On Wednesday, December 7, local time, Meta, OpenAI and Microsoft said at an AMD investor event that they will all use AMD's newly developed artificial intelligence chip, the Instinct MI300X. This shows that while Nvidia's GPU chips are critical to developing and deploying AI programs like ChatGPT, many tech companies are looking for alternatives.
If AMD's latest high-end chip, the Instinct MI300X, can start shipping early next year and meet demand from AI companies and related cloud service providers, it could reduce the cost of developing AI models and put competitive pressure on Nvidia's surging AI chip business.
"All the focus is on big processors and big Gpus for cloud computing," AMD CEO Zifeng Su said on Wednesday.
AMD says the MI300X is based on a new architecture that significantly improves performance. The biggest feature of this chip is that it has 192GB of high-performance HBM3 memory, faster data transmission, and can adapt to larger scale artificial intelligence models.
Su directly compared the MI300X and the systems it builds to Nvidia's leading AI GPU chip, the H100.
"This performance translates directly into a better user experience," Su said. "When you ask a model a question, you always want it to respond faster, especially as the responses are increasingly complex."
The main question for AMD is whether the company, which has been based on Nvidia, will reinvest time and money in another GPU vendor. "Accepting AMD is going to take some work," Mr. Su said.
AMD told investors and partners on Wednesday that the company has made improvements to the ROCm software suite. To compete with Nvidia's CUDA software, which is the industry benchmark, a key flaw was addressed. This original flaw in the ROCm software suite has been one of the main reasons why AI developers currently prefer Nvidia.
Chip prices are also important. AMD did not disclose pricing for the MI300X on Wednesday. Nvidia GPU chips currently sell for about $40,000 each, and Su said AMD's chips have to be cheaper than Nvidia's and have lower operating costs to convince customers to buy them.
AMD also said Wednesday that it has signed agreements with some of the companies that need GPU chips the most. According to a recent report from market research firm Omidia, Meta and Microsoft are the two largest buyers of Nvidia H100 GPU chips in 2023.
Meta says it will use the MI300X chip to handle AI inference workloads, such as AI stickers, image editing, and running voice assistants.
Microsoft Chief Technology Officer Kevin Scott said the company will provide access to the MI300X chip through Azure Web services.
Oracle's cloud computing will also use the chip.
OpenAI said it will support AMD's GPU chips in a software product called Triton. Triton is not a large language model like GPT, but also uses chip capabilities for AI research.
AMD has not yet made a forecast for mass sales of the chip, only projecting total data center GPU revenue of about $2 billion in 2024. For comparison, Nvidia's data center revenue exceeded $14 billion in the most recent quarter alone, though that figure includes other chip businesses besides Gpus.
However, AMD said that the size of the AI GPU chip market could climb to $400 billion in the next four years, double the company's previous expectations. This shows that expectations for high-end AI chips are high in the industry, and is exactly why AMD is now focusing investors' attention on the product line.
Su also said AMD doesn't think it needs to beat Nvidia to do better in the market.
"I think it's pretty clear that Nvidia is definitely in the majority right now," Su said of the AI chip market. "We think that number could exceed $400 billion by 2027." We can definitely get a piece of that."
Advertising statement: This article contains some external skip links (including not limited to hyperlinks, QR codes, passwords, etc.), used to convey more information, save the selection time, the results are for reference only, all the articles of IT home include this statement.