AMD has claimed that its Ryzen AI 300 series processors are outperforming Intel’s latest offerings in handling large language models (LLMs).
The company made this claim in a whitepaper published October 30 where it stated that its Ryzen AI 9 HX 375 processor can outshine Intel Core Ultra 7 258v processor in throughput and latency for AI workloads.
AMD revealed that despite having a slower RAM speed (7,500 MT/s) compared to Intel’s 8,533 MT/s, its Ryzen AI 9 HX 375 processor still outperformed Intel’s offering by up to 27% in terms of tokens per second, a key metric for LLM performance.
The whitepaper also highlights the Ryzen AI’s advantage in “time to first token” – the latency between submitting a prompt and the model starting to generate output. AMD claims its Ryzen AI 9 HX 375 processor can reduce latency in larger language models, achieving up to a 3.5x speedup in “time to first token” compared to Intel’s offerings.
AMD attributes this performance boost to specialized hardware, including the AMD XDNA 2 architecture-based neural processing units and the integrated graphics processing unit. The company also highlights the Variable Graphics Memory feature, which allows the iGPU to access more system RAM for improved throughput.
These reported AI performance gains in its Ryzen AI 300 series could be a message of hope for AMD investors, who may have been dampened by the company’s recent Q3 earnings report.
The underwhelming Q3 results led to a 7% drop in AMD’s stock price in after-hours trading yesterday, a similar fate suffered by Intel after it announced its Q3 earnings call earlier this month. Both companies have been going head to head in recent years in an attempt to outdo each other’s processor capabilities.
Despite the disappointing Q3 results, the advancements in AMD’s Ryzen AI 300 series could propel the semiconductor maker for a better Q4 call as it continues to compete in the evolving AI chip market currently led by Nvidia.