Micron in its latest research article has revealed the full potential of its upcoming GDDR6 memory, which will be equipped with the latest graphics cards from GPU manufacturers, especially NVIDIA. GDDR6 memory has been in Micron’s labs for quite some time now and seems to be almost ready to take its place on mass-produced graphics cards.
GDDR6
Micron analyses various aspects of GDDR6 memory, including silicon changes, channel enhancements and performance measurements. While GDDR6 follows an evolutionary path from GDDR5 and GDDR5X memory, there are still some significant changes in the underlying architecture to increase memory bandwidth while saving power.
One striking detail is that Micron has already determined in performance measurement that its GDDR6 memory could extend beyond the 16.5 GHz range. The result showed that with a slight but useful increase in I/O voltage, memory chips can reach speeds of up to 20 GHz, a significant jump from the 14 GHz target defined by JEDEC.
If we calculate, a 256-bit card with these speeds would be able to offer a bandwidth of 640 Gb/s that is close to the 652.8 Gb/s of Titan V, which uses HBM2 memory. A 384-bit card would almost reach the 1 Tb/s barrier with a bandwidth of approximately 960 GB/s, surpassing the NVIDIA Tesla V100 solution.
Micron mentions that they are very confident that the speeds that GDDR6 will offer will extend beyond the speeds defined by Jedec (14 GHz), so this is only an indication of the potential for overclocking that we would see in next-generation graphics cards.
Source
Wccftech