Samsung announces new GDDR6W memory to compete with HBM2

In the context: As manufacturers continue to squeeze every last drop of performance out of existing GDDR6 and GDDR6X memory modules, Samsung has announced a new and improved entry in the family, GDDR6W. Samsung claims that GDDR6W can compete with the bandwidth and speed of HBM2.

In 2016, Samsung and other manufacturers began manufacturing successors to fast (but imperfect) high bandwidth memory modules (HBMs). High Bandwidth Memory 2 (HBM2) seems to have solved all the problems with the previous generation by increasing capacity, speed and bandwidth. Unfortunately, the HBM2 never saw significant success in the desktop video card market.

The Fury and Vega card lines used HBM and HBM2 respectively. Unfortunately, each of them failed, and AMD returned to GDDR6 memory again, starting with the RX 5000 line. Some users were understandably disappointed with the rapid abandonment of HBM2.

Tuesday Samsung disclosed its latest entry in the GDDR6 family, GDDR6W. The South Korean tech giant wanted to bring some of the benefits of HBM2 to the already successful GDDR6 platform, especially the increased bandwidth. Based on the information and numbers that Samsung provides, GDDR6W could be a game changer in future GPUs.

Samsung is focusing heavily on VR and Metaverse applications. However, there’s no reason why GDDR6W won’t benefit future discrete graphics cards in general.

Samsung started by taking its existing GDDR6 platform and implementing what it calls “fan plate level packaging” (FOWLP). Instead of placing memory chips on a printed circuit board, they are mounted directly on a silicon wafer. The redistribution layers provide “thinner wiring patterns” and since the PCB is not involved, the modules will be thinner overall and better at dissipating heat.

The memory modules are dual-stacked, which Samsung says doubles the capacity and doubles the memory bandwidth. The memory limit is 24GB per die for standard GDDR6, so the new platform could theoretically support 48GB on a single graphics card.

These changes in module placement and overall die size have resulted in GDDR6W being 36% shorter than its GDDR6 counterpart. Due to the unchanged footprint, these modules can be “plugged into the same manufacturing processes” used in current GDDR6 products.

As seen above, the throughput of GDDR6W is very close to that of HBM2E chips. The current bandwidth limit of GDDR6X is around 1 TB per second, and GDDR6W greatly increases it by about 400 MB/s.

While we almost certainly won’t see GDDR6W on any Lovelace or RDNA3 graphics card, there’s always a chance Nvidia or AMD will swoop in and adopt GDDR6W in the next generation. However, Samsung’s new creation is a big step forward for GPUs, whether discrete or “for the Metaverse”.

Source link

Leave a Reply

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

Back to top button