Nvidia updates data center GPU roadmap, next-generation GB200 / GX200 will be released in 2025

Thank you, Mr. Air, a netizen of CTOnews.com, for your clue delivery! CTOnews.com, October 10, SemiAnalysis found that Nvidia recently updated its investor PPT presentation on NVIDIA's public and non-public strategy to establish a dominant position in the data center market.

After an in-depth study of this document, it was found that the company plans to integrate HBM3e storage, updates to PCI Express standards (6.0,7.0), and multi-GPU interconnection technologies that modify NVIDIA, such as NVLink.

Judging from NVIDIA's newly released official roadmap, NVIDIA is cautious about the specified date, which may be mainly due to the challenges of synchronizing its plans with foundry and preparing for software deployment.

It is worth noting that NVIDIA plans to launch a new generation of data center GPU architecture every year, for example, H200 will be used next year to replace the existing H100 (recently announced).

According to the roadmap, Nvidia is preparing to launch Hopper GH200 GPU in 2024, then Blackwell-based GB200 GPU sometime between 2024 and 2025, and eventually GX200 GPU after 2025.

Of course, these so-called GH200, GB200 and GX200 are all code names for GPUs, and the actual product names should be H200, B100, X100 and so on.

CTOnews.com Note: the "B" in the GB200 name stands for Blackwell, Nvidia's next-generation architecture, covering data centers and gaming products. By analogy, this "X" may just be a placeholder. After all, few scientists have surnames starting with "X", but it is too early to draw conclusions.