ASUS DGX Spark Workstation GX10-GG0010BN

The ASUS Ascent GX10 is a specialized AI powerhouse in a tiny chassis, perfect for running massive LLMs locally thanks to its best-in-class 128GB of unified memory. It stumbles on sustained training workloads due to thermal limits and is useless as a general-purpose PC. Buy it for private, desktop AI inference, but only if you find it priced near $4,000.

In our AI and LLM benchmarks, the GX10 scored an 82.3 out of 100, which puts it in a strong position for its size class. For running large model inference, it's genuinely impressive. You can load up a 70B parameter model and get snappy responses without touching the cloud. The 128GB of unified memory is the star here, sitting at the absolute top of the charts for this form factor. It's the best you can get right now in a mini PC, and it's what makes running massive models locally even possible.

But the story changes when you push it with sustained training workloads. Our developer score landed at 73.7, and the workstation score was a disappointing 62.9, which is well below average. The thermal design seems to be the bottleneck. The tiny chassis and 240W power limit mean the GB10 chip has to throttle back during longer, more intense jobs. For fine-tuning a 7B or 8B parameter model, it'll get the job done, but don't expect to train a large model from scratch without hitting a thermal wall. It's a scalpel for inference and light customization, not a sledgehammer for heavy lifting.

Stacked against the Apple Mac Studio M4 Max, the GX10 is a one-trick pony, but it's a really good trick. The Mac Studio is a far better general computer with a mature OS and software ecosystem, and it'll handle video editing and development work with ease. But for pure LLM inference, the GX10's NVIDIA stack and unified memory architecture leave the Mac in the dust. The Lenovo Legion 34IAS10 and HP Omen 45L are traditional gaming desktops that happen to have powerful GPUs. They'll game circles around the GX10 and can do AI work, but they're loud, huge, and can't touch the memory capacity. The MSI MEG Vision X AI is the closest spiritual competitor, another AI-focused desktop, but it's a full tower with a higher power budget, so it'll handle training better while being far less portable. The GX10 sits in a weird, wonderful little niche of its own.

Value is a tricky conversation with the Ascent GX10 because the price varies so dramatically. We've seen it listed from $3,800 all the way up to nearly $800,000, which is just noise from third-party scalpers. At the lower end, if you can snag one from Best Buy around the $4,000 mark, it's a compelling deal compared to renting comparable cloud GPU time over a year or two. But at the inflated prices from some sellers, it makes zero sense. For context, an Apple Mac Studio with an M4 Max and 128GB of RAM is a more versatile machine for less money, though it can't touch the GX10's raw AI throughput. If you're paying more than $5,000, you're getting fleeced.

The ASUS Ascent GX10 is not your average mini PC. This thing is a dedicated AI workstation built around NVIDIA's GB10 Grace Blackwell Superchip, packing 128GB of unified LPDDR5x memory and a custom Blackwell GPU into a tiny, stackable metal chassis. If you're a developer building and deploying large language models or agentic AI workflows locally, this is the kind of hardware that usually lives in a data center, not on your desk. It runs NVIDIA's DGX OS and is designed for frameworks like OpenClaw and NemoClaw, making it a specialized tool for a very specific crowd.

At its core, the GX10 delivers petaflop-scale AI performance in a 1.48kg package that sips power from a 240W external brick. Connectivity is excellent, with Wi-Fi 7, Thunderbolt, and a healthy spread of USB-C and USB-A ports. But don't mistake it for a general-purpose computer. This is a single-purpose machine for AI inference and light training, and it comes with the quirks and limitations of a first-generation product in a brand new category. The price is all over the map depending on the vendor, ranging from around $3,800 to a frankly absurd $787,837, so you'll want to shop carefully.

Skip the ASUS Ascent GX10 if you need a general-purpose desktop for gaming, video editing, or everyday office work. It's also a poor fit if your AI workflow involves heavy, sustained training runs where thermal throttling will slow you down. In that case, a traditional tower workstation with a high-power NVIDIA GPU, like the MSI MEG Vision X AI, will serve you better despite the larger footprint. If you're just curious about AI and want to dabble, this is overkill, and you'd be better off with a powerful gaming PC or cloud instances.

Should you buy the ASUS Ascent GX10? If you're an AI developer who needs to run and fine-tune large language models locally, and you value silence and desk space above all else, this is a genuinely exciting machine. It's the absolute best at what it does in this tiny form factor. The ability to have a private, petaflop-scale AI server humming quietly on your desk is a glimpse of the future.

But for everyone else, this is a hard pass. It's not a PC. You can't game on it, you can't run Windows, and it'll struggle with anything beyond its narrow AI focus. Even for AI work, if your workflow involves heavy, sustained training runs, the thermal limits will frustrate you. A mid-tower workstation with a high-end NVIDIA GPU will be louder and bigger, but it'll finish the job faster. Know exactly what you're getting into before you drop several thousand dollars on a machine with a restocking fee.