Intel's ex-chief of the GPU department begins his new journey at Oxmiq Labs, bringing world-class architects to deliver AI GPUs from scratch.
Raja Koduri Begins His New GPU Startup, Rebuilding the Entire GPU Ecosystem From Scratch, Prioritizing "Software First" Strategy to Enable Unmodified Python-Based CUDA Apps to Run on Non-NVIDIA Hardware
Roughly two years after retirement from Intel, the ex-chief GPU architect, Raja Koduri, who has previously worked for AMD as a SVP and Chief GPU Architect in the graphics department, is now on his new venture of producing AI GPUs from scratch. Raja founded a new startup called Oxmiq Labs, which will now leverage the experience of world-class GPU and AI architect veterans.
As per the press release, the company will now represent 500 years of collective experience and will utilize hundreds of patents to work on new AI hardware, which will be built from the ground up. Oxmiq won't just tweak the current GPU designs but will rebuild the entire GPU ecosystem from scratch with the "Atoms to Agents" mindset, developing low-level hardware and bringing transistor-level innovation to developing high-level AI systems.
With the increasing demands of multimodal AI, where various types of data such as text, images, and video interact in real time, GPUs with the capability to process these forms of data simultaneously and intelligently are the need of the hour. Since GPUs are flexible and programmable, unlike AI chips which serve a limited and specific task, Oxmiq's architecture aims to leverage that flexibility but improve the perfomrance and efficiency significantly.
Through hardware innovations such as OXCORE, which is a modular GPU core bringing Scalar, Vector, and Tensor units for various operations, Oxmiq will be delivering support for nano agents, Python acceleration, and CUDA/SIMD compatibility. OXCORE will be able to scale from tiny devices all the way to large data centers.
The OXCORE will utilize the OXQUILT chiplet architecture, which will enable the manufacturing of customized GPUs by mixing compute, memory, and interconnect components in the right ratio. This will help in the reduction of R&D and production time/costs compared to the traditional GPUs. However, as mentioned previously, Oxmiq's "Software First" strategy will be crucial to build an easy-to-use software stack, which will be compatible with both Oxmiq's GPU IP and third-party AI hardware like Tenstorrent.
We’re excited to partner with OXMIQ on their OXPython software stack,
OXPython’s ability to bring Python® workloads for CUDA to AI platforms like Wormhole™ and
Blackhole™ is great for developer portability and ecosystem expansion. It aligns with our goal of letting
developers open and own their entire AI stack.- Jim Keller, CEO of Tenstorrent
Through its flagship software called OXPython, it will allow NVIDIA CUDA applications to run on non-NVIDIA hardware without modifying the code, and will be utilized on Tenstorrent AI chips first. That said, Oxmiq's unified software ecosystem, called OXCapsule, will further help developers deploy the AI apps across many platforms by hiding the complexity of hardware.
OXMIQ has an impressive bold vision and world-class team,
The company’s GPU IP and software innovations will drive a new era of compute
flexibility across devices – from mobile to automotive to AI on the edge.- Lawrence Loh, SVP of MediaTek
That said, Oxmiq's strategy will be a licensing-first model, which will offer its IP licenses to others without having to build and sell full-fledged GPUs, avoiding the expensive tape-out process. Oxmiq has already raised $20 million in seed funding from companies like Mediatek, and it will be interesting to see the progress of this GPU startup, particularly when Raja claims that it is the first GPU startup in Silicon Vallenty in over 25 years.
About the author: Sarfraz Khan is a hardware reporter with a focus on PC components and the builder community. With years of experience writing about PC hardware and laptops, his work has been featured on several reputable technology publications. Sarfraz's hands-on experience is demonstrated through his first-person accounts of using and comparing different hardware configurations, providing practical and relatable insights for everyday users. His technical analysis is respected by peers in the enthusiast community and has been cited by specialized hardware sites such as Germany's Igor's Lab.
Follow Wccftech on Google to get more of our news coverage in your feeds.
