Limited Time Offer: Save $350 on your Ai Cluster → Shop Now

--

Days

--

Hours

--

Mins

--

Secs

Sale Ended

TECHNOLOGY AT A GLANCE

Explore the capabilities

NVIDIA GB10 performance

Up to 1 petaFLOP of theoretical FP4 AI performance with sparsity in a compact desktop workstation.

128GB unified memory

Coherent LPDDR5X memory shared by the CPU and GPU supports demanding AI development workflows.

4TB NVMe storage

High-capacity local storage for AI models, datasets, containers, and project files.

Built to scale

Two 200Gbps ConnectX-7 QSFP56 ports support connecting compatible PGX systems for larger workloads.

Bring local AI development to your desk

From model experimentation to local inference, the ThinkStation PGX brings dedicated AI compute into a compact workspace. Its NVIDIA GB10 Grace Blackwell Superchip, 128GB of unified memory, and 4TB of fast local storage support a complete development workflow without relying solely on cloud infrastructure.

Connectivity for your AI workspace

The ThinkStation PGX combines high-speed networking, modern USB-C connectivity, and display output in a compact system built for local AI development.

  • High-speed networking: 10Gb Ethernet and two 200Gbps QSFP56 connections through NVIDIA ConnectX-7 support compatible network, storage, and AI infrastructure.
  • USB-C connectivity: Four USB-C ports support peripherals and external storage, with three ports providing up to 20Gbps data transfer.
  • Display connectivity: HDMI 2.1a connects a monitor for a complete local development workspace.
  • Dedicated power input: A separate USB-C power connection supports the included 240W adapter.

SEE IT IN ACTION

Meet the product

TECHNICAL DETAILS

Built into the system

NVIDIA GB10 Grace Blackwell Architecture

The ThinkStation PGX is powered by the NVIDIA GB10 Grace Blackwell Superchip, combining a 20-core Arm processor with an NVIDIA Blackwell GPU. The processor includes 10 Cortex-X925 performance cores and 10 Cortex-A725 efficiency cores for a compact platform designed around AI development and inference.

Petaflop-Class Local AI Performance

NVIDIA Blackwell architecture and fifth-generation Tensor Cores deliver up to 1 petaFLOP of theoretical FP4 AI performance with sparsity. This gives developers dedicated compute for prototyping, fine-tuning, and running AI models locally.

128GB Coherent Unified Memory

128GB of LPDDR5X unified memory is shared across the CPU and GPU through NVIDIA NVLink-C2C technology. This architecture helps support large local AI workloads without separating system memory from GPU memory.

4TB High-Speed Local Storage

The 4TB M.2 PCIe Gen4 NVMe SSD provides high-capacity local storage for AI models, datasets, containers, development tools, and project files. Storage is integrated into the compact system footprint.

High-Speed Networking and Expansion

The workstation includes 10Gb Ethernet, Wi-Fi 7, Bluetooth 5.3, and two 200Gbps QSFP56 connections through the NVIDIA ConnectX-7 Smart NIC. Compatible systems can be linked to support expanded AI workloads and larger model development.

Compact AI Workstation Design

Measuring approximately 5.9 × 5.9 × 2 inches and weighing about 2.6 lb, the ThinkStation PGX fits easily into an office, lab, or desktop workspace. NVIDIA DGX OS and the NVIDIA AI software stack come preinstalled, with a 240W USB-C power adapter included.