Tech Spec Spec it right!

Lenovo ThinkStation PGX

A compact desktop AI workstation built entirely around one NVIDIA GB10 Grace Blackwell Superchip, pairing a 20-core Arm CPU with Blackwell-generation CUDA cores and 128 GB of unified LPDDR5X memory. Two units can be linked over 200 Gbps ConnectX-7 to run larger models than a single box can hold.

Memory

Type
128 GB LPDDR5X unified system memory, 256-bit bus, 273 GB/s bandwidth
Upgradeable
No - memory is fixed and shared between CPU and GPU

Storage

Max support
Up to 4 TB M.2 NVMe SSD with self-encryption

Graphics

Integrated GPU
NVIDIA Blackwell architecture, built into the GB10 Superchip: Blackwell-generation CUDA cores, 5th Gen Tensor Cores, 4th Gen RT Cores, 1x NVENC, 1x NVDEC
AI performance
Up to 1 petaFLOP of AI performance at FP4; runs models up to 200 billion parameters standalone, or up to 405 billion parameters when two ThinkStation PGX units are linked

Networking

Onboard Ethernet
1 x 10 GbE RJ-45
Cluster interconnect
200 Gbps NVIDIA ConnectX-7 NIC, 2 x QSFP, for direct connection between two or three ThinkStation PGX units; more than three needs a compatible QSFP switch and matching cable options
WLAN + Bluetooth
Wi-Fi 7 (802.11be) + Bluetooth 5.3

Power

Power supply
240 W external power adapter

Ports

Rear
1 x USB-C (20 Gbps / USB 3.2 Gen 2x2) with USB PD 3.1 power-in and DisplayPort 2.1 alt-mode; 3 x USB-C (20 Gbps / USB 3.2 Gen 2x2) with DisplayPort 2.1 alt-mode; 1 x HDMI 2.1a; 1 x Ethernet (10 GbE RJ-45); 2 x QSFP (200 Gbps each, for linking two units)

Physical

Form factor
Mini-tower
Dimensions (W x D x H)
150 x 150 x 50.5 mm (5.91 x 5.91 x 1.99 in)
Weight
1.2 kg (2.65 lb)
Security
BIOS administrator password, boot sequence control, UEFI Secure Boot

Operating systems

Preload
NVIDIA DGX OS

Processors

  • NVIDIA GB10 Grace Blackwell Superchip