NextSilicon's Maverick-2 accelerator could be the next big thing
NextSilicon just launched the Maverick-2 accelerator, and it's already making waves in high-performance computing.
Built on TSMC's 5nm tech with up to 192GB of HBM3e memory (96GB on PCIe cards, 192GB on OAM modules), this chip is live at places like Sandia National Labs.
It promises over 4x better performance-per-watt than GPUs and slashes energy costs by more than half—pretty impressive for anyone who cares about speed and sustainability.
The chip can run your existing code without any tweaks
This chip crushes massive data tasks, handling huge graph workloads (think 25GB+) that even top GPUs struggle with.
It runs regular C/C++, Fortran, and OpenMP code as-is—no annoying tweaks needed.
In real-world benchmarks, it matches leading GPUs but uses only half the power.
NextSilicon is all about that double-precision science life
Instead of chasing AI hype like NVIDIA, NextSilicon is focusing on double-precision science work—great for researchers or anyone running big simulations.
Its smart architecture adapts on the fly to whatever you throw at it, making data centers greener and more efficient.
Plus, its toolchain works with all the major programming frameworks (with support for CUDA and others planned), so you're not locked in if you want to switch things up down the road.