75 GiB2.29 GiB3±37%
Ornith-1.0-397BMoEUD-IQ2_M397B115.83 GiB0.07 GiB116.75 GiB2.29 GiB12±37%
gpt-oss-120b-abliteratedMoEQ8_0117B115.76 GiB0.09 GiB116.64 GiB2.40 GiB11±37%
GLM-4.5MoEUD-IQ2_M358B114.03 GiB0.81 GiB115.68 GiB3.36 GiB9±37%
GLM-4.7MoEUD-IQ2_M358B114.03 GiB0.81 GiB115.68 GiB3.36 GiB9±37%
MiniMax-M2.7MoEIQ4_XS229B114.00 GiB0.54 GiB115.33 GiB3.71 GiB11±37%
GLM-4.6MoEUD-IQ2_M357B113.56 GiB0.81 GiB115.21 GiB3.83 GiB9±37%
MiniMax-M2.1MoEIQ4_XS229B113.78 GiB0.54 GiB115.11 GiB3.93 GiB11±37%
MiniMax-M2MoEIQ4_XS229B113.78 GiB0.54 GiB115.11 GiB3.93 GiB11±37%
MiniMax-M2.5MoEIQ4_XS229B113.53 GiB0.54 GiB114.86 GiB4.18 GiB11±37%
Llama-4-Maverick-17B-128E-InstructMoEKV unresolvedUD-IQ1_S402B112.48 GiB0.42 GiB113.72 GiB5.32 GiB14±37%
Hermes-4-405BIQ2_XS406B111.15 GiB1.11 GiB113.34 GiB5.70 GiB2±12.9%
Hermes-3-Llama-3.1-405BIQ2_XS406B111.15 GiB1.11 GiB113.34 GiB5.70 GiB2±12.9%
Trinity-Large-ThinkingMoEIQ2_S399B112.03 GiB0.35 GiB113.21 GiB5.83 GiB13±37%
Nex-N2-ProMoEIQ2_S397B112.23 GiB0.07 GiB113.15 GiB5.89 GiB13±37%
Solar-Open2-250BMoEQ3_K_M250B111.63 GiB0.42 GiB112.88 GiB6.16 GiB12±37%
MiniMax-M2.7-BF16-ultra-uncensored-hereticMoEQ3_K_L229B110.22 GiB0.54 GiB111.55 GiB7.49 GiB11±37%
GLM-4.5-AirMoEQ8_0110B109.39 GiB0.40 GiB110.62 GiB8.42 GiB8±37%
GLM-4.5-Air-DerestrictedMoEQ8_0110B109.39 GiB0.40 GiB110.62 GiB8.42 GiB8±37%
GLM-4.6-REAP-268B-A32BMoEQ3_K_S269B108.47 GiB0.81 GiB110.12 GiB8.92 GiB8±37%
DeepSeek-V4-FlashMoEUD-IQ3_S291B109.25 GiB0.02 GiB110.12 GiB8.92 GiB13±37%
MiniMax-M3MoEIQ2_XXS427B108.60 GiB0.26 GiB109.68 GiB9.36 GiB11±37%
Qwen3-Coder-REAP-363B-A35BMoEUD-IQ1_M363B107.85 GiB0.54 GiB109.23 GiB9.81 GiB8±37%
DeepSeek-V4-Flash-0731MoEUD-IQ3_S304B108.10 GiB0.02 GiB108.96 GiB10.08 GiB13±37%
Mixtral-8x22B-v0.1MoEQ6_K141B107.60 GiB0.49 GiB108.95 GiB10.09 GiB4±37%
GLM-4.6-Derestricted-v3MoEIQ2_M357B107.14 GiB0.81 GiB108.79 GiB10.25 GiB9±37%
Llama-4-Scout-17B-16E-InstructMoEKV unresolvedQ8_0109B106.67 GiB0.42 GiB107.91 GiB11.13 GiB9±37%
NVIDIA-Nemotron-3-Super-120B-A12B-BF16MoEUD-Q6_K124B106.87 GiB0.19 GiB107.86 GiB11.18 GiB10±37%
MiniMax-M2.1-REAP-139B-A10BMoEI1-Q6_K139B106.40 GiB0.54 GiB107.73 GiB11.31 GiB9±37%
m51Lab-MiniMax-M2.7-REAP-139B-A10BMoEQ6_K139B106.40 GiB0.54 GiB107.73 GiB11.31 GiB9±37%
GLM-4.6VMoEQ8_0108B105.81 GiB0.40 GiB107.04 GiB12.00 GiB9±37%
Qwen3.5-REAP-212B-A17BMoEIQ4_XS212B105.39 GiB0.07 GiB106.31 GiB12.73 GiB11±37%
dots.llm1.instMoEQ5_K_M143B101.84 GiB2.18 GiB104.85 GiB14.19 GiB9±37%
Gemma-4-Dark-Gemistry-31BQ8_032.7B102.98 GiB0.68 GiB104.54 GiB14.50 GiB2±12.9%
c4ai-command-r-plus-08-2024Q8_0104B102.74 GiB0.56 GiB104.28 GiB14.76 GiB2±12.9%
gpt-oss-20b-hereticMoEQ8_020.9B102.72 GiB0.06 GiB103.57 GiB15.47 GiB6±37%
Huihui-gpt-oss-20b-BF16-abliteratedMoEQ8_020.9B102.72 GiB0.06 GiB103.57 GiB15.47 GiB6±37%
gemma-2-27b-itF3227.2B101.43 GiB0.63 GiB103.00 GiB16.04 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 q4_0 KV cache, the largest being Hy3 at IQ3_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.