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 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 16K context

largest quantization that fits, per model · 2091 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Qwen3-235B-A22B-Instruct-2507MoEIQ4_XS235B117.24 GiB0.83 GiB118.90 GiB0.14 GiB8±37%
Qwen3-235B-A22B-Thinking-2507MoEIQ4_XS235B117.24 GiB0.83 GiB118.90 GiB0.14 GiB8±37%
Qwen3.5-397B-A17BMoEUD-IQ2_M403B117.81 GiB0.13 GiB118.79 GiB0.25 GiB12±37%
Mistral-Small-4-119B-2603MoEQ8_0119B117.79 GiB0.10 GiB118.72 GiB0.32 GiB11±37%
GLM-4.7-REAP-218B-A32BMoEIQ4_NL218B116.20 GiB1.62 GiB118.65 GiB0.39 GiB6±37%
MiMo-V2.5MoEKV unresolvedUD-IQ3_XXS311B117.27 GiB0.53 GiB118.65 GiB0.39 GiB10±37%
Qwen3-235B-A22BMoEIQ4_XS235B116.89 GiB0.83 GiB118.54 GiB0.50 GiB8±37%
MiMo-V2-FlashMoEKV unresolvedI1-IQ3_XS310B117.12 GiB0.53 GiB118.50 GiB0.54 GiB10±37%
command-a-plus-05-2026-bf16MoEQ4_0219B117.42 GiB0.26 GiB118.48 GiB0.56 GiB8±37%
ERNIE-4.5-300B-A47B-PTUD-IQ3_XXS300B116.56 GiB0.95 GiB118.43 GiB0.61 GiB2±12.9%
Qwen3-VL-235B-A22B-ThinkingMoEIQ4_XS236B116.70 GiB0.83 GiB118.36 GiB0.68 GiB8±37%
Qwen3-VL-235B-A22B-InstructMoEIQ4_XS236B116.70 GiB0.83 GiB118.36 GiB0.68 GiB8±37%
Qwen3-235B-A22B-abliteratedMoEI1-IQ4_XS235B116.68 GiB0.83 GiB118.34 GiB0.70 GiB8±37%
DeepSeek-Coder-V2-Instruct-0724MoEIQ4_XS236B116.94 GiB0.30 GiB118.08 GiB0.96 GiB10±37%
DeepSeek-V2.5MoEIQ4_XS236B116.94 GiB0.30 GiB118.08 GiB0.96 GiB10±37%
DeepSeek-Coder-V2-InstructMoEIQ4_XS236B116.94 GiB0.30 GiB118.08 GiB0.96 GiB10±37%
step-3.5-flashQ4_1199B115.15 GiB1.98 GiB117.96 GiB1.08 GiB2±12.9%
Trinity-Large-PreviewMoEIQ2_M399B116.50 GiB0.49 GiB117.82 GiB1.22 GiB12±37%
Trinity-Large-TrueBaseMoEIQ2_M399B116.50 GiB0.49 GiB117.82 GiB1.22 GiB12±37%
Qwen3.5-REAP-262B-A17BMoEQ3_K_M262B116.42 GiB0.13 GiB117.40 GiB1.64 GiB11±37%
grok-2MoEIQ3_M270B115.25 GiB1.13 GiB117.32 GiB1.72 GiB3±37%
Ornith-1.0-397BMoEUD-IQ2_M397B115.83 GiB0.13 GiB116.81 GiB2.23 GiB12±37%
gpt-oss-120b-abliteratedMoEQ8_0117B115.76 GiB0.17 GiB116.72 GiB2.32 GiB11±37%
Step-3.7-FlashUD-Q4_K_M201B113.71 GiB1.98 GiB116.51 GiB2.53 GiB2±12.9%
GLM-4.5MoEUD-IQ2_M358B114.03 GiB1.62 GiB116.49 GiB2.55 GiB8±37%
GLM-4.7MoEUD-IQ2_M358B114.03 GiB1.62 GiB116.49 GiB2.55 GiB8±37%
GLM-4.6MoEUD-IQ2_M357B113.56 GiB1.62 GiB116.01 GiB3.03 GiB8±37%
MiniMax-M2.7MoEIQ4_XS229B114.00 GiB1.09 GiB115.87 GiB3.17 GiB10±37%
MiniMax-M2.1MoEIQ4_XS229B113.78 GiB1.09 GiB115.65 GiB3.39 GiB10±37%
MiniMax-M2MoEIQ4_XS229B113.78 GiB1.09 GiB115.65 GiB3.39 GiB10±37%
MiniMax-M2.5MoEIQ4_XS229B113.53 GiB1.09 GiB115.40 GiB3.64 GiB10±37%
Hermes-4-405BIQ2_XS406B111.15 GiB2.21 GiB114.45 GiB4.59 GiB2±12.9%
Hermes-3-Llama-3.1-405BIQ2_XS406B111.15 GiB2.21 GiB114.45 GiB4.59 GiB2±12.9%
Llama-4-Maverick-17B-128E-InstructMoEKV unresolvedUD-IQ1_S402B112.48 GiB0.84 GiB114.15 GiB4.89 GiB13±37%
Trinity-Large-ThinkingMoEIQ2_S399B112.03 GiB0.49 GiB113.34 GiB5.70 GiB13±37%
Solar-Open2-250BMoEQ3_K_M250B111.63 GiB0.84 GiB113.30 GiB5.74 GiB11±37%
Nex-N2-ProMoEIQ2_S397B112.23 GiB0.13 GiB113.22 GiB5.82 GiB13±37%
MiniMax-M2.7-BF16-ultra-uncensored-hereticMoEQ3_K_L229B110.22 GiB1.09 GiB112.10 GiB6.94 GiB10±37%
GLM-4.5-AirMoEQ8_0110B109.39 GiB0.81 GiB111.03 GiB8.01 GiB8±37%
GLM-4.5-Air-DerestrictedMoEQ8_0110B109.39 GiB0.81 GiB111.03 GiB8.01 GiB8±37%
GLM-4.6-REAP-268B-A32BMoEQ3_K_S269B108.47 GiB1.62 GiB110.93 GiB8.11 GiB8±37%
DeepSeek-V4-FlashMoEUD-IQ3_S291B109.25 GiB0.02 GiB110.12 GiB8.92 GiB13±37%
MiniMax-M3MoEIQ2_XXS427B108.60 GiB0.53 GiB109.95 GiB9.09 GiB11±37%
Qwen3-Coder-REAP-363B-A35BMoEUD-IQ1_M363B107.85 GiB1.09 GiB109.77 GiB9.27 GiB8±37%
GLM-4.6-Derestricted-v3MoEIQ2_M357B107.14 GiB1.62 GiB109.60 GiB9.44 GiB9±37%
Mixtral-8x22B-v0.1MoEQ6_K141B107.60 GiB0.98 GiB109.44 GiB9.60 GiB3±37%
DeepSeek-V4-Flash-0731MoEUD-IQ3_S304B108.10 GiB0.02 GiB108.96 GiB10.08 GiB13±37%
Llama-4-Scout-17B-16E-InstructMoEKV unresolvedQ8_0109B106.67 GiB0.84 GiB108.34 GiB10.70 GiB8±37%
MiniMax-M2.1-REAP-139B-A10BMoEI1-Q6_K139B106.40 GiB1.09 GiB108.28 GiB10.76 GiB9±37%
m51Lab-MiniMax-M2.7-REAP-139B-A10BMoEQ6_K139B106.40 GiB1.09 GiB108.28 GiB10.76 GiB9±37%
Llama-3_1-Nemotron-51B-InstructF1651.5B95.94 GiB11.25 GiB108.13 GiB10.91 GiB2±12.9%
NVIDIA-Nemotron-3-Super-120B-A12B-BF16MoEUD-Q6_K124B106.87 GiB0.39 GiB108.05 GiB10.99 GiB10±37%
GLM-4.6VMoEQ8_0108B105.81 GiB0.81 GiB107.45 GiB11.59 GiB8±37%
dots.llm1.instMoEQ5_K_M143B101.84 GiB4.36 GiB107.03 GiB12.01 GiB8±37%
Qwen3.5-REAP-212B-A17BMoEIQ4_XS212B105.39 GiB0.13 GiB106.38 GiB12.66 GiB11±37%
Llama-3_3-Nemotron-Super-49B-v1_5BF1649.9B92.89 GiB11.25 GiB105.09 GiB13.95 GiB2±12.9%
Valkyrie-49B-v2.1BF1649.9B92.89 GiB11.25 GiB105.09 GiB13.95 GiB2±12.9%
Llama-3_3-Nemotron-Super-49B-v1BF1649.9B92.89 GiB11.25 GiB105.09 GiB13.95 GiB2±12.9%
Gemma-4-Dark-Gemistry-31BQ8_032.7B102.98 GiB1.03 GiB104.89 GiB14.15 GiB2±12.9%
c4ai-command-r-plus-08-2024Q8_0104B102.74 GiB1.13 GiB104.84 GiB14.20 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 16,384 context with q4_0 KV cache, the largest being Qwen3-235B-A22B-Instruct-2507 at IQ4_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.