AMD mass-produces AI chip with higher inference ability than GPU by the tip of this 12 months… Stock price falls

-

Lisa Su AMD CEO (Photo = Shutterstock)

AMD released a brand new artificial intelligence (AI) chip and server chip, difficult Nvidia and Intel, the leaders in each market. Nonetheless, the market’s response appears to be somewhat cold.

Reuters and CNBC reported on the tenth (local time) that AMD announced its latest AI chip ‘Instinct MI325X’ at a brand new product launch event.

This chip, released by AMD as a successor to ‘MI300X’, is scheduled to enter mass production before the tip of the 12 months. Moreover, the corporate plans to release ‘MI350’ next 12 months and ‘MI400’ in 2026.

The Instinct MI325X chip is geared toward NVIDIA’s next-generation chip ‘Blackwell’, and is explained to have 1.8 times the memory capability and 1.3 times the bandwidth of NVIDIA’s ‘H200’, which is currently considered the perfect AI chip.

It was emphasized that the MI325 platform provides as much as 40% higher inference performance than NVIDIA H200 in Meta’s ‘Rama 3.1 model’. Specifically, it was claimed that AI models have a greater ability to generate or predict content than to process large-scale data.

If AMD’s latest AI chip is evaluated as comparable to Blackwell’s, it could put great pressure on Nvidia, which has earned about 75% cash in on GPU sales over the past 12 months.

CEO Lisa Su said, “AMD is working to reply to the demand for AI chips, with a market size of $500 billion by 2028.”

The explanation AMD had difficulty increasing its market share within the AI ​​chip field was not only due to hardware performance, but in addition due to Nvidia’s programming language ‘CUDA’. CUDA has turn into a regular amongst AI developers and is exerting influence in keeping developers in NVIDIA’s ecosystem.

Accordingly, AMD announced this week that it should improve its software, ROCm, in order that AI developers can easily convert more AI models to AMD AI chips.

Also, on the event, they announced a brand new server CPU line named ‘Turin’ and expressed their intention to focus on the server processor market dominated by Intel.

It was emphasized that the most recent server chip ‘EPYC9965’ has excellent overall performance, including a video transcoding time that’s 4 times faster than Intel’s server chip ‘Xeon’. Nonetheless, Intel announced that it has not yet been compared with the next-generation server chip ‘Xeon 6’ announced in September.

Despite the announcement of the brand new chip, the market response was cold, with AMD’s stock price plummeting by 4% on the Latest York Stock Exchange that day. The industry commented that this announcement was already known since early this 12 months and was nothing latest.

Specifically, experts identified that AMD continues to be a couple of 12 months behind Nvidia. “While it’s a big improvement over AMD’s predecessor, it should not threaten NVIDIA Blackwell in any way,” Bernstein Research analysts wrote.

Reporter Park Chan cpark@aitimes.com

ASK DUKE

What are your thoughts on this topic?
Let us know in the comments below.

0 0 votes
Article Rating
guest
0 Comments
Oldest
Newest Most Voted
Inline Feedbacks
View all comments

Share this article

Recent posts

0
Would love your thoughts, please comment.x
()
x