Oxmiq Labs Aims to Make High-Performance Chip Design Accessible to All
A former Intel and AMD engineer wants to massively reduce the costs of chip development. In his start-up Oxmiq Labs Components were developed based on the RISC-V architecture, which can be combined relatively easily.
Comprehensive development platform
The renowned GPU engineer Raja Koduri drastically simplifies access to high-performance chips for artificial intelligence (AI) and graphics-both for large data centers and for compact edge devices. Instead of producing your own hardware, OxmiQ develops central technologies and software solutions that can then be licensed to other companies Tech spot. Koduri, who previously worked at AMD, Intel and Apple, relies on a new form of chip development with Oxmiq.
Classically, the production of new graphics chips is extremely costly, which often excludes smaller companies. Oxmiq wants to change this by primarily offering intellectual property – such as the core architecture Oxcore. This is based on the open RISC-V commanding set architecture and can be flexibly adapted to different areas of application-from AI systems to robots.
Another important product of the company is Oxquilt, a modular system with which companies can put together individual system-on-chip solutions. This eliminates the effort of having to start every time. In addition, there is the software platform Oxcapsule, which helps developers to run AI and graphics software on a wide variety of hardware – without complex adjustments.
With money from Mediatek
Particularly noteworthy is Oxpython: This solution allows CUDA-based applications-originally written for Nvidia GPUs-to perform on other chips, first on the AI accelerator of the partner tenstorrent. Oxmiq also relies on advanced technologies such as in-memory computing and so-called “nano-agents” directly in chip design to further increase efficiency and speed.
With $ 20 million in start -up capital – including Mediatek – Oxmiq is financially well positioned. The license model is challenging, but the increasing need for flexible AI hardware could play the company on the cards. After all, estimates assume that the development of a specialized GPU on this basis can become up to 90 percent cheaper.
