Micron Confirms Mass Production of GDDR5X Memory

[Megalith]

Early adopters of the GeForce GTX 1080 will be the first to experience this new type of graphics DRAM. Original samples operated as high as 12 GB/s, but specifications now point toward a memory clock of 10 GB/s.

Micron’s first production GDDR5X chips (or, how NVIDIA calls them, G5X) will operate at 10 Gbps and will enable memory bandwidth of up to 320 GB/s for the GeForce GTX 1080, which is only a little less than the memory bandwidth of NVIDIA’s much wider memory bus equipped (and current-gen flagship) GeForce GTX Titan X/980 Ti. NVIDIA’s GeForce GTX 1080 video cards are expected to hit the market on May 27, 2016, and presumably Micron has been helping NVIDIA stockpile memory chips for a launch for some time now.

 

[westrock2000]

I thought HBM2 was forecasted to go up to 1TB/sec (possibly 2TB), where as we are hitting the upper limits of GDDR5.

 

[Ultima99]

I thought HBM2 was forecasted to go up to 1TB/sec (possibly 2TB), where as we are hitting the upper limits of GDDR5.

This has nothing to do with HBM.

 

[Deleted member 93354]

I thought HBM2 was forecasted to go up to 1TB/sec (possibly 2TB), where as we are hitting the upper limits of GDDR5.

It is. We hit the limits of gddr5. Gddr5x is a new chip design which allows much high throughput with same more latency.

 

[evilsofa]

The cost of HBM2 will be extreme for awhile, and the AMD Fury showed that we have only just barely hit the limit of GDDR5; in HardOCP's tests, Fury's HBM showed benefits only at game settings that were unplayable. GDDR5X will be enough bandwidth for those very few cards that need more than GDDR5 bandwidth until HBM2 becomes more cost effective.

 

[DejaWiz]

I love Micron/Crucial. So contributing to the market, yet seemingly underappreciated.

 

[Brent_Justice]

HBM has the potential for a lot more bandwidth, but HBM 1 was not utilized to its maximum potential in the Fury/FuryX, bandwicth was only 512GB/s far shy of the 1TB/s it is capable of, and far shy of HBM 2 which can go up to 2TB/s if utilized.

GDDR5X has an advantage in clock speed, where HBM has an advantage in the amount of bandwidth per pin. Remember it is bus width x clock speed that equals bandwidth. If you matted 10GHz GDDR5X on a 384-bit bus you'd have 960GB/sec of bandwidth, almost HBM 1 full bandwidth speeds.

So really what matters is how its used, what bus width it is matted to, and how well that bandwidth is utilized. Typically you are engine bottle necked, not memory bandwidth limited in GPUs.

 

[doubletake]

Bent, your math's a bit off; 10GHz GDDR5X on a 384 bit bus would only be 480GB/s.

 

[DejaWiz]

Bent, your math's a bit off; 10GHz GDDR5X on a 384 bit bus would only be 480GB/s.

Isn't it stacked? 480 * 2 = 960?

 

[doubletake]

Then how would the 1080 only be 320GB/s? The specs are 10GHz GDDR5X on a 256-bit bus.

 

[morningreis]

I thought HBM2 was forecasted to go up to 1TB/sec (possibly 2TB), where as we are hitting the upper limits of GDDR5.

Yes it is, but the cost and availability of HBM2 is bad, so Nvidia's answer was to have GDDR5X developed as a substitute. Not as good as HBM2, but still far better than GDDR5.