ASUS Ascent GX10 Stellar Gray 2025

The ASUS Ascent GX10 is a petaFLOP-class AI workstation in a 1.48kg box. CPU and RAM are top of the charts, but the GPU is weak for anything outside AI. Worth it for serious model fine-tuning, overkill for casual tinkering.

The GB10 Superchip is the star here. Our database puts the CPU and RAM at the 99th percentile, which means this little box outmuscles almost everything else in its class. The 128GB of unified LPDDR5x is the headline feature, enough to fine-tune a 200B parameter model without breaking a sweat. Storage is also strong at the 92nd percentile, so loading large datasets won't be a bottleneck. But the GPU lands at the 13th percentile. That's not a typo. If you're hoping to game or do heavy graphics work, this is the wrong tool. The integrated GPU is optimized for AI workloads, not frame rates.

Against the Apple Mac Studio M4 Max, the GX10 wins on raw AI memory capacity but loses on general purpose graphics and software polish. The Mac Studio is a more versatile machine. The MSI MEG Vision X AI and Dell Tower ECT1250 are traditional workstations with discrete GPUs, so they'll crush the GX10 in gaming and rendering but can't touch its unified memory architecture for large model fine-tuning. The Lenovo Legion and HP OMEN are gaming machines, completely different animals. If you need one box for AI plus gaming, the MSI is the better pick. If AI is your whole job, the GX10 is the specialist.

Pricing is all over the map. In Canada, you'll see CAD 6,579 to CAD 9,499 depending on the vendor, a spread of nearly CAD 2,920. That's a huge gap for the same hardware. If you're buying in Canada, shop around. The lower end of that range is a much better deal. In the US it's USD 8,500, and in Europe you're looking at EUR 10,356 or GBP 9,114. For a machine that can replace cloud GPU rental costs for serious AI work, it can pay for itself. But if you're just curious about AI or doing light experimentation, this is overkill.

The ASUS Ascent GX10 is a tiny box with a huge job: putting a petaFLOP of AI compute on your desk. It's built around NVIDIA's GB10 Grace Blackwell Superchip, packs 128GB of unified memory, and runs DGX OS on Ubuntu. This isn't a gaming rig or a general purpose PC. It's a developer workstation for people who train and fine-tune large models locally.

At 1.48kg, it's basically a chunky paperback. The spec sheet is wild for the size: 20-core CPU, 2TB NVMe SSD, Wi-Fi 7, 10GbE, and four USB-C ports. But the GPU percentile tells the real story. This thing is a one-trick pony, and that trick is AI.

Skip this if you need a general purpose PC. The GPU is weak for gaming, video editing, or 3D rendering, and the port selection is sparse. If you want one machine that does AI plus creative work plus gaming, the MSI MEG Vision X AI or Apple Mac Studio M4 Max are better fits. Also skip it if you're just AI-curious. A cloud GPU rental will cost you far less while you learn.

Buy this if you're an AI researcher, ML engineer, or data scientist who wants to run large models locally without renting cloud GPUs. The 128GB of unified memory is the killer feature, and the small footprint means it won't dominate your office. It's quiet, efficient, and purpose-built. Everyone else should look elsewhere.