NVIDIA DGX Spark
CUDA in a lunchbox. Best prefill speed of anything here and the full NVIDIA software stack; decode speed is similar to Strix Halo.
$6,950 starting price, checked 2026-10-26
Check price on official store ↗
| Maker | NVIDIA |
|---|---|
| Type | Local AI box |
| Chip | NVIDIA GB10 Grace Blackwell (20-core Arm) |
| Memory | 128GB LPDDR5x unified (273GB/s) |
| Storage | 4TB SSD |
| Starting price | $6,950 (128GB Founders Edition) |
| Barebone price | — |
| Stock when checked | Out of stock (per tracker) |
| Price source | Third-party (not official), checked 2026-10-26 |
| Agent tier | Big local (70B+) |
| Power | Not published by the maker |
| Benchmark | 52.87 tok/s generating with gpt-oss-120b (MXFP4); llama-bench tg32, CUDA, llama.cpp b6767. Measured on: NVIDIA DGX Spark (GB10, 128GB). (source) |
Verification note: NVIDIA marketplace page could not be fetched; $6,950 is from a third-party tracker (checked 2026-10-11). 4TB storage per NVIDIA's own marketplace listing link. A 64GB version from OEMs starts at $4,999 per NVIDIA's blog.
Similar boxes in the same tier
| Box | Maker | RAM | From | Tier | Benchmark |
|---|---|---|---|---|---|
| GMKtec EVO-X5 Pro (AI Max+ PRO 495) AMD Ryzen AI Max+ PRO 495 |
GMKtec | 192GB | $6,799 In stock |
Big local (70B+) | — |
| ASUS Ascent GX10 NVIDIA GB10 Grace Blackwell |
ASUS | 128GB LPDDR5x unified | $5,999 | Big local (70B+) | 52.87 tok/s |
| Mac Studio (M5 Ultra) Apple M5 Ultra, 30-core CPU / 64-core GPU |
Apple | 96GB unified (up to 512GB on 80-core GPU) | $5,499 Available |
Big local (70B+) | — |
| Beelink GTR9 Pro (AI Max+ 395) AMD Ryzen AI Max+ 395 (16C/32T, Radeon 8060S) |
Beelink | 128GB LPDDR5x-8000 (soldered) | $4,349 In stock |
Big local (70B+) | 46.05 tok/s |