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 4K 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 4K context

largest quantization that fits, per model · 2091 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Hy3MoEIQ3_XXS299B117.43 GiB0.66 GiB118.93 GiB0.11 GiB9±37%
Step-3.7-FlashQ4_1201B116.67 GiB1.34 GiB118.85 GiB0.19 GiB2±12.9%
MiMo-V2-FlashMoEKV unresolvedIQ3_XS310B117.68 GiB0.25 GiB118.78 GiB0.26 GiB11±37%
Qwen3.5-397B-A17BMoEUD-IQ2_M403B117.81 GiB0.06 GiB118.72 GiB0.32 GiB12±37%
Mistral-Small-4-119B-2603MoEQ8_0119B117.79 GiB0.05 GiB118.67 GiB0.37 GiB11±37%
command-a-plus-05-2026-bf16MoEQ4_0219B117.42 GiB0.27 GiB118.49 GiB0.55 GiB8±37%
Qwen3-235B-A22B-Instruct-2507MoEIQ4_XS235B117.24 GiB0.39 GiB118.46 GiB0.58 GiB8±37%
Qwen3-235B-A22B-Thinking-2507MoEIQ4_XS235B117.24 GiB0.39 GiB118.46 GiB0.58 GiB8±37%
MiMo-V2.5MoEKV unresolvedUD-IQ3_XXS311B117.27 GiB0.25 GiB118.37 GiB0.67 GiB11±37%
Qwen3-235B-A22BMoEIQ4_XS235B116.89 GiB0.39 GiB118.11 GiB0.93 GiB8±37%
ERNIE-4.5-300B-A47B-PTUD-IQ3_XXS300B116.56 GiB0.45 GiB117.93 GiB1.11 GiB2±12.9%
Qwen3-VL-235B-A22B-ThinkingMoEIQ4_XS236B116.70 GiB0.39 GiB117.92 GiB1.12 GiB8±37%
Qwen3-VL-235B-A22B-InstructMoEIQ4_XS236B116.70 GiB0.39 GiB117.92 GiB1.12 GiB8±37%
DeepSeek-Coder-V2-Instruct-0724MoEIQ4_XS236B116.94 GiB0.14 GiB117.92 GiB1.12 GiB10±37%
DeepSeek-V2.5MoEIQ4_XS236B116.94 GiB0.14 GiB117.92 GiB1.12 GiB10±37%
DeepSeek-Coder-V2-InstructMoEIQ4_XS236B116.94 GiB0.14 GiB117.92 GiB1.12 GiB10±37%
Qwen3-235B-A22B-abliteratedMoEI1-IQ4_XS235B116.68 GiB0.39 GiB117.90 GiB1.14 GiB8±37%
Trinity-Large-PreviewMoEIQ2_M399B116.50 GiB0.50 GiB117.83 GiB1.21 GiB12±37%
Trinity-Large-TrueBaseMoEIQ2_M399B116.50 GiB0.50 GiB117.83 GiB1.21 GiB12±37%
GLM-4.7-REAP-218B-A32BMoEIQ4_NL218B116.20 GiB0.76 GiB117.80 GiB1.24 GiB7±37%
Qwen3.5-REAP-262B-A17BMoEQ3_K_M262B116.42 GiB0.06 GiB117.33 GiB1.71 GiB11±37%
step-3.5-flashQ4_1199B115.15 GiB1.34 GiB117.32 GiB1.72 GiB2±12.9%
Ornith-1.0-397BMoEUD-IQ2_M397B115.83 GiB0.06 GiB116.74 GiB2.30 GiB12±37%
grok-2MoEIQ3_M270B115.25 GiB0.53 GiB116.72 GiB2.32 GiB3±37%
gpt-oss-120b-abliteratedMoEQ8_0117B115.76 GiB0.09 GiB116.64 GiB2.40 GiB11±37%
GLM-4.5MoEUD-IQ2_M358B114.03 GiB0.76 GiB115.64 GiB3.40 GiB9±37%
GLM-4.7MoEUD-IQ2_M358B114.03 GiB0.76 GiB115.64 GiB3.40 GiB9±37%
MiniMax-M2.7MoEIQ4_XS229B114.00 GiB0.51 GiB115.30 GiB3.74 GiB11±37%
GLM-4.6MoEUD-IQ2_M357B113.56 GiB0.76 GiB115.16 GiB3.88 GiB9±37%
MiniMax-M2.1MoEIQ4_XS229B113.78 GiB0.51 GiB115.08 GiB3.96 GiB11±37%
MiniMax-M2MoEIQ4_XS229B113.78 GiB0.51 GiB115.08 GiB3.96 GiB11±37%
MiniMax-M2.5MoEIQ4_XS229B113.53 GiB0.51 GiB114.83 GiB4.21 GiB11±37%
Llama-4-Maverick-17B-128E-InstructMoEKV unresolvedUD-IQ1_S402B112.48 GiB0.40 GiB113.70 GiB5.34 GiB14±37%
Trinity-Large-ThinkingMoEIQ2_S399B112.03 GiB0.50 GiB113.36 GiB5.68 GiB13±37%
Hermes-4-405BIQ2_XS406B111.15 GiB1.05 GiB113.28 GiB5.76 GiB2±12.9%
Hermes-3-Llama-3.1-405BIQ2_XS406B111.15 GiB1.05 GiB113.28 GiB5.76 GiB2±12.9%
Nex-N2-ProMoEIQ2_S397B112.23 GiB0.06 GiB113.15 GiB5.89 GiB13±37%
Solar-Open2-250BMoEQ3_K_M250B111.63 GiB0.40 GiB112.86 GiB6.18 GiB12±37%
MiniMax-M2.7-BF16-ultra-uncensored-hereticMoEQ3_K_L229B110.22 GiB0.51 GiB111.52 GiB7.52 GiB11±37%
GLM-4.5-AirMoEQ8_0110B109.39 GiB0.38 GiB110.60 GiB8.44 GiB8±37%
GLM-4.5-Air-DerestrictedMoEQ8_0110B109.39 GiB0.38 GiB110.60 GiB8.44 GiB8±37%
DeepSeek-V4-FlashMoEUD-IQ3_S291B109.25 GiB0.03 GiB110.14 GiB8.90 GiB13±37%
GLM-4.6-REAP-268B-A32BMoEQ3_K_S269B108.47 GiB0.76 GiB110.08 GiB8.96 GiB8±37%
MiniMax-M3MoEIQ2_XXS427B108.60 GiB0.25 GiB109.67 GiB9.37 GiB11±37%
Qwen3-Coder-REAP-363B-A35BMoEUD-IQ1_M363B107.85 GiB0.51 GiB109.20 GiB9.84 GiB8±37%
DeepSeek-V4-Flash-0731MoEUD-IQ3_S304B108.10 GiB0.03 GiB108.98 GiB10.06 GiB13±37%
Mixtral-8x22B-v0.1MoEQ6_K141B107.60 GiB0.46 GiB108.92 GiB10.12 GiB4±37%
GLM-4.6-Derestricted-v3MoEIQ2_M357B107.14 GiB0.76 GiB108.75 GiB10.29 GiB9±37%
Llama-4-Scout-17B-16E-InstructMoEKV unresolvedQ8_0109B106.67 GiB0.40 GiB107.89 GiB11.15 GiB9±37%
NVIDIA-Nemotron-3-Super-120B-A12B-BF16MoEUD-Q6_K124B106.87 GiB0.18 GiB107.85 GiB11.19 GiB10±37%
MiniMax-M2.1-REAP-139B-A10BMoEI1-Q6_K139B106.40 GiB0.51 GiB107.70 GiB11.34 GiB9±37%
m51Lab-MiniMax-M2.7-REAP-139B-A10BMoEQ6_K139B106.40 GiB0.51 GiB107.70 GiB11.34 GiB9±37%
GLM-4.6VMoEQ8_0108B105.81 GiB0.38 GiB107.02 GiB12.02 GiB9±37%
Qwen3.5-REAP-212B-A17BMoEIQ4_XS212B105.39 GiB0.06 GiB106.31 GiB12.73 GiB11±37%
Gemma-4-Dark-Gemistry-31BQ8_032.7B102.98 GiB0.95 GiB104.82 GiB14.22 GiB2±12.9%
dots.llm1.instMoEQ5_K_M143B101.84 GiB2.06 GiB104.73 GiB14.31 GiB9±37%
c4ai-command-r-plus-08-2024Q8_0104B102.74 GiB0.53 GiB104.25 GiB14.79 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 GiB0.76 GiB103.13 GiB15.91 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 4,096 context with q8_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.