Article 1RCJW HBM3: Cheaper, up to 64GB on-package, and terabytes-per-second bandwidth

HBM3: Cheaper, up to 64GB on-package, and terabytes-per-second bandwidth

by
Mark Walton
from Ars Technica - All content on (#1RCJW)
hbmchips-800x450.jpg

Enlarge

Despite first- and second-generation High Bandwidth Memory having made few appearances in shipping products, Samsung and Hynix are already working on a followup: HBM3. Teased at the Hot Chips symposium in Cupertino, Calfornia, HBM3 will offer improved density, bandwidth, and power efficiency. Perhaps most importantly though, given the high cost of HBM1 and HBM2, HBM3 will be cheaper to produce.

With conventional memory setups, RAM chips are placed next to each other on a circuit board, usually as close as possible to the logic device (CPU or GPU) that needs access to the RAM. HBM, however, stacks a bunch of RAM dies (dice?) on top of each other, connecting them directly with through-silicon vias (TSVs). These stacks of RAM are then placed on the logic chip package, which reduces the surface area of the device (AMD's Fury Nano is a prime example), and potentially provides a massive boost in bandwidth.

9bb6b7df-6be9-45fe-a2d1-6e4580a309f1-640

The tradeoff, though, as with most fancy packaging techniques, has been price and capacity. HBM1, as used in AMD's Fury graphics cards, was limited to 4GB stacks. HBM2, as used in Nvidia's workstation-only P100 graphics card, features higher density stacks up up to 16GB, but is prohibitively expensive for consumer cards.

Read 6 remaining paragraphs | Comments

index?i=dF4hM-xNMCU:_8ZwenTX2q4:V_sGLiPB index?i=dF4hM-xNMCU:_8ZwenTX2q4:F7zBnMyn index?d=qj6IDK7rITs index?d=yIl2AUoC8zA
External Content
Source RSS or Atom Feed
Feed Location http://feeds.arstechnica.com/arstechnica/index
Feed Title Ars Technica - All content
Feed Link https://arstechnica.com/
Reply 0 comments