- AMD launched the Instinct MI325X accelerator on Thursday, and shipping starts later this fall.
- The new chip offers better performance than Nvidia’s H200 yet lags behind Blackwell, analysts say.
- AMD’s stock dipped over 4% post-announcement, reflecting only modest excitement amid competition.
AMD CEO Lisa Su announced Thursday the launch of the Instinct MI325X accelerator, which is meant to compete against Nvidia. Industry analysts say it’s still behind by a year.
According to AMD, the new chip series offers 1.8 times capacity, 1.3 times bandwidth, and 30% better inference than the H200 tensor core GPU by Nvidia, which is used for high-performance computing. AMD plans to start production shipments for the new accelerator later this year.
While the GPU significantly improved over AMD’s previous offerings, Bernstein Research analysts wrote that it faces “zero challenge” against Nvidia’s upcoming Blackwell chip.
Nvidia announced in March the Blackwell series, which is set to ship in the fourth quarter of this year. The series has been backordered for at least a year as demand has exceeded supply.
“Even the company’s MI350X tease shows raw performance that, while on par with Blackwell on paper, arrives a year later, just in time to compete against Nvidia’s next offerings,” wrote Bernstein Research analysts. “Hence we do not see even AMD’s accelerated road map closing the competitive gap.”
While AMD boasted a robust lineup, there is “no near-term catalyst to change the dynamic” between the company and its chip rivals, wrote Vivek Arya, research analyst at Bank of America Securities.
“Training performance seems 1 year behind Blackwell (on par with H200) while inferencing is only slightly better,” Arya wrote.
After the AI announcements on Thursday, AMD stock dipped over 4%, which Bernstein analysts said potentially reflected “modest excitement” due to a lack of major new customers and a “somewhat disappointing” performance announcement amid Nvidia’s impending Blackwell launch.
AMD also took a jab at Intel in central processing units with the launch of its 5th generation EPYC processors, formerly codenamed “Turin,” for enterprise, AI, and cloud computing.
“We fully expect, as our competition launches their next generation of CPUs, and they ship in volume, that Turin will continue to be the leader for enterprise space,” said Su as she held up the chips for the audience in front of a slide that compared EPYC to Intel’s Intel’s 5th generation Xeon processors, codenamed Emerald Rapids.
Intel has recently announced a round of layoffs as it reported falling revenue growth.
