Last October, I wrote about Acer’s Veriton GN100 AI Mini Workstation, a compact desktop designed to bring serious AI horsepower to a machine small enough to sit on a desk. Now that same system is getting a public showcase. Acer says the Veriton GN100 will serve as the official hardware platform for the Spark Hack Series in New York, a developer competition presented by NVIDIA that runs April 10 through April 12.
The event will bring together machine learning engineers, startup founders, and systems developers who will build projects using open data from New York City. Up to 40 teams will compete across several tracks focused on real world problems, including human impact, environmental sustainability, and cultural access. The three winning teams will each receive a Veriton GN100 to keep so they can continue developing their work after the event wraps up.
The workstation itself is built on the NVIDIA GB10 Grace Blackwell Superchip platform and is designed to deliver server class AI performance in a small desktop footprint. Acer says the system can reach up to 1 petaFLOP of FP4 AI performance while offering 128GB of unified memory. Storage options go up to 4TB of NVMe, and connectivity includes Wi Fi 7, USB C ports, HDMI output, and an NVIDIA ConnectX 7 Smart NIC for high speed networking.
For the hackathon, Acer is highlighting some newer capabilities that expand what the workstation can handle locally. The Veriton GN100 can now link up to four systems together using a 200GbE RoCE switch. That scaling capability allows developers to run much larger AI models locally, with support for models up to 700 billion parameters compared to the earlier limit of roughly 405 billion.
Another major focus is support for NVIDIA’s NemoClaw reference stack. NemoClaw is designed for building autonomous AI agents capable of handling long running workflows. It combines open models such as Nemotron with the OpenShell runtime, allowing agent frameworks to operate inside controlled sandbox environments.
That approach allows developers to experiment with AI systems that do more than respond to prompts. These agents can plan multi step tasks, call specialized sub agents, and interact with technical tools across applications such as spreadsheets, CAD software, and creative production tools. Developers can also keep inference local or run hybrid deployments, which can reduce reliance on cloud processing and help maintain tighter control over data.
Acer is also introducing a management platform called Acer Sense Pro. The software provides centralized monitoring and troubleshooting through a single interface. Developers can track CPU, GPU, storage, and memory activity in real time while built in benchmarking tools measure AI performance using metrics such as tokens per second and time to first token.
Sense Pro also includes diagnostic tools, a knowledge base, and a local AI assistant intended to help developers identify and resolve hardware issues privately. The platform is designed to make it easier to evaluate different AI models and fine tune performance during development.
Hackathons like the Spark Hack Series tend to be interesting because they reveal what developers actually build when given powerful hardware and real data. Sometimes the results are impressive. Other times they highlight the limits of today’s AI tools.
Either way, putting the Veriton GN100 at the center of the event gives Acer a chance to show how its compact AI workstation performs outside the usual marketing slides.
