A lot of what nvidia does is hype their next technology. You got to give it to Jensen. Whatever he says, everyone parrots. A grade marketing + management. Heck even stocks run as soon as he says AI couple of times. In this case, he mentioned potential of ‘generative AI’ and every stock market analyst was parroting the same. It was actually chatgpt that set the ball rolling. Everyone, including lisa su and jensen are jumping on the bandwagon. Just an year back, jensen was hyping software generated by AI.
Their main advantage is their software platform which will be difficult to break. Luckily for other cos, TAM is big and software effort is being done by the industry to look for alternative hardware. Like what PyTorch 2.0 did.
OpenAI Triton only officially supports Nvidia GPUs today, but that is changing in the near future. Multiple other hardware vendors will be supported in the future, and this open-source project is gaining incredible steam. The ability for other hardware accelerators to integrate directly into the LLVM IR that is part of Triton dramatically reduces the time to build an AI compiler stack for a new piece of hardware.
A good summary is here
Nvidia’s colossal software organization lacked the foresight to take their massive advantage in ML hardware and software and become the default compiler for machine learning. Their lack of focus on usability is what enabled outsiders at OpenAI and Meta to create a software stack that is portable to other hardware. Why aren’t they the one building a « simplified » CUDA like Triton for ML researchers? Stuff like Flash Attention, why does it come out of Ph.D. students and not Nvidia?
| Subscribe To Our Free Newsletter |
