NVIDIA Announces Hopper Computing Architecture

NVIDIA has officially announced that it will host the new Spring GTC Graphics Technology Conference on March 21, 2022, in San Jose, California, with the usual keynote address by Founder and CEO Jen-Hsun Huang.

Also on stage will be NVIDIA CTO Macheal Kagan, Senior Vice President of R&D and Chief Scientist Bill Dally, Director of Machine Learning Anima Anandkumar, Vice President and Principal Director of Accelerated Computing Ian Buck, and Vice President of Healthcare Kimberly Powell, among other executives.

NVIDIA GTC 2022大会官宣:Hopper计算架构要来了

NVIDIA GTC 2022大会官宣:Hopper计算架构要来了

The NVIDIA GTC conference generally focuses on new hardware and software for HPC, AI, and autonomous driving, with a special focus on deep learning this time.

Although it is unlikely that the RTX 40 series of gaming cards will be announced, it will almost certainly be based on the next generation of computing architecture “Hopper”, as well as the corresponding accelerated computing card solutions.

微星GeForce RTX 3080 Ti SUPRIM X 12G 三区域光效,刀锋6代风扇,空气动力学,核心7热管

MSI GeForce RTX 3080 Ti SUPRIM X 12G tri-zone light effect, Blade 6 fan, aerodynamics, core 7 heatpipe

[Distributor] Store

[Product Price] 14999RMB

Hopper’s name comes from Herb, the great female programmer who is known as the mother of compilation.

NVIDIA GTC 2022大会官宣:Hopper计算架构要来了

Previous news shows that the Hopper architecture chip has been streamed, and it is speculated that the top core will be called GH100, which will be the first time in NVIDIA’s history to use MCM multi-core package with two chips integrated inside, and is expected to have a total of 288 SM stream multiprocessors, which is as much as 2.6 times more than A100, while the internal structure will also be overhauled, and the manufacturing process is expected to be TSMC 5nm.

AMD CDNA2/RDNA3, NVIDIA Xe-HPC will also be similar MCM multi-core package, which is based on CDNA2 architecture Instinct MI250/MI250X computing card has been released first, 6nm process, dual-core design, using 2.5D EFB bridging technology, the industry’s first multi-Die whole package (MCM), up to 14080 stream processor cores, 880 second-generation matrix cores, with 8192-bit 128GB HBM2e.

As for the next-generation gaming GPU, the codename is rumored to be “Ada Lovelace,” the only legitimate daughter of poet Byron, who is credited with being the first computer scientist to write the first computer program in history, and the process is also 5nm, with the release expected in the third quarter of this year.