Samsung has introduced its newest, third-generation HBM2 (second-gen excessive bandwidth reminiscence) at GTC 2019, which it calls Flashbolt. Capable of three.2Gb/s switch speeds and capability as much as 16GB in a single stack, Samsung’s HBM2E reminiscence dwarfs earlier generations of HBM2 reminiscence, and makes even the AMD Radeon VII’s quad-stack 16GB HBM2 operating at 2Gb/s appear somewhat civilian by comparability.
Despite a comparatively small footprint in comparison with GDDR5/6 merchandise, there are as much as eight stacks of reminiscence (8-Hi) inside each single HBM2 bundle, linked with by silicon vias (TSVs). AMD utilised HBM2’s tiny bundle measurement with its Vega structure, becoming 8GB of the stuff up into the RX Vega 64 and RX Vega 56. It then adopted that up in February with the launch of the Radeon VII, doubling reminiscence capability of its forebears to 16GB of HBM2. This was achieved with 4 reminiscence stacks (4-Hi a bit).
Yet that might be achieved with only one 8-Hi Samsung Flashbolt HBM2E bundle. Each die is able to an amazing 16Gb capability, stack up eight of ‘em and you’ve bought a 16GB bundle managing 410 GB/s information bandwidth.
“Flashbolt’s industry-leading performance will enable enhanced solutions for next-generation data centers, artificial intelligence, machine learning, and graphics applications,” Jinman Han, senior VP of reminiscence product at Samsung, says. “We will continue to expand our premium DRAM offering, and improve our ‘high-performance, high capacity, and low power’ memory segment to meet market demand.”
On goal: These are the best gaming mice round
But HBM2 is costly. AMD already fell foul to reminiscence pricing woes when attempting to roll out client focused HBM2 graphics playing cards, solely just lately managing to only about thread the value/efficiency needle with the Radeon VII. As such, this high-performance reminiscence tech garners essentially the most love from the burgeoning skilled GPU and IC market.
Nvidia stuffed 12GB of HBM2 inside the Titan V, however has by no means taken the plunge on the reminiscence tech within the mainstream market. Instead the inexperienced group opted for cheaper and lots succesful sufficient GDDR6 packages inside the RTX 20-series and 16-series graphics playing cards, with the only real exception of the GTX 1660 on GDDR5.
It was simply over a 12 months in the past that Samsung rolled out second-gen HBM2 nicknamed Aquabolt: 8GB HBM2 operating at 2.4Gb/s for a complete of 307.2GB/s. That was already a major step up over its first gen reminiscence, which needed to break HBM2 spec and improve voltage as much as 1.35V to interrupt the 2Gb/s barrier. So far, Samsung has stored quiet on Flashbolt voltages, however it might have to bump voltages as soon as once more to hit 3.2Gb/s.
Source
