Micron Introduces 36GB 12-Hi HBM3E Memory with 9.2 Gbps Speed

Micron has announced the start of production of its 12-Hi HBM3E memory, with a capacity of up to 36GB, speed of 9.2 Gbps and capable of reaching up to 1.2TB/s of bandwidth – with its manufacture aimed at use in artificial intelligence GPUs.

According to the manufacturer, this model allows use in larger AI models such as Llama 2 – with 70 billion parameters executed on a single processor. The 36GB capacity – 50% larger than 8-Hi HBM3E memories – also allows for a faster response time to avoid delays in communication between GPUs.

Another point highlighted by Micron for its 36GB HBM3E memory is that its power consumption is “significantly lower” than the 24GB HBM3E 8-Hi models. The manufacturer’s proposal is to bring to the market maximum performance with the lowest energy consumption, ensuring optimization for servers that increasingly demand more.

The company also claims that one of the models’ highlights is a fully programmable MBIST, allowing the system to run at full speed to its specifications – providing enhanced test coverage for accelerated validation, enabling faster time to market (TTM).

See more:

Considering that NVIDIA, Intel and AMD are bringing increasingly improved solutions for GPUs with artificial intelligence, there is a race against time between companies like Micron, Samsung and even SK Hynix to provide the most advanced hardware possible and stand out together in the tech industry.

With Micron’s announcement of its HBM3E 12-Hi with 36GB, they are ahead in this dispute and should soon be sending the memories to manufacturers who will use them in these GPUs for AI servers – which promises to have an even bigger leap between the end of 2024 and the beginning of 2025.

Fonte: TweakTown