Nvidia May Be Working With SK Hynix On GDDR6 Graphics Card Memory For Early 2018 Mass Production

sk hynix gddr6 early 2018 nvidia volta — Image used with permission by copyright holder

SK Hynix is the latest to introduce a GDDR6 product, stating that it will begin mass producing an 8Gb memory chip by early 2018. Samsung said something similar during the recent Hot Chips convention, stating that the mass production of its GDDR6 chips wouldn’t begin until 2018. They would have an I/O rate of 14 gigabits per second per pin, and a lower power consumption than GDDR5X technology, which only has a data rate of 12 gigabits per second. Micron is also pushing to mass produce a GDDR6 product by the end of the year.

SK Hynix said that its upcoming 8Gb memory chips will be manufactured using 2Xnm process technology, meaning anywhere between 20nm and 29nm. Each chip will also have an I/O data rate of 16 gigabits per second per pin, making it the “world’s fastest” to date. Throw in a 384-bit memory interface, and the GDDR6 tech will have the ability to process up to 768GB of graphics data per second. As a reference, GDDR5X memory used on the GeForce GTX 1080 Ti has an I/O transfer rate (bandwidth) of under 500GB per second.

“GDDR6 is a next generation graphics solution under development of standards at JEDEC, which runs twice as fast as GDDR5 having 10 percent lower operation voltage,” the company said. “As a result, it is expected to speedily substitute for GDDR5 and GDDR5X. SK Hynix has been collaborating with a core graphics chipset client to timely mass produce the GDDR6 for the upcoming market demands.”

Get your weekly teardown of the tech behind PC gaming

Current memory chips based on GDDR5 have 170 pins, 67 of which are specifically used for the I/O data exchange while the others are for grounding and power purposes. The more recent GDDR5X builds on that design by adding 20 pins to the form factor, providing an additional I/O boost. SK Hynix’s GDDR6 appears to have 180 pins.

The major difference between GDDR (Graphics Double Data Rate) and HBM (High Bandwidth Memory) is similar to what divides regular NAND and 3D NAND. On the GDDR5 front, chips are spread out horizontally in physical space, limiting the actual size of the graphics card and requiring “larger voltage circuity.” HBM tech takes the skyscraper approach by stacking memory chips vertically, which in turn provides better I/O bandwidth.

As for that “core graphics chipset client” comment, we’re betting it’s Nvidia. The company’s next-generation graphics chip architecture, code-named Volta, will supposedly use GDDR6 memory technology in 2018. The current crop of Pascal-based “GeForce” graphics cards for the gaming market use GDDR5 and GDDR5X while Nvidia’s “Tesla” solutions for the enterprise sector use HBM2 tech.

Meanwhile, AMD fully backs HBM technology, which was used on its Fuji-based graphics cards for the gaming market in 2015. AMD is also getting ready to unleash new graphics cards under its “Vega” chip design umbrella packing HBM2 on-board memory, and its sixth-generation “Navi” solutions could do the same. Yet AMD’s current Polaris-based value-priced RX 400 and RX 500 family of cards for the “mainstream” market relies on GDDR5, so we will have to play the wait-and-see game regarding the Navi-based lineup.