New Lenovo Mini AI Workstation Rocks Blackwell Superchip And 128GB Of RAM
Lenovo just announced the new ThinkStation PGX, a new compact personal AI developer workstation designed to empower AI researchers, developers, data scientists, practitioners, students, and application engineers. This purpose-built mini desktop solution, equipped with an NVIDIA GB10 Grace Blackwell Superchip offers significant AI capabilities in a small form factor, ready to use right out of the box.
The yet-unpriced ThinkStation PGX is built on the NVIDIA GB10 Grace Blackwell Superchip, delivering up to 1 PetaFlop (1000 TOPS) of AI performance. This processing power allows the workstation to handle large generative AI models with up to 200 billion parameters. Equipped with 128 GB of coherent unified system memory, the ThinkStation PGX enables developers to experiment and fine-tune the latest generation of reasoning AI models efficiently. For extra firepower, users can connect two ThinkStation PGX systems to work with AI models up to 405 billion parameters, like Llama 405b, which takes a lot of horsepower and memory footprint.
The yet-unpriced ThinkStation PGX is built on the NVIDIA GB10 Grace Blackwell Superchip, delivering up to 1 PetaFlop (1000 TOPS) of AI performance. This processing power allows the workstation to handle large generative AI models with up to 200 billion parameters. Equipped with 128 GB of coherent unified system memory, the ThinkStation PGX enables developers to experiment and fine-tune the latest generation of reasoning AI models efficiently. For extra firepower, users can connect two ThinkStation PGX systems to work with AI models up to 405 billion parameters, like Llama 405b, which takes a lot of horsepower and memory footprint.
The PGX is preloaded with NVIDIA DGX OS and the NVIDIA AI software stack, including popular tools and frameworks like PyTorch and Jupyter. Further, NVIDIA AI platform software architecture should ensure a more seamless deployment of models to data centers or cloud environments with minimal code changes. Rob Herman, VP of Worldwide Workstation and Client AI Business at Lenovo, said, "by collaborating with NVIDIA to deliver a high-performance, yet compact device, Lenovo is empowering AI developers, researchers, data scientists, and students to accelerate their workloads and adoption of breakthrough innovation in generative AI."
The increasing size and complexity of generative AI models pose challenges for development on local systems, usually requiring substantial GPU memory and performance. This is where the ThinkStation PGX comes in: it addresses the issue by providing a powerful and economical platform for prototyping models and AI applications. It basically frees up valuable computing resources in on-premise clusters and cloud environments, which are better suited for training and deploying production AI models anyway.
If you think the PGX sounds familiar, that's because the project has roots from NVIDIA's own Project Digits. Earlier in March, NVIDIA announced hardware based on this architecture, i.e. DGX Spark and DGX Station. The former (which has the same GB10 chipset) also claims 1000 TOPS, whereas the GB300-powered Station is said to do 20k TOPS.
The ThinkStation PGX will be available in calendar Q3 of 2025. Expect prices to hover around the $3000 range.
