Micron GDDR6X: after the 21 Gbps of the RTX 3090 Ti, the aim is to reach 24 Gbps

Micron he has announced the start of the volume production of the new chips 16 Gbit GDDR6X memory, used on board the new top of the NVIDIA range, the GeForce RTX 3090 Ti. With the double the capacity compared to the previous 8 Gbit solutions and able to ensure up to 15% higher performancethe new GDDR6X are set to 21 Gbps to allow the 3090 Ti to reach new levels of performance in both gaming and content creation software.

The US company, however, is already looking further and in its press release makes it known that in its roadmap there are GDDR6X memory chips up to 24 Gbps ready for “the data-hungry applications of the future”. Micron and NVIDIA have an exclusive relationship regarding GDDR6X, so it would be easy to deduce from that sentence that future GeForce RTX 4000 series solutions will push beyond 21 Gbps with peaks of 24 Gbps.

However, it is more plausible that NVIDIA chooses to stop again at 21 Gbps for its next top of the range, taking advantage of the volume production of Micron which will then already be underway for a few months. Chips beyond that speed may come later, perhaps when NVIDIA refreshes its offering with Ti / SUPER variants of the RTX 4000.

All that remains is to wait for new rumors or directly official information from NVIDIA. Regarding the GDDR6X it is good to remember that compared to GDDR6 they differ in the use of four-level pulse width modulation, better known as PAM4which will also be used in other standards such as PCIe 6.0 and DDR6.

Using PAM4, the GDDR6Xs are able to transfer data much faster, doubling the speed. Instead, the new technique applies four distinct signal (or rather, voltage) levels to transmit two bits of data to and from memory in a single cycle. As a result, each GDDR6X chip offers 84 GB / s bandwidth, which translates to nearly 1 TB / s on the RTX 3090.

The doubling of the data transmission speed in the channel occurs without doubling the required operating frequency, but not only: the GDDR6Xs consume less per transaction (pJ / bit) than the previous memories, and also have the ability to scale consumption towards the high or low, thus allowing precise control over overall performance and consumption. Precisely we are talking about 7.25 pJ / bit against 7.5 pJ / bit of the GDDR6. According to Micron, a 21Gbps GDDR6X requires 15% less power per bit transferred than a 14Gbps GDDR6.