Skip to content
Local AI Wizard
Your machine
Theme

DGX Spark · 128 GB unified

NVIDIA · DGX Spark

Workstations & servers128 GB unified memory

193 text models run well · 273 GB/s bandwidth

Models on this hardware

Start here

Text

Within 128 GB: 193 run well · 1 tight · 39 reviewed · 6 more with 256 GB

Text · LiveBench · overall

0–100 · higher is better

  • Color = publisher

Source: LiveBench 2026-06-25 · collected 2026-09-26 · Open models without a score are not drawn (no score ≠ 0).

Rows run well first, then tight fits; within each, reviewed models, then Hub downloads over 30 days, not quality. Sort a score column to rank by that source alone. No model runs only partly on this machine: it either fits in its shared memory or it does not.
Text models that run well within 128 GB
Smaug MiniAbacus AI · 27BQ4 · published params · source76.9Not on the LMArena text leaderboard of 2026-10-22~29 GBQ4 · published paramssource
Qwen3.8 27BAlibaba · 27B · LMArena 1,437Q4 · published params · source75.31,437 ± 5~29 GBQ4 · published paramssource
Qwen3.6 27BAlibaba · 27BQ4 · published params · source64.0Not on the LMArena text leaderboard of 2026-10-22~29 GBQ4 · published paramssource
Qwen3 14BAlibaba · 14.8BQ4 · published params · sourceNo comparable published resultNot on the LMArena text leaderboard of 2026-10-22~20 GBQ4 · published paramssource
Gemma 4 31BGoogle · 30.7B · LMArena 1,452Q4 · published params · sourceNo comparable published result1,452 ± 7~32 GBQ4 · published paramssource
Gemma 4 26B A4BGoogle · 25.2B · LMArena 1,437Q4 · published params · sourceNo comparable published result1,437 ± 7~27 GBQ4 · published paramssource
Qwen3 8BAlibaba · 8.2BQ4 · published params · sourceNo comparable published resultNot on the LMArena text leaderboard of 2026-10-22~15 GBQ4 · published paramssource
Qwen3.5 9BAlibaba · 9BQ4 · published params · sourceNo comparable published resultNot on the LMArena text leaderboard of 2026-10-22~15 GBQ4 · published paramssource
8 of 194

LMArena ratings in this table: LMArena 2026-10-22 · collected 2026-10-25 · dataset · CC BY 4.0

Tight fit: it runs, but leaves less than 12.8 GB free beyond the 8 GB the system keeps (the memory figure already includes that reserve).

Every reviewed text model here 39

Every one that runs here, well or slowly, scored or not; 19 of 39 have a LiveBench or LMArena result. Run well first, then tight fits, then slow; within each, by Hub downloads over 30 days, read 2026-10-25, not by quality.

Reviewed text models that run on DGX Spark · 128 GB unified
Qwen3 14BAlibaba · 14.8BNot in LiveBench 2026-06-25Not on the LMArena text leaderboard of 2026-10-22~20 GB9.0Apache 2.010.4M
Gemma 4 31BGoogle · 30.7BNot in LiveBench 2026-06-251,452~32 GB18.3Apache 2.08.4M
Gemma 4 26B A4BGoogle · 25.2BNot in LiveBench 2026-06-251,437~27 GB16.9Apache 2.08.2M
Qwen3 8BAlibaba · 8.2BNot in LiveBench 2026-06-25Not on the LMArena text leaderboard of 2026-10-22~15 GB5.0Apache 2.08.1M
Qwen3.8 27BAlibaba · 27B75.31,437~29 GB16.5Apache 2.07.8M
Qwen3.5 9BAlibaba · 9BNot in LiveBench 2026-06-25Not on the LMArena text leaderboard of 2026-10-22~15 GB5.7Apache 2.06.1M
gpt-oss-20bOpenAI · 21BNot in LiveBench 2026-06-251,317~24 GB11.6Apache 2.05.8M
Llama 3.1 8B InstructMeta · 8BNot in LiveBench 2026-06-251,211~14 GB4.9Llama 3.14.8M
8 of 39

With 256 GB you also run… 6

The next memory size on this ladder runs these well. Figures are the requirement on this hardware type.

Browse Text models

Adjust the estimate: quantization, context, reserveloads the live catalog

Vision

Within 128 GB: 150 run well · 10 reviewed · 4 more with 256 GB

Vision · LMArena

rating · higher is better

  • Color = publisher

Source: LMArena 2026-10-19 · collected 2026-10-25 · dataset · CC BY 4.0 · Models without a rating are left out, not scored as zero.

Rows run well first, then tight fits; within each, reviewed models, then Hub downloads over 30 days, not quality. Sort a score column to rank by that source alone. No model runs only partly on this machine: it either fits in its shared memory or it does not.
Vision models that run well within 128 GB
Gemma 4 31BGoogle · 30.7B · Image to text · LMArena 1,262Q4 · published params · source1,262 ± 6~32 GBQ4 · published paramssource
Qwen3.8 27BAlibaba · 27B · Image to text · LMArena 1,242Q4 · published params · source1,242 ± 8~29 GBQ4 · published paramssource
Gemma 4 26B A4BGoogle · 25.2B · Image to text · LMArena 1,242Q4 · published params · source1,242 ± 6~27 GBQ4 · published paramssource
gemma-3-27b-itgoogle · 27.43B · Image to text · LMArena 1,160Q4 · Hugging Face count · source1,160 ± 8~29 GBQ4 · Hugging Face countsource
GLM-4.5Vzai-org · 107.71B · Image to text · LMArena 1,153Q4 · Hugging Face count · source1,153 ± 12~89 GBQ4 · Hugging Face countsource
Llama-4-Scout-17B-16E-Instructmeta-llama · 108.6B · Image to text · LMArena 1,128Q4 · Hugging Face count · source1,128 ± 10~90 GBQ4 · Hugging Face countsource
Qwen2.5-VL-32B-InstructQwen · 33.45B · Image to text · LMArena 1,119Q4 · Hugging Face count · source1,119 ± 15~34 GBQ4 · Hugging Face countsource
Molmo2-8Ballenai · 8.7B · Image to text · LMArena 1,112Q4 · Hugging Face count · source1,112 ± 20~15 GBQ4 · Hugging Face countsource
8 of 150

LMArena ratings in this table: LMArena 2026-10-19 · collected 2026-10-25 · dataset · CC BY 4.0

With 256 GB you also run… 4

The next memory size on this ladder runs these well. Figures are the requirement on this hardware type.

12 more vision models in the catalog have no usable memory figure or parameter count for this profile. Browse Vision models

Image

77 within 128 GB · 1 more with 256 GB

Image · Prompt alignment · GenEval

0–1 · higher is better

Source: GenEval · 2023 · reviewed 2026-10-05 · Published quality in the original benchmark preset; not measured on this hardware.

Image models within 128 GB
Stable Diffusion XL 1.0Stability AI · 2.6B · Text to imageExact hardware documented · TensorRT / NGC PyTorch · source0.55Not on the LMArena text-to-image leaderboard of 2026-10-07Documented path without a memory figureExact hardware documentedTensorRT / NGC PyTorch · source
Stable Diffusion 1.5Stability AI · 0.98B · Text to image16-bit est. · sum of parts · source0.43Not on the LMArena text-to-image leaderboard of 2026-10-07~5 GB16-bit est. · sum of partssource
Emu3.5-ImageBAAI · 34.1B · Text + image to image16-bit est. · params · sourceNo comparable published resultNot reviewed for LMArena~84 GB16-bit est. · paramssource
Ming-Image-0.1-DesigninclusionAI · 6.2B · Text to imageConditional path · CUDA · sourceNo comparable published resultNot reviewed for LMArena~80 GBConditional pathCUDA · source
Qwen-Image-EditQwen · 20.4B · Image to image16-bit est. · params · sourceNo comparable published resultNot reviewed for LMArena~51 GB16-bit est. · paramssource
Qwen-Image-Edit-2509Qwen · 20.4B · Image to image16-bit est. · params · sourceNo comparable published resultNot reviewed for LMArena~51 GB16-bit est. · paramssource
Qwen-Image-Edit-2511Qwen · 20.4B · Image to image16-bit est. · params · sourceNo comparable published resultNot reviewed for LMArena~51 GB16-bit est. · paramssource
Qwen-Image-LayeredQwen · 20.4B · Text + image to image16-bit est. · params · sourceNo comparable published resultNot reviewed for LMArena~51 GB16-bit est. · paramssource
8 of 77

LMArena ratings in this table: LMArena 2026-10-07 · collected 2026-10-25 · dataset · CC BY 4.0

With 256 GB you also run… 1

The next memory size on this ladder. Figures are the requirement on this hardware type.

17 more image models in the catalog have no usable memory figure or parameter count for this profile. Browse Image models

Video

73 within 128 GB · 2 more with 256 GB
Video models within 128 GB
MiniMax H3MiniMax · 33B · Text to video16-bit est. · params · source~82 GB16-bit est. · paramssource
Cosmos-H-Surgicalnvidia · 15.2B · Image to videoConditional path · CUDA · source~80 GBConditional pathCUDA · source
Wan-Dancer-14BWan-AI · Image to videoPublisher minimum · source~80 GBPublisher minimumsource
Wan2.2 T2V-A14BWan-AI · 27B · Text to videoConditional path · CUDA · source~80 GBConditional pathCUDA · source
Wan2.2-I2V-A14BWan-AI · 27B · Image to videoConditional path · CUDA · source~80 GBConditional pathCUDA · source
Wan2.2-I2V-A14B-DiffusersWan-AI · 14.3B · Image to videoConditional path · CUDA · source~80 GBConditional pathCUDA · source
Wan2.2-S2V-14BWan-AI · 16.3B · Image to videoConditional path · CUDA · source~80 GBConditional pathCUDA · source
Wan2.2-T2V-A14B-DiffusersWan-AI · 14.3B · Text to videoConditional path · CUDA · source~80 GBConditional pathCUDA · source
8 of 73

How this list works Documented execution paths on this hardware first, then publisher minimums (CUDA or Apple MPS).

With 256 GB you also run… 2

The next memory size on this ladder. Figures are the requirement on this hardware type.

20 more video models in the catalog have no usable memory figure or parameter count for this profile. Browse Video models

Audio

134 within 128 GB

Audio · Word error rate · Open ASR

% · lower is better

Source: Word error rate · Open ASR · reviewed 2026-10-20 · Published quality in the original benchmark preset; not measured on this hardware.

Audio models within 128 GB
Whisper large-v3-turboOpenAI · 0.81B · Speech to textConditional path · CUDA · source7.83~6 GBConditional pathCUDA · source
Whisper large-v3OpenAI · 1.55B · Speech to textConditional path · CUDA · source7.44~10 GBConditional pathCUDA · source
Step-Audio-R1.1stepfun-ai · 33.5B · Speech to text16-bit est. · params · sourceNo comparable published result~83 GB16-bit est. · paramssource
Voxtral-Small-24B-2507mistralai · 24.3B · Speech to text16-bit est. · params · sourceNo comparable published result~61 GB16-bit est. · paramssource
Kimi-Audio-7Bmoonshotai · 9.8B · Text to speech16-bit est. · params · sourceNo comparable published result~26 GB16-bit est. · paramssource
Kimi-Audio-7B-Instructmoonshotai · 9.8B · Text to speech16-bit est. · params · sourceNo comparable published result~26 GB16-bit est. · paramssource
AuKtencent · 1.5B · Text to speechConditional path · CUDA · sourceNo comparable published result~25 GBConditional pathCUDA · source
AuK-Flashtencent · Text to speechConditional path · CUDA · sourceNo comparable published result~25 GBConditional pathCUDA · source
8 of 134

33 more audio models in the catalog have no usable memory figure or parameter count for this profile. Browse Audio models

Embeddings

91 run well within 128 GB

Embeddings · MTEB (Multilingual, v2) · mean(task)

0–100 · higher is better

  • Color = publisher

Source: MTEB results (curated rows, sourced per model)

Embeddings models that run well within 128 GB
Qwen3-Embedding 8BAlibaba (Qwen)8-bit · published params · source70.6~12 GB8-bit · published paramssource
Qwen3-Embedding 4BAlibaba (Qwen)8-bit · published params · source69.5~7 GB8-bit · published paramssource
Harrier OSS v1 (0.6B)Microsoft8-bit · published params · source69.0~3 GB8-bit · published paramssource
Qwen3-Embedding 0.6BAlibaba (Qwen)8-bit · published params · source64.3~3 GB8-bit · published paramssource
Multilingual E5 Large Instructintfloat8-bit · published params · source63.2~3 GB8-bit · published paramssource
EmbeddingGemma 2Google8-bit · published params · source61.4~3 GB8-bit · published paramssource
EmbeddingGemma (300M)Google8-bit · published params · source61.1~3 GB8-bit · published paramssource
BGE-M3BAAI8-bit · published params · source59.6~3 GB8-bit · published paramssource
8 of 91

Browse Embeddings models

Runs well within 128 GB of unified memory · planning estimates, not measured speed bundled catalog copy · 1,777 repositories · checked

Measurements

Published tokens per second, exactly as measured; never estimated.

Published measurements on this hardware
gpt-oss 20B MXFP4 MoEggml-org/gpt-oss-20b-GGUFDGX Spark · 1× NVIDIA GB1083.43 ± 0.59MXFP4 MoE
0 / 32 · Not reported by llama-bench
CUDA
llama.cpp 11fb327bf (7941); CUDA
Flash Attention enabled
llama.cpp bench
2026-09-25
gemma-3-4b-it-qat Q4_0ggml-org/gemma-3-4b-it-qat-GGUF · llama-ben…DGX Spark · 1× NVIDIA GB1081.05 ± 0.20Q4_0
0 / 32 · Single llama-bench process;…
llama.cpp CUDA
llama.cpp commit 11fb327bf
Flash Attention enabled
llama.cpp benches/dgx-spark · maintainer result
2026-10-25
Qwen3.5 35B A3B MXFP4Qwen3.5-35B-A3B MXFP4 GGUF (the source name…DGX Spark · 1× NVIDIA GB10 · 128 GB unified63.13MXFP4 (MoE; 35B total / 3B active)
0 / 128 · Single llama-bench process;…
llama.cpp CUDA (SM121a, source build)
llama.cpp b8922 (commit 13d36cf89); cma…
Flash Attention enabled
nabe2030/dense-27b-31b-dgx-spark (GitHub) · community benchmark
2026-10-09
Qwen3 Coder 30B A3B Q8_0ggml-org/Qwen3-Coder-30B-A3B-Instruct-Q8_0-…DGX Spark · 1× NVIDIA GB1061.06 ± 0.23Q8_0
0 / 32 · Not reported by llama-bench
CUDA
llama.cpp 11fb327bf (7941); CUDA
Flash Attention enabled
llama.cpp bench
2026-09-25
Llama 2 7B Q4_0Llama 2 7B Q4_0 (scoreboard model; the scor…Nvidia GB10 · 128 GB (memory size not stated by the source; this chip is sold with one memory size) · coopmat260.39 ± 0.05Q4_0
0 / 128 · Single llama-bench process;…
Vulkan
llama.cpp commit 657e011
Flash Attention enabled
llama.cpp Vulkan scoreboard · community result
2026-10-25
gpt-oss-120b MXFP4 MoEggml-org/gpt-oss-120b-GGUF · llama-bench mo…DGX Spark · 1× NVIDIA GB1058.72 ± 0.20MXFP4 MoE
0 / 32 · Single llama-bench process;…
llama.cpp CUDA
llama.cpp commit 11fb327bf
Flash Attention enabled
llama.cpp benches/dgx-spark · maintainer result
2026-10-25
Nemotron-3-Nano-4B Q8_0ggml-org/Nemotron-3-Nano-4B-GGUF · llama-be…DGX Spark · 1× NVIDIA GB1052.85 ± 0.12Q8_0
0 / 32 · Single llama-bench process;…
llama.cpp CUDA
llama.cpp commit 1bbec6a75
Flash Attention enabled
llama.cpp benches/nemotron · maintainer result
2026-10-25
GLM-4.7-Flash Q8_0ggml-org/GLM-4.7-Flash-GGUF · llama-bench m…DGX Spark · 1× NVIDIA GB1048.68 ± 0.12Q8_0
0 / 32 · Single llama-bench process;…
llama.cpp CUDA
llama.cpp commit 11fb327bf
Flash Attention enabled
llama.cpp benches/dgx-spark · maintainer result
2026-10-25
Qwen3.8 27B NVFP4 + DSparkRadixArk/Qwen3.8-27B-NVFP4DGX Spark · 1× GB1036.6NVFP4 + DSpark
128 / 128 · C1
SGLang
sparkrun v0.3.4; lmsysorg/sglang@sha256…
NVIDIA Developer Forum
2026-09-25
Qwen2.5-Coder-7B-Q8_0ggml-org/Qwen2.5-Coder-7B-Q8_0-GGUF · llama…DGX Spark · 1× NVIDIA GB1029.43 ± 0.02Q8_0
0 / 32 · Single llama-bench process;…
llama.cpp CUDA
llama.cpp commit 11fb327bf
Flash Attention enabled
llama.cpp benches/dgx-spark · maintainer result
2026-10-25
Nemotron 3 Super 120B A12B Q4_Kggml-org/Nemotron-3-Super-120B-GGUF · Q4_K …DGX Spark · 1× NVIDIA GB10 · 128 GB unified19.94 ± 0.16Q4_K (reported as 'Q4_K' by llama-bench; GGUF from ggml-org)
0 / 32 · Single llama-bench process;…
llama.cpp CUDA
llama.cpp build 04a65daab (8268); n_uba…
Flash Attention enabled
llama.cpp benches/nemotron · DGX Spark
2026-10-09
Qwen3.8 27B NVFP4 + MTPInferact/Qwen3.8-27B-NVFP4DGX Spark · 1× GB1018.5NVFP4 + MTP (3 speculative tokens)
128 / 128 · C1
vLLM official
sparkrun v0.3.4; vllm/vllm-openai:qwen3…
NVIDIA Developer Forum
2026-09-25
Qwen3.6 27B Q4_K_MQwen3.6-27B Q4_K_M GGUF, 15.65 GiB (Unsloth…DGX Spark · 1× NVIDIA GB10 · 128 GB unified11.85 ± 0.01Q4_K_M (15.65 GiB)
0 / 128 · Single llama-bench process;…
llama.cpp CUDA (SM121a, source build)
llama.cpp b8922 (commit 13d36cf89); cma…
Flash Attention enabled
nabe2030/dense-27b-31b-dgx-spark (GitHub) · community benchmark
2026-10-09
Qwen3.6 27B BF16Qwen3.6-27B BF16 GGUF, 50.10 GiB (Unsloth b…DGX Spark · 1× NVIDIA GB10 · 128 GB unified4.54 ± 0.00BF16 (50.10 GiB)
0 / 128 · Single llama-bench process;…
llama.cpp CUDA (SM121a, source build)
llama.cpp b8922 (commit 13d36cf89); cma…
Flash Attention enabled
nabe2030/dense-27b-31b-dgx-spark (GitHub) · community benchmark
2026-10-09

Specifications

21 sourced fields; the source is named once per group, and a field from another source links to it. Official specifications · Memory specification checked: 2026-09-30 · manufacturer source · Sourced facts JSON

Unified memory128 GBShared by CPU and GPU · source
Memory bandwidth273 GB/sManufacturer spec · not model speed · source
Launch price (MSRP)USD 3,999.00US · launched · source · not a tracked offer
Where to buyNVIDIA MarketplaceManufacturer link · this configuration · US · recorded

Markers: the best LiveBench-scored text model that runs well here, and the lightest text model that does not run well here but runs well with 256 GB. Q4 planning estimates, not measured fits.

Selected configuration

Latest check · Manufacturer source

Memory capacity
128 GB
Memory type
128 GB LPDDR5x unified
Chip / processor
NVIDIA GB10 Grace Blackwell
Memory bandwidth
273 GB/s
CPU
20-core Arm (10 Cortex-X925 + 10 Cortex-A725)
CPU cores
20
GPU
NVIDIA Blackwell
Memory interface
256 bit
CUDA cores
6,144
Included external PSU rating
240 W
AI performance (TOPS, manufacturer)
1,000 TOPS · FP4, sparse

Family specification

Latest check · Manufacturer source

Storage options (family)
1 TB or 4 TB NVMe M.2
Ports
4 × USB-C · HDMI 2.1a · 10Gb Ethernet · ConnectX-7
Weight
1.2 kg
Wi-Fi
Wi-Fi 7
Bluetooth
Bluetooth 5.4
Dimensions (cm)
15.0 × 15.0 × 5.05 (L × W × H)
Operating system
NVIDIA DGX OS
CUDA compute capability
12.1
Introduced
2025
Release date
2025-10-15

128 GB coherent unified memory. Check ARM64 runtime and checkpoint support. Capacity alone does not establish throughput or compatibility.

Photo: Official manufacturer image. · Manufacturer source

Memory bandwidth is not measured model speed; verify runtime and checkpoint compatibility. Hardware specification reviewed: 2026-10-18.

How estimates work
Candidates
Reviewed pages use published parameter counts; automatic text candidates use official repository tensor counts (packed quantized weights and unsupported architectures excluded).
Planning
Q4 weights + 20% allowance + reserve: 25% of unified or system memory (at least 2 GB, at most 8 GB; 8 GB here) or 2 GB per dedicated GPU. Adjustable in “Adjust the estimate”. No KV cache is calculated from context.
Fit tiers
Runs well: the estimate fits the fast memory (VRAM, or unified memory within the macOS GPU share). Runs slowly: within the machine's memory but outside the fast part; a graphics card offloads layers to PC RAM (the listed RAM, or 32 GB assumed). Doesn't run: above both. Tiers are planning estimates, never a speed.
Vision
Vision-language models use the Q4 planning estimate from the parameter count (published by the model page, or read from Hugging Face). Classic vision models (classify, detect, segment) use a 16-bit estimate. No vision benchmark is tracked, so rows carry no score.
Media and embeddings
Media rows use documented execution paths (exact hardware or platform) and publisher minimums; a published requirement above this memory moves a model to the next tier. Without either, a 16-bit estimate from the published parameter count (× 1.2 + 2 GB), labelled as such. Embeddings use 8-bit weights + 20% + 2 GB.

Local AI basics: parameters, Q4 and memory · Full model catalog · Data and methodology

Alternatives

Side-by-side comparisons, then other machines on the same memory ladder (unified memory); each page lists the models that fit it.