facebook rss twitter

Micron customers begin sampling GDDR5X memory chips

by Mark Tyson on 24 March 2016, 10:01

Tags: Micron (NASDAQ:MU)

Quick Link: HEXUS.net/qaczmg

Add to My Vault: x

Back in early Feb Micron wrote a blog post about its intermediary GDDR5 memory technology and how it would hit mass production by summer. At that time we heard that the first GDDR5X chips would start sampling in spring – and now, nicely on schedule, daffodils at the ready – spring has come.

The Hot Hardware team claim to have got word direct from Micron that it has indeed started sampling GDDR5X to its clients. No other source was linked, and I can't find any official mention of this milestone on the Micron site at the time of writing. The sampling stage is an important step and according to that report we probably won't hear anything else from Micron until GDDR5X memory packing graphics cards are launched as summer heats up.

Recapping GDDR5X memory capability highlights; it can offer data rates of 13Gbps to 16Gbps, it is largely backwards compatible with GDDR5 designs, and it provides improved power consumption running at a lower voltage. That's not bad for an intermediary technology, if the price is right.

As Micron is said to have both 8Gb (1GB) and 16Gb (2GB) densities available for partners to sample, these partners will be able to cobble 8GB and 16GB graphics cards, ready for summer. As time moves on HBM will largely take over graphics RAM duties in the AiB industry but there are rumours that Nvidia, at least, will kit-out some high-end and mid-high range graphics cards with GDDR5X for release this summer.



HEXUS Forums :: 11 Comments

Login with Forum Account

Don't have an account? Register today!
With HBM here and reducing power consumption noticeably, I just cannot get excited about faster GDDR5….
shaithis
With HBM here and reducing power consumption noticeably, I just cannot get excited about faster GDDR5….

I guess it's more a case of something is better than nothing - nvidia don't have HBM and will have to wait for HBM2.
But at the moment, the power savings are substantially more useful than the extra bandwidth…..have we seen any benchmarks yet where the bandwidth makes a difference?
shaithis
But at the moment, the power savings are substantially more useful than the extra bandwidth…..have we seen any benchmarks yet where the bandwidth makes a difference?

Fury X vs 390X perhaps? Guess hard to tease out the effect of more stream processors - AMD will claim they wouldn't add them if they were bandwidth limited probably.

Maybe 390x vs 290x a better comparison.
Until HBM2 is available, HBM is limited to 4GB due to interposer size limits. Your choices are 4GB of fast RAM, or 8GB+ of a-bit-slower RAM. We've seen that more RAM is a more effective speed boost than faster RAM for CPU tasks, and it's unlikely that will change for current GPU tasks.