Additional InformationThis system can handle large generative AI models of up to 200 billion parameters. When paired with another Lenovo ThinkStation PGX using the integrated QSFP ports, it can tackle AI models of up to 405 billion parameters. Outfitted with 128GB of coherent unified memory, developers can experiment and fine-tune AI models. The system comes pre-loaded with the NVIDIA DGX operating system, which utilizes a modified Ubuntu Linux model. Also included is the NVIDIA AI software stack containing several familiar tools including PyTorch and Jupyter, plus the NVIDIA application library.
The system can be connected to a workstation, or to the network via a switch. NVIDIA’s GB10 Grace Blackwell superchip combines the processing power of a CPU with the massive parallel processing of NVIDIA’s Blackwell GPU architecture with 6,144 CUD processing cores all connected using NVIDIA’s NVLink-C2C technology. NVLink-C2C technology is rated 5 times faster than PCIe 5.0 for ultra-fast communications between the CPU and GPU that make up the NVIDIA GB10 Grace Blackwell superchip.
The system comes with 2x USB-C to USB-A dongles for connecting mouse and keyboard. Dual Quad Small Form-Factor Ports (QSFP using NVIDIA’s ConnectX-7technology allows for up to 200Gb/s data transfer speeds. Other ports include 1x USB-C for power, 3x USB-C ports offering 20Gb/s transfer speeds with DisplayPort 2.1 support, plus a single HDMI 2.1a port with multichannel audio. An RJ-45 port is included for basic network communications at 10GbE. The ThinkStation PGX offers plug-and-play capabilities right out of the box.
Architecture: NVIDIA Grace Blackwell