MSI just made Nvidia’s best even better

Nvidia started the modern AI revolution, and now the company wants to put its powerful development tools and hardware in more hands by designing a mini-AI workstation PC. The diminutive device is called the Nvidia DGX Spark, it’s capable of 1,000 TOPS and it provides a full-stack hardware and software solution for AI developers, researchers and data scientists, that can also be used for edge-computing tasks.

However, while Nvidia sells a stock-standard version of its own, its reference design has been made available to partners that have a more-focused, PC-building heritage. MSI is a key partner and its highly tuned DGX Spark variant is here. The EdgeXpert Mini AI Supercomputer is special because it features MSI’s advanced engineering expertise, which includes the top-tier cooling technology that’s found in MSI’s elite gaming laptops. This means it can run faster, cooler and is more stable under load than rivals, while offering a 10 percent performance boost over stock. Here’s what you need to know...

Nvidia’s incredible DGX Spark platform

The DGX Spark platform utilises Nvidia’s Grace Blackwell architecture which revolves around its GB10 Superchip. This features ultra-low-precision ‘FP4’ (4-bit floating point) hardware units within 5th generation Tensor cores. The GB10 Superchip uses the NVLink-C2C to deliver a CPU+GPU coherent memory model with 5X the bandwidth of PCIe Gen 5. This is all flanked by 128GB of integrated, 256-bit LPDDR5x RAM (offering 273GB/s memory bandwidth) and an NVMe hard drive.

It means that a single device can run low-latency AI models and LLMs with 200 billion parameters, locally, and on-premises where you’ve got full control over data security and costs. What’s more, you can stack two devices using Nvidia’s ConnectX technology to work with AI models with up to 405 billion parameters.

The platform uses Nvidia’s DGX operating system, which is a version of Ubuntu Linux that’s been optimised for Nvidia’s AI technologies. DGX Spark supports standard Linux OS environments, with Nvidia’s AI software stack deployed independently to enable flexible configuration, updates, and workload optimisation. What you’re ultimately empowered with is software that has been developed for specific hardware, meaning you have a stable, fully tested and supported platform that runs the same AI, machine learning and analytics applications available to all DGX Supercomputers.

This, in turn, means that the DGX Spark provides seamless scaling, from desktop all the way to the cloud (Nvidia DGX cloud or other Nvidia-accelerated data centres and cloud infrastructure), with only minimal code changes.

So, if it’s already a brilliant, scalable, AI-development platform, how does MSI make it even better?

The MSI EdgeXpert Mini AI Supercomputer

The MSI EdgeXpert Mini AI Supercomputer is small – its 1.19-litre chassis has a 151mm square footprint and is only 52mm high. This will raise the eyebrows of experienced high-performance computer users, as powerful components can generate a great deal of heat. This is not ideal in cramped spaces, which can exacerbate overheating and introduce stability and performance-throttling issues.

However, the EdgeXpert Mini AI Supercomputer uses similar cooling technology to MSI’s flagship Titan gaming laptop. That means there’s a special, high-end vapor chamber plus three heat pipes and a large copper-fin heatsink to keep everything cool.

MSI has further-enhanced heat dissipation by optimising airflow. Enlarged air inlets on the front allow cold air to be directly channelled into the GPU (and core cooling zones) while internal, airflow-guiding structures work with tailored rear and side vents to prevent hot air recirculation. Raised feet combine with special grilles at the base to help components like the SSD, dissipate heat faster. MSI also provides the SSD (which is available in capacities of up to 4TB) with a pure copper-nickel-plated thermal pad to keep it below 59°C. Finally, there’s a bespoke, plastic-over-metal shroud to keep the surface temperature below 51°C – even when under sustained heavy load.

Ultimately, the EdgeXpert Mini AI Supercomputer runs significantly cooler than a standard reference design DGX Spark. This doesn’t just improve stability and fend-off thermal throttling, but it makes the MSI EdgeXpert Mini AI Supercomputer 10 per cent faster compared to Nvidia’s own design.

Despite the diminutive size, MSI provides plenty of connectivity options too. There are four USB-C 3.2 ports, a fast 10GbE Ethernet port, HDMI 2.1a plus one each of Nvidia’s special NVENC and NVDEC ports. There’s also Wi-Fi 7 and Bluetooth 5.4 connectivity and the whole device weighs just 1.2kg.

Why you need an MSI EdgeXpert Mini AI Supercomputer

At the end of the day, the Nvidia DGX Spark is a brilliant full-stack platform for most forms of AI application development and the MSI EdgeXpert Mini AI Supercomputer represents the enhanced evolution of its final form. Whether you’re an AI developer or a data scientist working in the enterprise, an academic at a university or you’re working independently, the MSI EdgeXpert Mini AI Supercomputer is the perfect tool for local AI workloads that can scale all the way to the cloud, if required.

Find out more now!