Home » Markets » AI

JEDEC HBM4 high-bandwidth memory standard addresses next-gen AI, HPC

Addressing the growing use of AI and high-performance computing, JEDEC has published a new high-bandwidth memory (HBM) DRAM standard: HBM4.

JEDEC HBM4 high-bandwidth memory standard addresses future AI, HPC

It aims to improve the performance, efficiency, and scalability of the memory required to power the next generation of AI, HPC (high-performance computing), and graphics workloads. For example, the next generation AI training and inference systems.

HBM4

With transfer speeds up to 8 Gb/s across a 2048-bit interface, HBM4 boosts total bandwidth up to 2TB/s, highlights JEDEC.


The standard also doubles the number of independent channels per stack. Specifically, from 16 channels (HBM3) to 32 channels with 2 pseudo-channels per channel. This provides designers with more flexibility and independent ways to access a cube (the stacking of DRAM chips), says the Solid State Technology Association.


In terms of power efficiency, the JESD270-4 spec supports vendor specific VDDQ (0.7V, 0.75V, 0.8V or 0.9V) and VDDC (1.0V or 1.05V) levels. This should result in lower power consumption and improved energy efficiency.

Finally, for capacity, HBM4 supports 4-high, 8-high, 12-high and 16-high DRAM stack configurations with 24Gb or 32Gb die densities. This will provide for a higher cube density of 64GB (32Gb 16-High).

AI

“High performance computing platforms are evolving rapidly and require innovation in memory bandwidth and capacity,” said Barry Wagner. He is Director of Technical Marketing at NVIDIA and the JEDEC HBM Subcommittee Chair.

“Developed in collaboration with technology industry leaders, HBM4 is designed to drive a leap forward in efficient, high performance computing for AI and other accelerated applications.”

Bandwidth

For its part, Google emphasises the importance of memory bandwidth.

“[This] is one of the key pillars of performance for AI computing systems,” said Nikhil Jayaram, VP of Google Cloud Silicon. “JEDEC HBM4 represents the big step in bandwidth that Google needs for next generation training and inference systems. We look forward to the advances in AI that HBM4-based systems will enable.”

To download JESD270-4, visit the JEDEC website.

Image: Google

See also: JEDEC updates Universal Flash Storage and Memory Interface standards

Alun Williams

Web Editor of Electronics Weekly, he is the author of the Gadget Master, Eyes on Android and Electro-ramblings blogs and also covers space technology news. He has been working in tech journalism for worryingly close to thirty years. In a previous existence, he was a software programmer.

Leave a Reply

Your email address will not be published. Required fields are marked *

*