NVIDIA
Operating mode
Use this to bias workload recommendations toward responsiveness, background autonomy, lighter serving, or multi-GPU scale-out.
Current mode
Balanced
Balanced for general local use. Keeps the ranking neutral across personal and serving workflows.
NVIDIA DGX Spark 128GB is a compact Grace Blackwell personal AI system with 128 GB of coherent unified memory and the NVIDIA CUDA software stack preinstalled. It is aimed at developers who want to prototype, fine-tune, and run much larger local models than fit on 24 GB or 48 GB consumer GPUs, but without jumping straight to a rack-scale server.
Official product page ↗Beyond LLMs
What AI tasks this GPU can handle — from text generation to image and video creation.
| Capability | Status | Representative Model |
|---|---|---|
| LLM Chat (7B) | Runs natively | Llama 3.1 8B Q4 |
| LLM Coding (30B) | Runs natively | Qwen 3 30B Q4 |
| LLM Large (70B) | Runs natively | Llama 3.1 70B Q4 |
| Image Gen (SDXL) | Runs natively | SDXL 1.0 FP16 |
| Image Gen (Flux) | Runs natively | Flux.1 Dev FP16 |
| Image Gen (SD 3.5) | Runs natively | SD 3.5 Large FP16 |
| Video Short (25f) | Runs natively | LTX Video 2B |
| Video Long (100f) | Runs natively | Wan Video 14B |
购买建议
本地 AI 的绝佳选择
能良好运行 50 个顶级模型中的 36 个 — 本地推理的全能之选。
128.0 GB
Unified memory
最适合此 GPU 的模型
What will limit you first
The model fits in shared memory, but shared-memory bandwidth is now the real limiter.
Fit does not mean dedicated-VRAM speed
Unified or shared memory can make a model technically fit, but sustained tokens per second may still trail a discrete high-bandwidth GPU with less total memory.
Shared-memory contention still exists
The OS, browser, and inference runtime all compete for the same physical memory pool, so real-world headroom is less forgiving than raw capacity suggests.
Best upgrade itinerary
Prioritize bandwidth, not only capacity
If this workload feels slow, the next useful step is often a GPU tier with materially faster memory bandwidth rather than only a small bump in capacity.
Unlocks 2 additional models that do not fit on the current setup.
想要更多余量？ NVIDIA H200 141GB (141.0 GB VRAM) 是下一步升级选择。
Chat
SThis model is still usable for chat, but it is not the most specialized pick. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, ollama, lm-studio.
Coding
SThis model is a direct match for coding. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, ollama, lm-studio.
Agentic Coding
SThis model is still usable for agentic-coding, but it is not the most specialized pick. It belongs to a current frontier family for local AI. It should run, but memory headroom will be limited. Known channels: huggingface, lm-studio.
Reasoning
SThis model is a direct match for reasoning. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, ollama, lm-studio.
RAG
SThis model is a direct match for rag. It belongs to a current frontier family for local AI. It should run, but memory headroom will be limited. Known channels: huggingface, lm-studio.
Qwen 3.5 122B A10BQwen3-Coder-NextQwen3-Coder 30B A3B InstructQwen 3.6 35B A3BQwen3-VL 30B A3B Instruct
Command A 111B
Mistral Small 4 119BQwen 3.5 35B A3BDevstral 2 123B InstructQwen 2.5 VL 72BQwen 3 30B A3BQwen 3.5 9BQwen 3.5 27BQwen 3.5 4B
GPT-OSS 120BMagistral Small 2507Qwen 3 8BQwen 3 32BQwen 3.6 27BDevstral Small 2 24B InstructGPT-OSS 20BQwen 3 14BPixtral Large 124BNemotron Cascade 2 30B A3B
Phi-4-reasoning-plus 14BNemotron 3 Nano 30BLeanstral 119B A6BDevstral Small 1.1Phi-4 Mini Reasoning 4B
Gemma 4 31BGemma 4 26B A4BNemotron Nano 8BMinistral 3 14B
EXAONE 4.0 32B
Jina Embeddings v3
BGE M3Qwen 3.5 397B A17B
Kimi K2.5Kimi K2.6
DeepSeek V4 ProDeepSeek V4 Flash
GLM-5.1GLM-5DeepSeek V3.2Qwen 3 235B A22BQwen3-Coder 480B A35B InstructDeepSeek Coder V2 236BDeepSeek R1 671BDeepSeek V3.1 671B触手可及
高质量模型，只需稍多一点内存
Image & Video Generation
51 of 52 models can generate images or video on your NVIDIA DGX Spark 128GB
| Model | Max Resolution | Gen Time | Grade |
|---|---|---|---|
| SD TurboImage | 512×512 | ~8.9s | S |
| Stable Diffusion 1.5Image | 512×768 | ~17.8s | S |
| Realistic Vision v5.1Image | 512×768 | ~17.8s | S |
| DreamShaper 8Image | 512×768 | ~17.8s | S |
| LCM DreamShaper v7Image | 512×768 | ~5.3s | S |
| PixArt-SigmaImage | 1024×1024 | ~1m 11s | S |
| FramePack I2VVideo | 1280×720 | ~2m 11s/frame | S |
| SDXL TurboImage | 512×512 | ~8.9s | S |
| SDXL LightningImage | 1024×1024 | ~26.7s | S |
| Stable Diffusion XL 1.0Image | 1024×1024 | ~1m 11s | S |
| Playground v2.5Image | 1024×1024 | ~1m 47s | S |
| RealVisXL v5.0Image | 1024×1024 | ~1m 20s | S |
| DreamShaper XLImage | 1024×1024 | ~1m 20s | S |
| Juggernaut XL v9Image | 1024×1024 | ~1m 20s | S |
| Animagine XL 3.1Image | 1024×1024 | ~1m 20s | S |
| Pony Diffusion V6 XLImage | 1024×1024 | ~1m 20s | S |
| Animagine XL 4.0Image | 1024×1024 | ~1m 20s | S |
| Illustrious XLImage | 1024×1024 | ~1m 20s | S |
| Wan Video 2.1 1.3BVideo | 480×832 | ~52s/frame | S |
| Stable Diffusion 3.5 MediumImage | 1024×1024 | ~2m 5s | S |
| Flux.2 Klein 4BImage | 1024×1024 | ~21.4s | S |
| LTX Video 2BVideo | 1280×720 | ~1m 2s/frame | S |
| KolorsImage | 1024×1024 | ~2m 22s | S |
| Stable CascadeImage | 1024×1024 | ~2m 58s | S |
| AuraFlow v0.3Image | 1536×1536 | ~5m 21s | S |
| Stable Diffusion 3.5 LargeImage | 1024×1024 | ~6m 32s | S |
| Stable Diffusion 3.5 Large TurboImage | 1024×1024 | ~1m 11s | S |
| CogVideoX 2BVideo | 720×480 | ~1m 2s/frame | S |
| HunyuanVideoVideo | 720×1280 | ~2m 11s/frame | S |
| ChromaImage | 1024×1024 | ~1m 11s | S |
| Z-Image TurboImage | 1536×1536 | ~1m 14s | S |
| Flux.1 DevImage | 1024×1024 | ~5m 21s | S |
| Flux.1 SchnellImage | 1024×1024 | ~1m 2s | S |
| LTX Video 13BVideo | 1280×720 | ~2m 11s/frame | S |
| Flux.1 Kontext DevImage | 1024×1024 | ~5m 56s | S |
| AnimateDiff v1.5.3Video | 512×768 | ~32.5s/frame | S |
| Cosmos Diffusion 7BVideo | 1024×576 | ~1m 42s/frame | S |
| CogVideoX 5BVideo | 720×480 | ~1m 29s/frame | S |
| Wan2.2 TI2V 5BVideo | 832×480 | ~1m 29s/frame | S |
| Flux.2 Klein 9BImage | 1024×1024 | ~35.6s | S |
| Flux.1 Fill DevImage | 1024×1024 | ~5m 3s | S |
| Mochi 1 PreviewVideo | 848×480 | ~1m 58s/frame | S |
| HunyuanVideo 1.5Video | 720×1280 | ~1m 49s/frame | S |
| Helios 14BVideo | 1280×720 | ~2m 15s/frame | S |
| SkyReels V2 14BVideo | 1280×720 | ~2m 15s/frame | S |
| Wan Video 2.1 14BVideo | 720×1280 | ~2m 15s/frame | S |
| Wan Video 2.2 14BVideo | 720×1280 | ~2m 15s/frame | S |
| Qwen ImageImage | 1024×1024 | ~1m 60s | S |
| Qwen Image EditImage | 1024×1024 | ~1m 60s | S |
| Flux.2 DevImage | 1024×1024 | ~56m 10s | S |
| MAGI-1Video | 1280×720 | ~2m 47s/frame | S |
| HunyuanImage 3.0Image | 256×256 | ~3m 31s | F |
Image models estimated at 1024×1024 (28 steps, FP16). Video models estimated at 768×512 (25 frames, 30 steps, FP16). Actual performance varies with runtime and system load.
Upgrade paths
See what you unlock with more powerful hardware
升级选项
NVIDIA H200 141GB下一步升级Unlocks 2 additional models that do not fit on the current setup.
Lifts average decode speed across fitting models by about 423%.
~$30,000 MSRP
NVIDIA B200 180GBNVIDIA 升级Unlocks 8 additional models that do not fit on the current setup.
Lifts average decode speed across fitting models by about 539%.
~$30,000 MSRP
Unlocks 12 additional models that do not fit on the current setup.
Lifts average decode speed across fitting models by about 433%.
~$20,000 MSRP
Unlocks 13 additional models that do not fit on the current setup.
Lifts average decode speed across fitting models by about 498%.
~$8,000 MSRP
NVIDIA DGX Spark 128GB (128 GB unified memory) can run these top models: Qwen 3.5 122B A10B (score: 88/100), Qwen3-Coder-Next (score: 88/100), Qwen3-Coder 30B A3B Instruct (score: 87/100). See the full compatibility list above.
NVIDIA DGX Spark 128GB ships with 128 GB of unified memory, with roughly 108.8 GB realistically usable for AI inference after OS and runtime overhead.
Yes, NVIDIA DGX Spark 128GB is excellent for running LLMs locally with top compatibility scores above 80/100.
For coding on NVIDIA DGX Spark 128GB, we recommend Qwen3-Coder-Next. It achieves 11.1 tokens per second with 256K context window. This model is a direct match for coding. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, ollama, lm-studio.
There are 4 upgrade path(s) from NVIDIA DGX Spark 128GB: NVIDIA H200 141GB, NVIDIA B200 180GB. Upgrading would unlock larger models and faster inference speeds.
Yes, NVIDIA DGX Spark 128GB with 109 GB of usable memory can run Flux.1 Dev at FP16 natively. Flux is a 12B parameter diffusion transformer that produces high-quality images. You can also run the Schnell variant for faster generation.
NVIDIA DGX Spark 128GB (128 GB unified memory) can handle various AI generation tasks beyond LLMs. For image generation, SDXL and Stable Diffusion 3.5 run well. Flux.1 Dev also runs natively for state-of-the-art image quality. For video, LTX Video 2.3 can generate short clips. Check the AI Capability Matrix above for detailed compatibility.
NVIDIA DGX Spark 128GB is excellent for AI image generation. With 109 GB of usable memory, it runs all major diffusion models including Flux.1, SDXL, and Stable Diffusion 3.5 at full precision. You can generate high-resolution images quickly and even handle video generation models.
Yes, NVIDIA DGX Spark 128GB with 109 GB of usable memory can run Qwen 3.5 27B at Q8 (near-lossless, ~28.9 GB) or even FP16 (~55.4 GB) depending on your context needs. This setup provides an excellent experience with this model. Use Ollama or vLLM for best results.
With 109 GB VRAM on NVIDIA DGX Spark 128GB, use Q8_0 for most models — it is near-lossless and you have the memory for it. For 70B+ models, Q6_K offers excellent quality. Reserve Q4_K_M for 100B+ models or when you need maximum context length.
NVIDIA DGX Spark 128GB has enough memory for many local LLMs, but bandwidth still matters a lot for real speed. Once a model fits, a faster-memory GPU can feel significantly better than a slower setup with similar capacity.
Compare with similar
