Nvidia, Intel, and AMD unveiled major AI chip and infrastructure technologies at Computex 2024.

At Computex 2024 in Taiwan, Nvidia, Intel and AMD made major announcements signaling that the semiconductor industry is entering a new era of accelerated development. Nvidia CEO Jensen Huang revealed an expanded roadmap of new GPUs, CPUs and other chips for future data centers. AMD CEO Lisa Su announced an expanded data center accelerator chip roadmap starting with the 288-GB Instinct MI325X GPU coming later this year, along with plans to release the Ryzen AI 300 chips for AI-enabled laptops, the Ryzen 9000 desktop series, and the 192-core EPYC "Turin" processors. Intel CEO Pat Gelsinger announced the new Xeon 6 E-core CPUs, the upcoming Lunar Lake processors for AI-enabled PCs, and revealed ecosystem momentum and platform pricing for Intel's upcoming Gaudi 3 AI chip. Nvidia also announced the expansion of its MGX modular designs for Blackwell GPUs, the launch of its NIM inference microservices, and plans to boost AI application development for PCs running on its RTX GPUs.

Nvidia outlined an aggressive strategy to release new data center GPU architectures annually, representing an acceleration from its previous two-year release cycle. Huang stated the company's philosophy: "Our basic philosophy is very simple: build the entire data center scale, disaggregate it and sell it to you in parts on a one-year rhythm, and we push everything to the technology limits." The roadmap follows Nvidia's March announcement of its Blackwell architecture, which enables up to 30 times greater inference performance and consumes 25 times less energy for massive AI models compared to the Hopper architecture from 2022. In 2025, Nvidia plans to release Blackwell Ultra, described by Huang as "pushed to the limits," along with an updated Spectrum Ultra X800 Ethernet Switch. In 2026, the company will debut Rubin, an all-new GPU architecture using HBM4 memory, alongside a second-generation CPU called Vera following its Arm-based Grace architecture, the NVLink 6 Switch doubling chip-to-chip bandwidth to 3600 GBps, the CX9 SuperNIC capable of 1600 GBps, and the X1600 generation of InfiniBand and Ethernet switches. Huang emphasized: "All of these chips that I'm showing you here are all in full development, 100 percent of them, and the rhythm is one year at the limits of technology, all 100 percent architecturally compatible."

Through its MGX modular reference design platform, Nvidia is enabling server vendors to "quickly and cost-effectively build more than 100 system design configurations" by choosing their preferred GPU, CPU and data processing unit combinations. The newly revealed GB200 NVL2 single-node server design contains two Blackwell GPUs and two Grace CPUs for integration into existing data centers. The MGX platform supports other Blackwell-based products including the B100 GPU, B200 GPU, GB200 Grace Blackwell Superchip and GB200 NVL72, and will also support Intel's upcoming Xeon 6 P-core processors and AMD's upcoming EPYC "Turin" processors, with both companies planning "for the first time, their own CPU host processor module designs." AMD revealed its next-generation Ryzen 9000 desktop processors using the new Zen 5 architecture, with the flagship 16-core Ryzen 9 9950X beginning shipments in July.