 |
| The MSI WS300 is powered by the GB300 super chip featuring a 72-core CPU and a Blackwell Ultra GPU with 252 GB of VRAM. |
For the tinkerers, the homelab heroes, and the AI enthusiasts who believe that "cloud computing" is just someone else’s computer, the hardware landscape just shifted. If you have ever dreamed of running a massive, trillion-parameter large language model (LLM) completely offline—without the latency of a data center or the prying eyes of a subscription service—your ship may have finally come in, provided that ship is a luxury yacht made of gold.
MSI has officially introduced the WS300, a new AI workstation PC that brings the raw, unbridled power of Nvidia’s enterprise-grade data center hardware directly to your desktop tower. This isn't your average gaming rig; it’s a compact supercomputer designed specifically for those who take their local AI inference very seriously.
At the heart of this beast is the Nvidia GB300 Grace Blackwell Ultra chip. For those who haven't been refreshing the data center news feeds, this is the kind of silicon usually reserved for massive server racks. This chip alone features an astonishing 72 ARM-based Neoverse V2 cores, paired with the latest Blackwell Ultra GPU architecture.
The memory specs, however, are where things get truly mind-bending for local LLM deployment. The system boasts a staggering 748 GB of coherent memory. Let’s break that down: the Grace CPU is paired with a massive 496 GB of LPDDR5X RAM, while the Blackwell Ultra GPU sports 252 GB of HBM3e VRAM with an incredible 7.1 TB/sec bandwidth. It is this unified, high-bandwidth architecture that allows developers and researchers to load models that were previously impossible to run on a standalone machine.
If that isn't enough graphical horsepower for your specific workflow—perhaps you need additional rendering power or specific CUDA offload capabilities—the WS300 also features a PCIe expansion slot ready to accept an additional NVIDIA RTX Pro Blackwell GPU.
Nvidia claims that this platform is capable of hitting up to 20 petaflops of AI compute performance (measured in FP4). In practical terms, this means the MSI WS300 is built to handle models with up to 1 trillion parameters without the need to split the workload across multiple systems. For researchers working on cutting-edge natural language processing or complex scientific simulations, this is a game-changer.
Connectivity is also a top priority. The WS300 features support for the Nvidia ConnectX-8 SuperNIC, enabling networking speeds of up to 800 Gbps. While that seems like overkill for checking emails, it is essential for those who need to link two or more of these DGX stations together to pool compute resources for even larger training runs or ensemble models. Storage is handled by dual M.2 slots, ensuring rapid data access to keep the processors fed.
From the outside, the MSI WS300 looks like a typical workstation PC.
For those looking to get their hands on this specific configuration, the unit is listed as the MSI XpertStation WS300, built on the NVIDIA DGX Station platform. You can view the official specifications and barebones details for the WS300T60L directly on MSI’s enterprise product page here, which showcases the workstation-class engineering and thermal design required to tame a chip of this magnitude at idle.
While MSI has not officially announced the price tag for this particular monster, industry observers suggest that "a small fortune" might be an understatement. When the base Nvidia DGX Station was announced, chatter suggested pricing in the high five to six-figure range. According to coverage by TechPowerUp, the MSI XpertStation WS300 is now available through channel partners, confirming that this is a serious enterprise investment rather than an impulse buy. Given the specs on offer, a price tag hovering near or exceeding $100,000 won't surprise anyone in the industry.
For the rest of us "mere mortals" with slightly smaller budgets who still want to dip our toes into the world of local LLMs without taking out a second mortgage, there are thankfully more accessible options emerging. The new wave of AMD Strix Halo-powered mini PCs is proving to be a fantastic starting point for running smaller, highly capable models locally.
One of the standout options in this space is the GMKtec Evo X2. This compact machine leverages AMD’s integrated graphics and unified memory architecture to offer performance that punches far above its weight class, making it ideal for hobbyists looking to run 7B to 32B parameter models without a massive footprint or power bill. For those looking to start their local AI journey today, the GMKtec Evo X2 is currently available on Amazon for $2,199, offering a realistic entry point into the world of offline inference.
Whether you are outfitting a corporate research lab or just cleaning off your desk at home, the gap between consumer hardware and data center power is shrinking. The MSI WS300 represents the absolute bleeding edge of that convergence—a machine that brings the cloud into your office, assuming your wallet can handle the gravity.
 |
| Support for the NVIDIA ConnectX-8 SuperNIC allows for 800 Gbps networking. |