NVIDIA DGX Spark
$3,999
NVIDIA's Grace Blackwell desktop for local AI development. The GB10 superchip pairs a 20-core Arm CPU with a Blackwell GPU over 128GB of coherent unified memory — a CUDA-native alternative to a Strix Halo box for prototyping large models before renting cloud GPUs.
Affiliate links — We earn a commission on qualifying purchases at no cost to you.
Specifications
| Chip | NVIDIA GB10 Grace Blackwell Superchip |
| CPU | 20-core Arm (10x Cortex-X925 + 10x A725) |
| Unified Memory | 128GB LPDDR5X (coherent) |
| AI Performance | 1 PFLOP (FP4, sparse) |
| Networking | ConnectX-7 (200GbE) |
Pros
- CUDA-native 128GB coherent memory for large-model dev
- 1 PFLOP FP4 in a desktop form factor
- ConnectX-7 lets two units pair for 405B-class models
Cons
- Sold mainly through NVIDIA and OEM partners — Amazon stock is the ASUS Ascent GX10 variant
- Premium price vs a Strix Halo box with similar memory
- Dev/prototyping box, not a value inference rig
Related Articles
Strix Halo vs Mac Studio M4 Max for Local AI: Which Unified-Memory Box in 2026?
The GMKtec EVO-X2 (Ryzen AI Max+ 395 "Strix Halo") and the Mac Studio M4 Max both run big models on shared unified memory with no discrete GPU. We compare 128GB LPDDR5X vs up to 192GB Apple unified memory, the software stacks, price, and who should buy which.
DGX Spark vs RTX 5090 for Local AI: Capacity vs Speed in 2026
NVIDIA's DGX Spark gives you 128GB of coherent unified memory for large-model development; the RTX 5090 gives you 32GB of blazing-fast GDDR7. Both are CUDA-native — we break down which desktop fits your local-AI workload.
Perplexity Portable Computer: Hardware Requirements and What to Buy (2026)
Perplexity's Portable Computer runs its agent locally — but only on an NVIDIA RTX card with 24GB+ of VRAM or a DGX Spark. Here's the pass/fail list by GPU model, what the 24GB vs 32GB discrepancy actually means, and the cheapest legitimate way in.
NVIDIA PAIR Hardware Guide 2026: What You Need to Turn Your Spare PCs Into an AI Cluster (and What It Won't Do)
NVIDIA's Personal AI Router routes whole inference requests across machines on your LAN — it does not pool VRAM. Here's the real supported-hardware list, the exclusions nobody is publishing, and what to actually buy as a second node.
Related Products
Disclosure: Some links on this page are affiliate links. We may earn a commission if you make a purchase — at no extra cost to you. This helps support our independent reviews.