GDDR6X graphics memory 21 Gbps and high capacity from Micron from 2021

GDDR6X graphics memory 21 Gbps

The launch of the new Nvidia GeForce RTX 3000 is also a milestone for Micron. For the first time, GDDR6X is used, which is the successor to GDDR6 with higher clock rates and ultimately more bandwidth. In the coming year, faster patterns and double capacity are on the agenda.

The first chips are only 1 GByte in size, so that many have to be used on the flagship solutions. The GeForce RTX 3090 therefore not only uses twelve positions on the top, but also has to take twelve places on the back in order to be able to provide 24 GB of graphics memory. The next professional solutions of the Quadro family on the basis of the ampere architecture should offer up to 48 GByte GDDR6X - this is only possible with correspondingly larger memory chips.

GDDR6X Speed and bandwidths in development

In terms of speeds, what has been around for weeks in the rumors is ultimately true. Nvidia uses conservative 19.5 Gbps for its first solutions, Micron has specified the solutions with 19 to 21 Gbps. With the latter specification, the mark of 1 TB/s memory bandwidth falls on the appropriate memory interface. The voltage of the same size memory chips as GDDR6 is 1.25 to 1.35 volts as before. Nevertheless, the chips should consume less energy at 50 percent more bandwidth.

GDDR6X power consumption

GDDR6X power consumption

For the first time, Micron uses the so-called PAM4 method (4-level pulse amplitude modulation), which will also be used in classic RAM in the future and replaces the PAM2 method (also called NRZ). There are four voltage levels for encoding and two bits per clock can be transmitted, effectively doubling the bandwidth at the same frequency. In other words, the predecessor GDDR6 would have to clock twice as high in the classical method in order to enable the same data rate per pin.

PAM4 doubles the bandwidth internally

GDDR6X with PAM4 compared to GDDR6 with PAM2

It’s not the first time that Nvidia and Micron have joined forces to advance a storage standard. Already GDDR5X was a solution of the two manufacturers, which Nvidia could use exclusively on their cards. The extent to which such a clause also exists with GDDR6X is not yet known. But at least a certain time frame of exclusive availability for Nvidia is more than likely.