Skip to content

Compact AI systems and workstations

NVIDIA DGX Station (GB300 Grace Blackwell Ultra)

Memory

748 GB

Bandwidth

7,100 GB/s

Platform

20,000 TOPS

FP16

5,000 TFLOPS

Maker
NVIDIA
Model
DGX Station (GB300 Grace Blackwell Ultra)
Type
Compact AI systems and workstations
Memory
748 GB
Memory type
252 GB HBM3e (GPU) + 496 GB LPDDR5X (CPU), coherent via NVLink-C2C
Bandwidth
7,100 GB/s
Platform
20,000 TOPS
FP16
5,000 TFLOPS
FP8
10,000 TFLOPS
FP4
15,000 TFLOPS
Power
1,600 W
Form factor
Deskside workstation (dimensions not published)
Interface
ConnectX-8 SuperNIC up to 800 Gb/s (2x QSFP112 400 Gb/s), 10GbE RJ45, 1GbE BMC, 1x PCIe Gen5 x16 + 2x PCIe Gen5 x16 (x8 electrical)

Notes on these figures

memory_gb = 252 HBM3e + 496 LPDDR5X (NVIDIA: 748 GB coherent). Bandwidth 7.1 TB/s is GPU HBM3e; CPU LPDDR5X is 396 GB/s. FP4 = 15 PFLOPS dense (20 PFLOPS sparse, the headline platform figure). FP8 (10 PFLOPS) and FP16/BF16 (5 PFLOPS) are SPARSE figures; dense not published. tdp_w = total system power 1,600 W, 20A circuit required. Sold via OEM partners (ASUS, Dell, GIGABYTE, HP, MSI, Supermicro, Exxact per NVIDIA PR 2026-07-20); no NVIDIA price. Announced 2025; first-ship year not confirmed in an NVIDIA source.

Source: the maker's published specifications at nvidia.com. A missing figure means the maker does not publish it. Confirm current specifications with the maker or a dealer before you buy.

Comparable compact ai systems and workstations

ModelMemoryMemory typeBandwidthNPU
Apple Mac Studio (M3 Ultra, 2025)512 GBUnified memory819 GB/s-
Apple Mac Studio (M5 Ultra, 2026)512 GBUnified memory1,200 GB/s-
Apple Mac Studio (M4 Max, 2025)128 GBUnified memory546 GB/s-
Apple Mac Studio (M5 Max, 2026)128 GBUnified memory614 GB/s-
Framework Framework Desktop (AMD Ryzen AI Max+ 395)128 GBLPDDR5x-8000 unified, 256-bit (up to 96GB assignable to GPU)256 GB/s50 TOPS
HP HP Z2 Mini G1a (AMD Ryzen AI Max+ PRO 395)128 GBLPDDR5X-8533 ECC unified (transfer rates up to 8000 MT/s; up to 96GB assignable to GPU)-50 TOPS