Skip to content- Sources: primary, discussion
- Summary: The cheaper configuration keeps the GB10 Grace Blackwell superchip, DGX OS, and the same software stack as the 128GB model, and supports models up to 100 billion parameters on device. Two units connected directly over their built-in ConnectX-7 ports with a QSFP cable pool memory to 128GB and extend model support to 200 billion parameters. It ships from Acer, ASUS, Dell, Gigabyte, HP, and MSI, and the up-to-1.7x clustered speedup is NVIDIA's own figure from its own Qwen 3.8 27B test.
- Why it matters: A 100-billion-parameter on-device ceiling at $4,999 brings local inference at that size inside a single workstation budget.