Financial Times reports that several major tech companies are beginning to support AI software trends that do not rely solely on NVIDIA’s CUDA. CUDA has long been NVIDIA’s secret weapon, granting them a significant advantage in the AI chip market. However, the majority of popular software requires CUDA to operate, leading to NVIDIA monopolizing the AI chip market and becoming a costly dependence for tech companies in need of such chips.
According to Financial Times, this initiative is known as Triton, developed by OpenAI since 2021 to allow AI software to run on a variety of chips. Major tech companies supporting Triton include Meta, Microsoft, and Google, all of which are NVIDIA customers developing their own AI chips.
Currently, Triton supports AMD Instinct for running, with Intel Gaudi soon to follow. An example of Triton in use is Meta utilizing it with their custom-designed MTIA chip, yielding positive results. Triton is not the only software project aiming to break free from the dependence on CUDA, as the UXL Foundation, backed by Intel, and based on Intel’s oneAPI, as well as Modular’s Mojo programming language project, are also striving to create AI development languages that are easier to use than CUDA.
Source: Financial Times
TLDR: Major tech companies are supporting AI software trends that do not rely on NVIDIA’s CUDA, with initiatives like Triton aiming to diversify chip usage for AI software development.
Leave a Comment