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 64K 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 64K context

largest quantization that fits, per model · 2091 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Mistral-Small-4-119B-2603MoEQ8_0119B117.79 GiB0.40 GiB119.01 GiB0.03 GiB10±37%
DeepSeek-Coder-V2-Instruct-0724MoEIQ4_XS236B116.94 GiB1.19 GiB118.97 GiB0.07 GiB9±37%
DeepSeek-V2.5MoEIQ4_XS236B116.94 GiB1.19 GiB118.97 GiB0.07 GiB9±37%
DeepSeek-Coder-V2-InstructMoEIQ4_XS236B116.94 GiB1.19 GiB118.97 GiB0.07 GiB9±37%
MiniMax-M2.1MoEIQ4_XS229B113.78 GiB4.36 GiB118.92 GiB0.12 GiB8±37%
MiniMax-M2MoEIQ4_XS229B113.78 GiB4.36 GiB118.92 GiB0.12 GiB8±37%
command-a-plus-05-2026-bf16MoEQ4_0219B117.42 GiB0.68 GiB118.90 GiB0.14 GiB8±37%
MiniMax-M2.5MoEIQ4_XS229B113.53 GiB4.36 GiB118.67 GiB0.37 GiB8±37%
Trinity-Large-PreviewMoEIQ2_M399B116.50 GiB1.28 GiB118.61 GiB0.43 GiB11±37%
Trinity-Large-TrueBaseMoEIQ2_M399B116.50 GiB1.28 GiB118.61 GiB0.43 GiB11±37%
step-3.5-flashQ4_K199B110.56 GiB7.04 GiB118.43 GiB0.61 GiB2±12.9%
Qwen3.5-397B-A17BMoEIQ2_XS403B116.79 GiB0.53 GiB118.17 GiB0.87 GiB12±37%
Qwen3.5-REAP-262B-A17BMoEQ3_K_M262B116.42 GiB0.53 GiB117.80 GiB1.24 GiB11±37%
Qwen3-235B-A22B-abliteratedMoEI1-Q3_K_L235B113.46 GiB3.30 GiB117.60 GiB1.44 GiB7±37%
GLM-4.7-REAP-218B-A32BMoEIQ4_XS218B110.09 GiB6.47 GiB117.40 GiB1.64 GiB5±37%
Ornith-1.0-397BMoEUD-IQ2_M397B115.83 GiB0.53 GiB117.21 GiB1.83 GiB12±37%
gpt-oss-120b-abliteratedMoEQ8_0117B115.76 GiB0.64 GiB117.19 GiB1.85 GiB10±37%
Step-3.7-FlashQ4_K_S201B109.05 GiB7.04 GiB116.92 GiB2.12 GiB2±12.9%
Llama-4-Maverick-17B-128E-InstructMoEKV unresolvedUD-IQ1_S402B112.48 GiB3.38 GiB116.68 GiB2.36 GiB10±37%
MiMo-V2-FlashMoEKV unresolvedIQ3_XXS310B113.14 GiB2.11 GiB116.10 GiB2.94 GiB9±37%
Solar-Open2-250BMoEQ3_K_M250B111.63 GiB3.38 GiB115.83 GiB3.21 GiB9±37%
GLM-4.6-REAP-268B-A32BMoEQ3_K_S269B108.47 GiB6.47 GiB115.78 GiB3.26 GiB6±37%
grok-2MoEQ3_K_S270B109.94 GiB4.50 GiB115.38 GiB3.66 GiB3±37%
MiniMax-M2.7-BF16-ultra-uncensored-hereticMoEQ3_K_L229B110.22 GiB4.36 GiB115.37 GiB3.67 GiB8±37%
GLM-4.5MoEUD-IQ2_XXS358B107.95 GiB6.47 GiB115.26 GiB3.78 GiB6±37%
GLM-4.7MoEUD-IQ2_XXS358B107.95 GiB6.47 GiB115.26 GiB3.78 GiB6±37%
GLM-4.6MoEUD-IQ2_XXS357B107.48 GiB6.47 GiB114.79 GiB4.25 GiB6±37%
GLM-4.6-Derestricted-v3MoEIQ2_M357B107.14 GiB6.47 GiB114.45 GiB4.59 GiB6±37%
Trinity-Large-ThinkingMoEIQ2_S399B112.03 GiB1.28 GiB114.13 GiB4.91 GiB12±37%
Nex-N2-ProMoEIQ2_S397B112.23 GiB0.53 GiB113.61 GiB5.43 GiB12±37%
GLM-4.5-AirMoEQ8_0110B109.39 GiB3.23 GiB113.45 GiB5.59 GiB7±37%
GLM-4.5-Air-DerestrictedMoEQ8_0110B109.39 GiB3.23 GiB113.45 GiB5.59 GiB7±37%
dots.llm1.instMoEQ5_K_S143B95.08 GiB17.44 GiB113.35 GiB5.69 GiB4±37%
Qwen3-Coder-REAP-363B-A35BMoEUD-IQ1_M363B107.85 GiB4.36 GiB113.04 GiB6.00 GiB6±37%
Mixtral-8x22B-v0.1MoEQ6_K141B107.60 GiB3.94 GiB112.39 GiB6.65 GiB3±37%
MiniMax-M2.1-REAP-139B-A10BMoEI1-Q6_K139B106.40 GiB4.36 GiB111.55 GiB7.49 GiB7±37%
m51Lab-MiniMax-M2.7-REAP-139B-A10BMoEQ6_K139B106.40 GiB4.36 GiB111.55 GiB7.49 GiB7±37%
MiniMax-M3MoEIQ2_XXS427B108.60 GiB2.11 GiB111.53 GiB7.51 GiB10±37%
Llama-4-Scout-17B-16E-InstructMoEKV unresolvedQ8_0109B106.67 GiB3.38 GiB110.87 GiB8.17 GiB7±37%
DeepSeek-V4-FlashMoEUD-IQ3_S291B109.25 GiB0.02 GiB110.12 GiB8.92 GiB13±37%
MiMo-V2.5MoEKV unresolvedUD-IQ3_S311B106.98 GiB2.11 GiB109.93 GiB9.11 GiB10±37%
GLM-4.6VMoEQ8_0108B105.81 GiB3.23 GiB109.87 GiB9.17 GiB7±37%
Hermes-4-405BIQ2_XXS406B99.91 GiB8.86 GiB109.85 GiB9.19 GiB2±12.9%
Hermes-3-Llama-3.1-405BIQ2_XXS406B99.91 GiB8.86 GiB109.85 GiB9.19 GiB2±12.9%
NVIDIA-Nemotron-3-Super-120B-A12B-BF16MoEUD-Q6_K124B106.87 GiB1.55 GiB109.21 GiB9.83 GiB9±37%
DeepSeek-V4-Flash-0731MoEUD-IQ3_S304B108.10 GiB0.02 GiB108.96 GiB10.08 GiB13±37%
Qwen3-VL-235B-A22B-ThinkingMoEQ3_K_M236B104.72 GiB3.30 GiB108.86 GiB10.18 GiB7±37%
Qwen3-VL-235B-A22B-InstructMoEQ3_K_M236B104.72 GiB3.30 GiB108.86 GiB10.18 GiB7±37%
Qwen3-235B-A22BMoEQ3_K_M235B104.72 GiB3.30 GiB108.86 GiB10.18 GiB7±37%
Qwen3-235B-A22B-Thinking-2507MoEQ3_K_M235B104.72 GiB3.30 GiB108.86 GiB10.18 GiB7±37%
Qwen3-235B-A22B-Instruct-2507MoEQ3_K_M235B104.72 GiB3.30 GiB108.86 GiB10.18 GiB7±37%
MiniMax-M2.7MoEUD-IQ4_NL229B103.15 GiB4.36 GiB108.29 GiB10.75 GiB8±37%
Mistral-Medium-3.5-128BQ6_K_L128B101.13 GiB6.19 GiB108.27 GiB10.77 GiB2±12.9%
c4ai-command-r-plus-08-2024Q8_0104B102.74 GiB4.50 GiB108.22 GiB10.82 GiB2±12.9%
Hy3MoEQ2_K299B101.28 GiB5.63 GiB107.74 GiB11.30 GiB7±37%
Gemma-4-Dark-Gemistry-31BQ8_032.7B102.98 GiB3.14 GiB107.00 GiB12.04 GiB2±12.9%
Qwen3.5-REAP-212B-A17BMoEIQ4_XS212B105.39 GiB0.53 GiB106.77 GiB12.27 GiB11±37%
ERNIE-4.5-300B-A47B-PTQ2_K_L300B101.76 GiB3.80 GiB106.48 GiB12.56 GiB2±12.9%
gemma-2-27b-itF3227.2B101.43 GiB3.46 GiB105.83 GiB13.21 GiB2±12.9%
gpt-oss-20b-hereticMoEQ8_020.9B102.72 GiB0.43 GiB103.94 GiB15.10 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 65,536 context with q4_0 KV cache, the largest being Mistral-Small-4-119B-2603 at Q8_0. 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.