JEDEC Looks To Push GDDR Limits

Article I just found on GDDR5 upgrade to GDDR5X

"The JESD232 Graphics Double Data Rate (GDDR5X) SGRAM standard is designed to meet the increasing need for more memory bandwidth in graphics, gaming, compute, and networking applications, according to Micron's Michael Richter, who helped develop the standard under the auspices of JEDEC."


" JEDEC wanted to leverage the existing ecosystem for GDDR5 so that it wasn't a radical change, said Richter. For example, it will be an easy transition from GDDR5 to GDDR5X for target applications such as network gear. “It's not a radical change. A major goal by JEDEC was to make as few changes as possible," he said. That's why the new specification isn't dubbed GDDR6 or 7.

GDDR5X specifies key elements related to the design and operability of memory chips for applications requiring very high memory bandwidth, targeting data rates of 10 to 14 Gb/s, which is double that of GDDR5, said Richter. However, it uses the same, proven pseudo open drain (POD) signaling as GDDR5 to allow for a smooth transition from its predecessor. “Over time we figured out we can run this memory a lot faster." "

http://www.eetimes.com/document.asp?doc_id=1328804

 

However, the tech is essentially still prone to large power draw, and looking forward, I don't see it surviving past this or maybe next year at the latest.

As gpu's increase in performance, current high-end hardware will end up in lower end sections, plus when you take into account that the industry is shifting towards less space use in mobile environments (and even desktops), the only proper way forward is to go with HBM due to power savings, massively higher bandwidth (which is essential for integrated gpu's btw) and of course much lower space requirements.

Regular GDDR5 would do the trick, while the money would be far better spent investing in HBM which has already debuted and will be mass produced this year - so, obviously, we shouldn't really be having any issues.

I can see GDDR5 being used in GPU's that don't need large bandwidth, but with space requirements going down (not up), the industry might just end up phasing out GDDR5 and GDDR5X that much faster.

 

HBM FTW!

 

GDDR5X Puts Up a Fight Against HBM, AMD and NVIDIA Mulling Implementations

There's still a little bit of fight left in the GDDR5 ecosystem against the faster and more energy-efficient HBM standard, which has a vast and unexplored performance growth curve. The new GDDR5X standard offers double the bandwidth per-pin compared to current generation GDDR5, without any major design or electrical changes, letting GPU makers make a seamless and cost-effective transition to it.

n a presentation by a DRAM maker leaked to the web, GDDR5X is touted as offering double the data-rate per memory access, at 64 byte/access, compared to 32 byte/access by today's fastest GDDR5 standard, which is currently saturating its clock/voltage curve at 7 Gbps. GDDR5X breathes a new lease of live to the ageing DRAM standard, offering 10-12 Gbps initially, with a goal of 16 Gbps in the long term. GDDR5X chips will have identical pin layouts to their predecessors, and hence it should cost GPU makers barely any R&D to implement them.

When mass-produced by companies like Micron, GDDR5X is touted to be extremely cost-effective compared to upcoming HBM standards, such as HBM2. According to a Golem.de report, both AMD and NVIDIA are mulling GPUs that support GDDR5X, so it's likely that the two could reserve expensive HBM2 solutions for only their most premium GPUs, and implement GDDR5X on their mainstream/performance solutions, to keep costs competitive.

http://www.techpowerup.com/217027/g...m-amd-and-nvidia-mulling-implementations.html