A few days ago NVIDIA announced Grace, a server CPU with ISA ARM and that will use next generation ARM Neoverse cores. This CPU is not designed to compete in the PC market and much less in that of smartphones due to its large size and consumption.
Intel Responds to NVIDIA Grace Announcement
Although NVIDIA Grace will not be launched until 2023 we have to put Intel’s response in context, and that is while NVIDIA was announcing its future ARM CPU for servers, Intel shares had a drop in their price due to the announcement. Which has led Intel to give an answer to calm its investors.
That is why through Yahoo Finance, Pat Gelsinger, current CEO of Intel, has given a response to the announcement of the NVIDIA CPU:
NVIDIA today announced some new chips for the data center market, with a particular focus on accelerating artificial intelligence applications with standard chips known as central processing units or CPUs. The stock market has reacted quickly. What is our position with them?
We announced Ice Lake (a new server processors) in the past week with extremely positive feedback. And with Ice Lake, we have an extraordinary expansion of capabilities for AI. It is NVIDIA that is not responding to us and not us to them. Clearly, the idea of an improved CPU for AI is the domain where Intel is leading.
We also have our Habana product line, a chipmaker specializing in AI that Intel bought in 2019, which is an aggressive move, and the computing deal with Amazon is proof of that. So clearly, I would say that the idea for CPUs comes from Intel. We are now building the AI into them and we hope this is an area where we are on offense, not defensive in the future.
Intel has plans to integrate your AMX units into your CPUs, these units perform the same work as systolic arrays or Tensor units. Which are widely used to accelerate the different artificial intelligence algorithms.
What does Intel have to fear from NVIDIA?
NVIDIA today it is no longer a company that makes graphics processors, but have made the leap to AI where the graphics they are just a small part. The reason for this is obvious, they see it as a much bigger business opportunity. The key to this is the implementation of its Tensor Cores in its GPUs and the development of units such as the NVDLA accelerator, but a Tensor Core is nothing more than a systolic array that can not only be implemented in a GPU but also in a CPU.
Although Pat Gelsinger’s words contain much of the truth, the problem for Intel is the software ecosystem based on CUDA and other NVIDIA tools in the high-performance computing market. It is the dependence of much of the software on NVIDIA tools that Intel has led them to have to develop their response in the form of OneAPI development technologies.
So NVIDIA’s advantage over Intel is in the software, which is a much more difficult stumbling block to overcome than NVIDIA will have once it has completed the ARM purchase.