GDDR7 capacity stagnates at only 2GB! First 3GB in the future!

Recently, the JEDEC organization officially announced the GDDR7 video memory technical specification, which is a significant improvement in all aspects of the indicators, but I didn't expect it to stagnate in terms of capacity density, so I can only look forward to the future.

GDDR7 is upgraded to a quad-channel transfer architecture, increasing bandwidth per pin to 32-48Gbps, equivalent to a full 2-3 times that of GDDR6/6X, with up to 1.5TB/s of bandwidth at 256-bit bitwidth, and also supports on-chip ECC, while the voltage has been lowered from 1.35V to 1.2V to further save power consumption.

In addition, the signal modulation has been reduced from PAM-4 to PAM-3 to reduce the burden, and the package has been changed to a 266 FBGA.

The first batch of GDDR7 memory will have a single capacity of only 2GB (16Gb), exactly the same as today's GDDR6/6X, so the NVIDIA RTX 50 series and AMD RX 8000 series, which will be the first to carry it, will still need quite a bit of memory to reach the mega capacity.

Luckily, JEDEC is also planning for higher capacity densities, with 3GB, 4GB, 6GB, and even 8GB coming one after another in the future, with 3GB being the first time such an anticlimactic capacity has ever appeared.

However, when mass production will depend on the progress of giants such as Samsung, Micron, SK Hynix, and the willingness of NVIDIA and AMD to adopt.

It is worth mentioning that the GDDR6 era has also been designed with a single 1.5GB capacity, but it has never been mass-produced.

GDDR5/6/7 Channel Architecture Diagrams

GDDR7 On-Chip ECC Flow Schematic

GDDR7 Pin Definition Schematic