AMD launched a brand new artificial-intelligence chip on Thursday that’s taking direct purpose at Nvidia’s knowledge middle graphics processors, generally known as GPUs.
The Intuition MI325X, because the chip is named, will begin manufacturing earlier than the top of 2024, AMD mentioned Thursday throughout an occasion asserting the brand new product. If AMD’s AI chips are seen by builders and cloud giants as a detailed substitute for Nvidia’s merchandise, it might put pricing stress on Nvidia, which has loved roughly 75% gross margins whereas its GPUs have been in excessive demand over the previous yr.
Superior generative AI equivalent to OpenAI’s ChatGPT requires large knowledge facilities stuffed with GPUs with a purpose to do the required processing, which has created demand for extra firms to offer AI chips.
Up to now few years, Nvidia has dominated nearly all of the information middle GPU market, however AMD is traditionally in second place. Now, AMD is aiming to take share from its Silicon Valley rival or at the very least to seize a giant chunk of the market, which it says will likely be value $500 billion by 2028.
“AI demand has really continued to take off and truly exceed expectations. It is clear that the speed of funding is continuous to develop in every single place,” AMD CEO Lisa Su mentioned on the occasion.
AMD did not reveal new main cloud or web prospects for its Intuition GPUs on the occasion, however the firm has beforehand disclosed that each Meta and Microsoft purchase its AI GPUs and that OpenAI makes use of them for some functions. The corporate additionally didn’t disclose pricing for the Intuition MI325X, which is often bought as a part of a whole server.
With the launch of the MI325X, AMD is accelerating its product schedule to launch new chips on an annual schedule to higher compete with Nvidia and reap the benefits of the growth for AI chips. The brand new AI chip is the successor to the MI300X, which began delivery late final yr. AMD’s 2025 chip will likely be referred to as MI350, and its 2026 chip will likely be referred to as MI400, the corporate mentioned.
The MI325X’s rollout will pit it in opposition to Nvidia’s upcoming Blackwell chips, which Nvidia has mentioned will begin delivery in vital portions early subsequent yr.
A profitable launch for AMD’s latest knowledge middle GPU might draw curiosity from traders which can be searching for further firms which can be in line to learn from the AI growth. AMD is just up 20% thus far in 2024 whereas Nvidia’s inventory is up over 175%. Most business estimates say Nvidia has over 90% of the marketplace for knowledge middle AI chips.
AMD inventory fell 3% throughout buying and selling on Thursday.
AMD’s largest impediment in taking market share is that its rival’s chips use their very own programming language, CUDA, which has develop into commonplace amongst AI builders. That primarily locks builders into Nvidia’s ecosystem.
In response, AMD this week mentioned that it has been enhancing its competing software program, referred to as ROCm, in order that AI builders can extra simply swap extra of their AI fashions over to AMD’s chips, which it calls accelerators.
AMD has framed its AI accelerators as extra aggressive to be used circumstances the place AI fashions are creating content material or making predictions reasonably than when an AI mannequin is processing terabytes of knowledge to enhance. That is partially as a result of superior reminiscence AMD is utilizing on its chip, it mentioned, which permits it to server Meta’s Llama AI mannequin sooner than some Nvidia chips.
“What you see is that MI325 platform delivers as much as 40% extra inference efficiency than the H200 on Llama 3.1,” mentioned Su, referring to Meta’s large-language AI mannequin.
Taking over Intel, too
Whereas AI accelerators and GPUs have develop into essentially the most intensely watched a part of the semiconductor business, AMD’s core enterprise has been central processors, or CPUs, that lay on the coronary heart of almost each server on the earth.
AMD’s knowledge middle gross sales through the June quarter greater than doubled previously yr to $2.8 billion, with AI chips accounting for less than about $1 billion, the corporate mentioned in July.
AMD takes about 34% of whole {dollars} spent on knowledge middle CPUs, the corporate mentioned. That is nonetheless lower than Intel, which stays the boss of the market with its Xeon line of chips. AMD is aiming to alter that with a brand new line of CPUs, referred to as EPYC fifth Gen, that it additionally introduced on Thursday.
These chips are available numerous totally different configurations starting from a low-cost and low-power 8-core chip that prices $527 to 192-core, 500-watt processors supposed for supercomputers that value $14,813 per chip.
The brand new CPUs are significantly good for feeding knowledge into AI workloads, AMD mentioned. Almost all GPUs require a CPU on the identical system with a purpose to boot up the pc.
“Immediately’s AI is absolutely about CPU functionality, and also you see that in knowledge analytics and a variety of these varieties of functions,” Su mentioned.
