AMD announced products for AI data centers, including the EPYC 9006 server CPU and the Helios AI rack server.
On July 24, 2026, Japan time, AMD held a keynote speech at its AI-related product launch event, 'Advancing AI 2026,' where it announced products such as the server CPU 'EPYC 9006' and the rack-scale system 'Helios.'
AAI 2026: AMD CEO Lisa Su Launches End-to-End AI Solutions (Keynote Replay) - AMD Newsroom
Advancing AI 2026 | Build What's Next with @AMD - YouTube
High-performance GPUs are essential for AI training and inference. Furthermore, when entrusting AI with tasks such as application development or administrative work, CPU performance is just as important as GPU performance. For this reason, NVIDIA, a strong rival of AMD, has developed its own CPU, ' Vera ,' and has begun production of the 'Vera Rubin NVL72,' which combines it with the AI-specialized GPU 'Rubin.'
NVIDIA announces details of its AI-focused GPU 'Rubin' and AI-focused CPU 'Vera,' with the AI rack 'Vera Rubin NVL72' boasting a processing performance of 2400 TFLOPS in FP64 precision and integrating Groq's high-speed inference chip - GIGAZINE
The 6th generation EPYC 'EPYC 9006 series (codename: Venice)' is a server CPU developed to compete with Vera.
The EPYC 9006 series is based on the Zen 6 architecture and manufactured using a 2nm process. It features up to 512 threads per CPU and a maximum memory bandwidth of 1.6TB/s.
The AI agent's execution performance is touted as 1.7 times that of the previous generation model, significantly surpassing Vera. Furthermore, it is touted as delivering 1.4 times the performance of the previous generation model in general-purpose use cases beyond AI.
And then there's the 'Helios' rack system, which combines the EPYC 9006 with the AI-specialized GPU 'MI455X'.
The MI455X boasts up to 34 times faster token processing speed compared to the MI355X. The processing speed for particularly long tokens has significantly improved.
Helios is positioned as a competitor to Vera Rubin. The graph below compares the token processing throughput of Helios and Vera Rubin. Helios outperforms Vera Rubin by 15% in low-load processing and by 10% in high-load processing.
The Ryzen AI Embedded X100 series processors for AI and the KRIA development board equipped with the X100 were also announced. The X100 is a processor that integrates the CPU, GPU, and NPU, and is intended for industrial products such as robots and medical devices.
Kria has also announced the availability of the 'Kria AI Robotics Developer Platform,' a development kit that includes fans, cases, and other accessories for Kria.
Furthermore, AMD announced a partnership with Cerebras, a developer of high-speed inference systems, to ship ultra-low latency AI systems in the second half of 2026.
AMD partners with Cerebras, a high-speed AI inference system provider, to ship ultra-low latency AI systems in the second half of 2026 - GIGAZINE
Related Posts:
in Hardware, Posted by log1o_hf