JEDEC updates HBM for bigger memory stacks and faster transfer speeds

JEDEC recently updated its HBM (High-Bandwidth Memory) recently. The new JESD235A standard will allow two-high, four-high and eight-high TSV stacks with each having the capacity between 1GB to 8GB.

This is a significant boost compared to the earlier HBM standard that is implemented in AMD Radeon R9 Fury X. The Fury X’s HBM has four stacks with 128GB/s effective bandwidth. The new standard will be based on 1024-bit based eight-channel interface like the first generation HBM and up to 256 GB/s transfer speed.

The HBM DRAM update is not going to be useful for gaming, but also for servers, networking and client applications which require peak bandwidth, bandwidth-per-watt capacity-per-area. from the CPU and GPU makers. Considering the adopted display resolution and datasets are becoming much larger, having large memory with less footprint and faster transfer speed becomes very crucial to keep up with the demand. As one will expect, this will enable GPU makers to have much smaller and high-performing graphic card for its size, thus enabling a small form-factor as a gaming powerhouse.

The currently available cards (with the exception of AMD Radeon Fury and Fury X) have PCB and cooling systems extended even higher than 11 inches at times. HBM should enable most of the units to qualify as ‘big things come in small packages’. Thanks to HBM implementation, Fury X’s reference card PCB measures up to 7.5″ wide.

The group emphasized that the pseudo-channel architecture has improved its actual bandwidth speed. It also has a feature to alert the user should the stacks exceed maximum temperature levels. The controller will take the necessary steps to bring its temperatures down to acceptable levels.

What’s also interesting is reports indicates Nvidia using GDDR5X instead of the current High-Bandwidth module implementation (judging by the size of its BGA chip) on the GP104 Pascal core. It’s still unclear which direction Nvidia will take for other Pascal graphic card variants, but we know the AMD has signed up with Samsung to significantly ease the manufacturing process for its future.

Source: 1

Leave a Reply

Your email address will not be published.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

Beyond News. Beyond Reviews. Beyond Guides & Recommendations.

Join the never ending discussions on never ending topics.
Previous Article

Nvidia Pascal GP104 core may use GDDR5 (or GDDR5X) instead of HBM

Next Article
amd bapco

AMD takes a jab at SYSmark 2014 benchmark, deems it unreliable

Related Posts