Earlier this year Micron announced that they had begun shipping samples of GDDR5X, a natural evolution of the current GDDR5 VRAM standard, to it’s partners, namely AMD and NVIDIA. Just now they’ve confirmed that they’ve entered mass production for those partners. That means that the forthcoming GPUs that’ll feature the new higher bandwidth memory technology won’t be held back by memory production issues.
GDDR5X is hitting high yields, meaning it won’t be the weakest link in the supply line
The increase in performance with GDDR5X is actually not insignificant, coming very close to the bandwidth offered by HBM while on a smaller memory bus and while using a similar amount of power. In addition, this technology allows for memory configurations larger than 4GB, which is seen as being limiting despite that not necessarily always being strictly the case.
VRAM Memory Technology
|Example||NVIDIA GeForce GTX 980 Ti (GDDR5)||AMD R9 290X (GDDR5)||AMD R9 Fury X (HBM)||HBM2 (Theoretical)||NVIDIA GeForce GTX 1080 (GDDR5X)|
|Bandwidth Per Pin||7Gb/s||5Gb/s||1Gb/s||2Gb/s||14Gb/s|
|Chips per Stack||12||16||4||4||8|
|Estimated Power Consumption||31.5W||30W||14.6W||???||20W|
NVIDIA thus far is the first manufacturer to officially release the specs of their newest product, their flagship making good use of GDDR5X to help feed the memory of their Pascal-based GTX 1080. AMD, too, is rumored to be using GDDR5X primarily for their flagship Polaris 10 based card instead of HBM so they can include higher memory capacities at similar speeds to the first generation of HBM. The former technology was partly responsible for some of the supply issues surrounding the release of AMD’s Fury X cards.
Thankfully this news is rather positive, leading us to fully support the notion that the Polaris line of cards, in addition to Pascal-based cards, are still on time for a June and end of may launch time-frame respectively.