Supercharged speeds – Micron ships HBM4 memory to “key customers”

Micron’s new HBM4 memory provides a 60% speed boost over its last-generation memory

Micron has confirmed that it has started shipping its next-generation HBM4 memory to “key customers” ahead of its production ramp in 2026. With its new 12-high 36GB HBM4 modules, Micron has delivered 60% more performance than its last-generation HBM3E memory modules. This boost will fuel the next generation of AI hardware advancement, offering boosted performance and power efficiency.

With 2.0 TB/s of memory bandwidth per chip, a single HBM4 module can deliver more memory bandwidth than an RTX 5090 graphics card. Nvidia’s GeForce RTX 5090 uses sixteen GDDR7 memory chips to deliver 1792 GB/s of memory bandwidth. With HBM4, Micron can deliver more bandwidth and higher memory capacities on a single chip.

With HBM4, Micron also boasts a 20% boost in power efficiency when compared to its HBM3E products. This is great news for datacenters, as increased performance/watt results in reduced power and cooling costs.

The increased data throughput of HBM4 memory will help to accelerate the inferencing performance of new AI accelerators. With a 60% boost in bandwidth, HBM4 memory will deliver a major leap in performance for those who adopt it. Unfortunately, HBM memory is primarily used by enterprise and data centre products. That means that consumer-grade GPUs are unlikely to benefit from this tech anytime soon.

You can join the discussion on Micron’s super-fast HBM4 memory shipping to customers on the OC3D Forums.