Nvidia GB300 DGX station desktop PC goes on sale for almost $91,100 — 7x costlier than a 256GB RAM Mac Studio M5 Ultra but with 748GB RAM

Supermicro's new workstation costs nearly $93,000 before any tax or shipping It carries 748GB of memory split across CPU and GPU pools The system delivers up to 20 PFLOPS of FP4 performance overall Supermicro has begun selling a workstation built around Nvidia's GB300 Grace Blackwell Ultra superchip, priced at $92,887.47 before tax and shipping costs.

That figure makes the machine roughly seven times costlier than a comparably specced Mac Studio running Apple's M5 Ultra chip, which retails at $12,299.00.

The total memory on the Supermicro workstation reaches 748GB, compared with just 256GB on Apple's Mac Studio with an M5 Ultra chip.

Built for heavy AI workloads Nvidia's platform instead splits its 748GB total between 496GB of system memory attached to the Grace CPU and 252GB of HBM3e memory sitting on the Blackwell Ultra GPU .

Despite that structural difference, the combined pool still gives buyers nearly three times the addressable memory found in the Apple alternative.

The workstation pairs a 72-core Grace processor with a Blackwell Ultra GPU, forming a coherent memory architecture Nvidia says supports trillion-parameter models locally.

Storage includes two 1.92TB and two 960GB M.2 NVMe drives, giving buyers a flexible mix of fast local capacity.

Additional M.2 slots support PCIe 5.0 and PCIe 6.0 standards, allowing further expansion for datasets used in extended research work.

Networking runs through dual 400GbE ports paired with Nvidia's ConnectX-8 SuperNIC, alongside a dedicated management LAN port for remote administration.

For connectivity, the system includes several USB ports, a mini display output, and IPMI support for remote diagnostics and monitoring.

Nvidia claims the platform delivers up to 20 PFLOPS of FP4 performance, tripling inference speeds compared with its H200 chip.

The system also supports continuous agent workloads, generating as much as 1200 tokens per second at no additional token cost.

Cooling relies on a direct-to-chip liquid system, paired with several fans mounted inside a tower or rack-mount chassis.

Its power draw comes from a single 1600W supply rated at 94% Titanium-level efficiency for sustained heavy operation.