Room For Innovation On GDDR5: Higher Density, Faster Data Rates Coming With 'GDDR5X'

With the adoption of 4K monitors on the rise, and the release of consumer-grade virtual reality just around the corner, the need for larger amounts of faster memory has never been so high. Graphics cards have been released with ever-increasing frame buffers year over year to accommodate high resolutions. Graphics cards today come equipped with as much as 8 GB, and in some cases even 12 GB of memory. Until now, 4 Gb modules were the highest density available, forcing manufacturers to work with limited frame buffers, or build large PCBs to house all the memory modules.

With today's announcement, Micron has released memory chips with 8 Gb density, effectively doubling the maximum memory capacity of upcoming graphics cards and other products that use GDDR5 memory, such as game consoles.

On the other hand, it is now also possible to reduce the size of board needed, while maintaining the same amount of graphics memory, making it possible to design cards that take up a smaller footprint.

In addition to launching higher density GDDR5 modules, Micron also hinted at what's to come in 2016. Currently, 7 Gbps GDDR5 is what we see on many graphics cards offered today. Micron said that 8 Gbps modules are currently in production, but it sees this as the absolute peak for GDDR5 in its current form. The company said it observed that command address protocoling and array speed were the two limiting factors, while the interface had plenty of additional headroom.

In order to surpass the 8 Gbps barrier of GDDR5, without completely building a new memory technology from the ground up, Micron doubled the pre-fetch of GDDR5 from eight data words for each memory access, to 16 data words per access. Doing this resulted in 10 to 12 Gbps on the first generation, and the company expects to be able to surpass 14 Gbps, and perhaps reach 16 Gbps and beyond as the technology is refined.

Micron will make a formal announcement about this new memory technology some time in 2016, but what we know so far is that the company is calling it "GDDR5X," and it will be significantly faster than current offerings. The company wanted to make adoption as simple as possible, so it retained as much of the GDDR5 commands and protocols as possible. Additionally, Micron is not keeping this as a proprietary option, and has instead approached JEDEC to make GDDR5X a widely available standard.

8 Gb GDDR5 modules are available now for hardware manufacturers. Micron expects availability of products using GDDR5X to start hitting the market in the second half of 2016.

Follow Kevin Carbotte @pumcypuhoy. Follow us @tomshardware, on Facebook and on Google+.

 Kevin Carbotte is a contributing writer for Tom's Hardware who primarily covers VR and AR hardware. He has been writing for us for more than four years. 

  • boju
    Another round of referbs that actually matter for a change? before Pascal's 3D chips arrive which will be 9x faster than standard GDDR5 apparently.
    Reply
  • ragenalien
    This will be good for keeping the cost down and performance up on some of the lower/mid range graphics cards. But HBM has so much more potential for the high end.
    Reply
  • TallestJon96
    I expect to see a lot of memory segmentation. Lowest end gets ddr3, low to mid range gets gddr5, slightly higher gets gddr5x, and the highest end gets HBM2. Hopefully in the next year or two, ddr3 gets dropped, and we have a simple gddr5 for the low to mid end, and HBM 2 for the higher end.
    Reply
  • airborn824
    Another round of referbs that actually matter for a change? before Pascal's 3D chips arrive which will be 9x faster than standard GDDR5 apparently.

    Pascal is going to use HBM which is already available on AMD Fury and NANO GPUs. IT is a new stackable vRAM that was developed by AMD and Hynix
    Reply
  • deksman
    This attempt at restructuring GDDR5 comes a bit late.
    Should have been released before.
    AMD already transitioned to HBM (which is far superior to GDDR5 in terms of saving the physical space reducing the GPU size, as well as lowering power consumption whilst offering huge increases in speed and bandwidth).

    I might only see this being used in Nvidia's Pascal, but even that is questionable as Nvidia has intentions of releasing Pascal with HBM2 (once it becomes available to them).

    On that note, AMD has priority access to HBM2 (much like with current HBM1) which is in limited supply, and will be used on their new GPU's with (finally) new architecture and lower manuf. process.

    On that end, Nvidia will have to wait until Hynix releases sufficient amount of HBM2 to them, which probably means delays for Pascal.
    Alternatively, in order to release 'something' next year in any appreciable time frame, Nvidia might either use GDDR5X for Maxwell rebrands, or will possibly use them for Pascal.

    Since Nvidia has more monetary resources than AMD, I can see them using GDDR5X as a way to release Maxwell rebrand cards to have something on the market, and then finally Pascal (once HBM2 becomes available to them).
    Reply
  • CRITICALThinker
    16559657 said:
    I expect to see a lot of memory segmentation. Lowest end gets ddr3, low to mid range gets gddr5, slightly higher gets gddr5x, and the highest end gets HBM2. Hopefully in the next year or two, ddr3 gets dropped, and we have a simple gddr5 for the low to mid end, and HBM 2 for the higher end.

    Somehow I feel that we will be seeing DDR4 before we see GGDR5 on the low end. This is good for midrange though, as it looks as if we may be seeing some higher amounts of memory on those cards.

    Reply
  • boju
    16560080 said:
    Another round of referbs that actually matter for a change? before Pascal's 3D chips arrive which will be 9x faster than standard GDDR5 apparently.

    Pascal is going to use HBM which is already available on AMD Fury and NANO GPUs. IT is a new stackable vRAM that was developed by AMD and Hynix

    Oh cool, I thought this kind of vram wont be available until next gen or at least the era of Pascal, didn't realise it's already here. I still had my head in the GDDR5 sandbox, thanks for pointing that out.
    Reply
  • renz496
    16560701 said:
    This attempt at restructuring GDDR5 comes a bit late.
    Should have been released before.
    AMD already transitioned to HBM (which is far superior to GDDR5 in terms of saving the physical space reducing the GPU size, as well as lowering power consumption whilst offering huge increases in speed and bandwidth).

    but HBM is very expensive right now. and looking ad Fury X and 980Ti that massive bandwidth provided by HBM are not an advantage right now. this could be cheaper alternative without sacrificing performance.

    16560701 said:
    On that note, AMD has priority access to HBM2 (much like with current HBM1) which is in limited supply, and will be used on their new GPU's with (finally) new architecture and lower manuf. process.

    that priority access probably will not be much of an issue to nvidia due to this:

    http://www.kitguru.net/components/graphic-cards/anton-shilov/samsung-expects-graphics-cards-with-6144-bit-bus-48gb-of-hbm-memory-onboard/
    Reply
  • whassup
    With the emergence of HBM GDDR5 is obsolete in almost every feature. Still GDDR5 can be used to replace the GDDR3 present in some of the entry level graphics cards. The GDDR3 should be completely flushed out from the GPU segment.
    Reply
  • HyperMatrix
    Not a bad idea. GDDR5 still beats out HBM for memory latency. At 12gbps, it should be more than sufficient for 4K gaming. And may actually perform better than HBM due to the difference in architecture (slow and wide vs. fast and narrow). Especially on a 384-bit bus like the 980ti/Titan X.
    Reply