Hardware Detail desktopCurrent

NVIDIA DGX Spark 128GB

Compact Grace Blackwell desktop system for local agents, large-model inference, and AI development.

NVIDIANVIDIA DESKTOP2025-10128GB unified

AI Memory

128GB unified

Usable model memory reserves headroom for OS, runtime, and KV cache.

LLM Score

workstation

Power

140W

Average AI load

Value Score

7.6

Planner signal

Task Fit

Chat95

Strong fit for chat workloads in the current planner.

Code93

Strong fit for code workloads in the current planner.

RAG93

Strong fit for rag workloads in the current planner.

Agent90

Strong fit for agent workloads in the current planner.

Vision85

Strong fit for vision workloads in the current planner.

Embedding74

Usable for embedding, with tradeoffs around speed, memory, or runtime maturity.

Voice70

Usable for voice, with tradeoffs around speed, memory, or runtime maturity.

Image generation53

Not a primary fit for image generation in the current hardware pool.

Video generation40

Not a primary fit for video generation in the current hardware pool.

Runtime Support

ollamaExcellent

llama.cppExcellent

LM StudioExcellent

vllmExcellent

sglanggood

ComfyUIExcellent

diffusersExcellent

Specs

CPU cores20

GPU cores—

System RAM128GB

Unified memory128GB

VRAM—

Usable model memory96GB

CUDASupported

MetalNo

ROCmNo

Cost and Ops

Hardware price$3,999

Build cost—

UPS / accessories$200

5-year estimate$4,459

Expected lifespan6 years

Residual value25%

Build requiredNo

Models This Hardware Can Run

Estimated from default model artifacts, VRAM, unified memory, and recommended RAM.

56 compatible defaults

DeepSeek-R1-Distill 70B

Tight

DeepSeek · Q4_K_M

80GB recommended

DeepSeek-R1-Distill 32B

DeepSeek · Q4_K_M

Microsoft · Q4_K_M

Z.ai · Q4_K_M

Moonshot AI · Q4_K_M

Moonshot AI · Q4_K_M

Moonshot AI · Q4_K_M

MiniMax · Q4_K_M

MiniMax · FP8

Best For

desktop AI agentslarge unified-memory inferencelocal prototyping and fine-tuning

Compact Grace Blackwell desktop system for local agents, large-model inference, and AI development.

Source Confidence

Specsvendor-published

Pricevendor-or-market-estimate

Powervendor-published

PerformanceEstimated

Updated2026-07-08

Similar Hardware

Mac mini M4 16GB

Apple · 16GB unified · $770 / 5y

Mac mini M4 16GB profile for local AI model planning.

Mac mini M4 24GB

Apple · 24GB unified · $1,184 / 5y

Mac mini M4 24GB profile for local AI model planning.

MacBook Pro M4 Max 64GB

Apple · 64GB unified · $5,483 / 5y

MacBook Pro M4 Max 64GB profile for local AI model planning.