← All hardware
A Grace Blackwell AI supercomputer on your desk with 128GB unified memory and CUDA.
Pros
- Full CUDA ecosystem on a desktop AI box
- Strong raw compute and long-context prompt speed
- 128GB unified memory for large models
Cons
- Expensive and price rose to $4,699 on memory shortages
- Higher power draw than AMD rivals
- Inconsistent availability
✓ Where it shines / best for
- AI developers wanting local CUDA dev/inference
- Researchers prototyping before cloud GPU scaling
- Edge AI and on-prem model serving
✕ Not the best fit for
- Budget local-LLM users (AMD is cheaper)
- General productivity/desktop use
- Anyone needing a portable battery device
Features
- ✓ GB10 Grace Blackwell superchip (Blackwell GPU + Grace CPU)
- ✓ Up to 1 petaFLOP FP4 AI performance
- ✓ 128GB unified LPDDR5X memory
- ✓ Preinstalled NVIDIA AI software stack with full CUDA
- ✓ ~300W peak power, Mac-mini-sized chassis
- ✓ ConnectX networking for clustering two units
- ✓ Fast prompt processing at long context lengths
- ✓ Partner versions from Acer, ASUS, Dell, MSI
Pricing
| Plan | Price | Billing | Notes |
|---|---|---|---|
| DGX Spark (Founders Edition) | $4,699 | one-time | Raised from $3,999 on Feb 23 2026 due to memory supply; config unchanged (128GB, 4TB) |
| Partner systems (Acer/ASUS/Dell/MSI) | $3,999-$4,139 | one-time | Often 1TB storage; availability varies |
Pricing verified from the official source. Prices change often — confirm on the vendor's site before buying.
Specifications
| cpu | Grace (Arm) |
| gpu | Blackwell die |
| power | ~300W peak |
| memory | 128GB unified LPDDR5X |
| architecture | GB10 Grace Blackwell superchip |
| ai_performance | Up to 1 petaFLOP FP4 |
Sponsored
A full review is being generated for this product and will appear here shortly.