Micron To Offer 2X GDDR5 Speed In 2016

I’m still fuzzy about the relationship between GDDR5X and GDDR6. A lot of reports seem to suggest that the latter is merely the former.

With speeds from 10 to 14 Gb the next generation memory will be much faster and provide much needed bandwidth. Micron can easily call this memory GDDR6 as we have heard people from the graphics industry already using the term. The new memory will continue to use traditional component form factor, similar to GDDR5, reducing the burden and complexity of design and manufacturing.

Great Scott!

That's not awfully exciting knowing what's coming h1 2016

Going by public slides released months ago, Pascal seems to use 4 stacks of HBM2, with each stack providing 204GB/s bandwidth (816GB/s via 1024-bit wide interface total).

By comparison, on a traditional memory interface, a 256-bit wide GDDR5X memory subsystem using 8 of those 14Gb chips would have about 896GB/s bandwidth. One of the benefits of GDDR5X is the price advantage, so the picture isn't as cut and dried as you suspect.

It's unlikely that HBMx memory will be used across entire product lines, so GDDR5X could have other uses in mid-range cards and other applications (frame buffer for embedded devices and other uses).
----

I looked at micron's web site and this is called GDDR5X. It's based on GDDR5 and isn't the same as "GDDR6" whenever that is finalized.

^ jeebus, that's a fuckton of bandwidth.

As we've seen in the tests of the Fury here on HardOCP, that much bandwidth currently only becomes needed on settings that are not yet playable, even with 2 Furys in Crossfire:

"This is the tipping point, this is the point at which the higher bandwidth on the HBM memory is finally showing us what its potential can do in terms of performance. The problem is of course performance is nowhere near playable at this ridiculously intense setting."

http://www.hardocp.com/article/2015/10/06/amd_radeon_r9_fury_x_crossfire_at_4k_review/3

This is why you'll only see HBM on the flagship cards from both Nvidia and AMD that have a hope of actually utilizing it. It's wasted on anything less.

When will they finally move from that aging 28nm process? Its hard to believe all the smartphone soc's are hogging the lines.

gxp500 said:

When will they finally move from that aging 28nm process? Its hard to believe all the smartphone soc's are hogging the lines.

Click to expand...

I read that the AMD Arctic Island GPUs coming next year (which will use HBM2) are gonna be 14nm/16nm FinFET with 3D-stacked transistors. http://wccftech.com/amd-1416nm-arctic-islands-launching-summer-2016/

And I think nVidia said Pascal is gonna be 16nm (FinFET as well, I think).
http://wccftech.com/nvidia-pascal-volta-architecture-gp100-4k-60fps-graphic-card/

So yeah it looks like we are finally fucking done with 28nm. Maybe next year we'll actually see single GPUs capable of running games at 4K with max in-game settings at very good framerates.

DrezKill said:

I read that the AMD Arctic Island GPUs coming next year (which will use HBM2) are gonna be 14nm/16nm FinFET with 3D-stacked transistors. http://wccftech.com/amd-1416nm-arctic-islands-launching-summer-2016/

And I think nVidia said Pascal is gonna be 16nm (FinFET as well, I think).
http://wccftech.com/nvidia-pascal-volta-architecture-gp100-4k-60fps-graphic-card/

So yeah it looks like we are finally fucking done with 28nm. Maybe next year we'll actually see single GPUs capable of running games at 4K with max in-game settings at very good framerates.

Click to expand...

So at E# they should start showing 16K