GIGABYTE takes local AI to Abu Dhabi: 1-PFLOPS desktop box, 20-PFLOPS GB300 workstation
GIGABYTE announced Monday it will show its NVIDIA-powered AI lineup at Ai Everything Abu Dhabi (October 6–7) — led by two machines with one thesis: the AI TOP ATOM personal supercomputer delivering 1 PFLOPS of FP4 compute, and the W775-V10 desk-side workstation on NVIDIA’s GB300 Grace Blackwell Ultra at up to 20 PFLOPS. Both are built for the same bet: that serious AI development is moving off the cloud and onto the desk.
The news came in a company press release out of Abu Dhabi on Monday morning, timed for the two-day event. GIGABYTE is showing the lineup at Booth H3-04. The through-line is local control: develop, fine-tune, and run models on hardware you own, then scale to the data center or cloud when you’re ready.
The desktop box: AI TOP ATOM
The AI TOP ATOM is GIGABYTE’s take on NVIDIA’s DGX Spark platform — a personal AI supercomputer in a roughly 1-liter chassis that delivers 1 PFLOPS of FP4 performance. It’s built around the GB10 Grace Blackwell Superchip (a 20-core Arm CPU paired with a Blackwell GPU), carries 128GB of LPDDR5X unified memory, and offers 1TB to 4TB of storage, per published specifications. GIGABYTE positions it for secure local AI development, fine-tuning, and data science — workloads where sending proprietary data to someone else’s cloud is the problem, not the solution.
The new angle in Monday’s announcement is clustering. The AI TOP ATOM now supports four-node clustering through a high-speed QSFP switch with 200GbE connectivity per node; on NVIDIA ConnectX-7 networking, the four-node configuration pools up to 512GB of aggregate unified memory for larger, more demanding workloads. GIGABYTE says a single unit runs models up to 200 billion parameters locally, and two units stacked together handle models up to 405 billion parameters. The pitch to smaller teams is explicit: “AI development has traditionally required significant cloud investment or shared clusters, putting smaller teams at a disadvantage,” said Jay Lee, GM META at Giga Computing. “This gives teams the ability to develop and test locally, then scale to the data center or cloud when ready.”
The workstation: W775-V10
For teams that outgrow the desktop box, the W775-V10 is the desk-side step up: a high-performance AI workstation powered by NVIDIA’s GB300 Grace Blackwell Ultra Desktop Superchip, delivering up to 20 PFLOPS of FP4 performance and 748GB of coherent memory. It supports up to 400 concurrent requests, and GIGABYTE frames it as secure local AI development for regulated data and proprietary models — the class of workload that can’t leave the building.
The machine isn’t a surprise: Giga Computing previewed a W775-V10-L01 prototype with the GB300 chip at SIGGRAPH last August. Monday’s release turns the prototype into a productized desk-side workstation and puts it in front of buyers this week in Abu Dhabi.
Why it matters
Two trends are colliding here. First, the hardware to run serious models locally keeps getting denser: a 1-PFLOPS box the size of a lunchbox was a fantasy five years ago, and a 20-PFLOPS desk-side workstation with three-quarters of a terabyte of coherent memory would have been a data-center SKU. Second, the reasons to keep AI work in-house — data sovereignty, regulated workloads, the economics of steady-state inference versus bursty cloud rental — keep getting stronger.
GIGABYTE isn’t alone in the DGX Spark lane — Dell, ASUS, HP and others build their own variants on the same silicon — but it’s one of the few pushing the full ladder, from the clustered desktop to the GB300 workstation to enterprise servers, under one brand. This site covered a smaller 64GB AI TOP ATOM variant just last week. The Abu Dhabi showcase is the next rung: local AI is no longer a hobbyist niche, it’s a product line.