SK Hynix Develops HBM3E For Blistering AI Memory Bandwidth And NVIDIA Is On Board
SK hynix has developed what it claims is the world's best performing HBM3e memory with throughput of up to 1.15 terabytes (TB) per second. Translated into real-world workloads, that's fast enough to process more than 230 Full HD 1080p movies (each one 5GB in size) in a single second. That's kind of mind-boggling when put into that kind of perspective.
The memory chip maker is also touting a 10% improvement in heat dissipation, which it achieved by adopting a cutting-edge packaging technique called Advanced Mass Reflow Molded Underfill (MR-MUF). At a high level, this involves attaching chips to circuits on a lower substrate and then filling gaps with a liquid material when stacking those chips, as opposed to laying a film.
As much as we'd like to see the latest HBM solutions appear on consumer products, the focus is really on pricier data center hardware, especially as the industry increases its efforts on AI.
"With its experience as the supplier of the industry’s largest volume of HBM products and the mass-production readiness level, SK hynix plans to mass produce HBM3E from the first half of next year and solidify its unrivaled leadership in AI memory market," SK hynix says.
SK hynix already has a willing buyer in NVIDIA, which recently announced an updated Grace Hopper GH200 GPU infused with HBM3e memory.
"We have a long history of working with SK hynix on High Bandwidth Memory for leading edge accelerated computing solutions," said Ian Buck, Vice President of Hyperscale and HPC Computing at NVIDIA. "We look forward to continuing our collaboration with HBM3E to deliver the next generation of AI computing."