Lisa Su displays an AMD Instinct MI300 chip as she delivers a keynote address at CES 2023 in Las Vegas, Nevada, Jan. 4, 2023.
David Becker | Getty Images
Meta, OpenAI, and Microsoft said at an AMD investor event Wednesday they are going to use AMD’s newest AI chip, the Instinct MI300X. It’s the most important sign thus far that technology firms are looking for alternatives to the expensive Nvidia graphics processors which have been essential for creating and deploying artificial intelligence programs such as OpenAI’s ChatGPT.
If AMD’s latest high-end chip is nice enough for the technology firms and cloud service providers constructing and serving AI models when it starts shipping early next yr, it could lower costs for developing AI models and put competitive pressure on Nvidia’s surging AI chip sales growth.
“The entire interest is in big iron and big GPUs for the cloud,” AMD CEO Lisa Su said Wednesday.
AMD says the MI300X relies on a new architecture, which regularly leads to significant performance gains. Its most virtue is that it has 192GB of a cutting-edge, high-performance variety of memory known as HBM3, which transfers data faster and can fit larger AI models.
Su directly compared the MI300X and the systems built with it to Nvidia’s most important AI GPU, the H100.
“What this performance does is it just directly translates right into a higher user experience,” Su said. “While you ask a model something, you need it to come back faster, especially as responses get more complicated.”
The most important query facing AMD is whether or not firms which have been constructing on Nvidia will invest the time and money to add one other GPU supplier. “It takes work to adopt AMD,” Su said.
AMD on Wednesday told investors and partners that it had improved its software suite called ROCm to compete with Nvidia’s industry standard CUDA software, addressing a key shortcoming that had been one among the first reasons AI developers currently prefer Nvidia.
Price may also be vital. AMD didn’t reveal pricing for the MI300X on Wednesday, but Nvidia’s can cost around $40,000 for one chip, and Su told reporters that AMD’s chip would have to cost less to purchase and operate than Nvidia’s so as to persuade customers to buy it.
Who says they’ll use the MI300X?
AMD MI300X accelerator for artificial intelligence.
On Wednesday, AMD said it had already signed up a number of the firms most hungry for GPUs to use the chip. Meta and Microsoft were the 2 largest purchasers of Nvidia H100 GPUs in 2023, according to a recent report from research firm Omidia.
Meta said it would use MI300X GPUs for AI inference workloads such as processing AI stickers, image editing, and operating its assistant.
Microsoft’s CTO, Kevin Scott, said the corporate would offer access to MI300X chips through its Azure web service.
Oracle‘s cloud may also use the chips.
OpenAI said it could support AMD GPUs in one among its software products, called Triton, which is not an enormous large language model like GPT but is utilized in AI research to access chip features.
AMD is not forecasting massive sales for the chip yet, only projecting about $2 billion in total data center GPU revenue in 2024. Nvidia reported greater than $14 billion in data center sales in probably the most recent quarter alone, although that metric includes chips aside from GPUs.
Nevertheless, AMD says the entire marketplace for AI GPUs could climb to $400 billion over the following 4 years, doubling the corporate’s previous projection. This shows how high expectations are and how coveted high-end AI chips have turn out to be — and why the corporate is now focusing investor attention on the product line.
Su also suggested to reporters that AMD doesn’t think that it needs to beat Nvidia to do well out there.
“I believe it’s clear to say that Nvidia has to be the overwhelming majority of that at once,” Su told reporters, referring to the AI chip market. “We consider it might be $400 billion-plus in 2027. And we could get a pleasant piece of that.”