Micron announces mass production of GDDR6 RAM

The availability of a new type of memory chip is often a harbinger of new graphics processors. Micron's announcement that the company has started volume production of next-generation GDDR6 memory could be a precursor to the next go-round of cards for gamers and for data scientists in the fields of deep learning and AI. The company specifically says it has begun mass production of 8-Gb memory chips (1 GB). The company's catalog shows that initial GDDR6 RAM will have speeds ranging from 10 to 14 Gbps per pin at voltages from 1.25 to 1.35 V. An eventual 16-Gbps version will target up to 64 GB/s of bandwidth from one package.

Just like every other time the number after the letters “GDDR” has gone up, Micron is saying the fresh silicon delivers a significant performance boost and more bandwidth compared to the previous generation of graphics RAM. The application of the new silicon to graphics cards and deep-learning hardware is inevitable, but the company also says that high-end networking hardware, RAID controllers, and automotive systems will also take advantage of the increased memory bandwidth provided by GDDR6. The manufacturer goes on to tout the 448 GB/s available in auto-qualified GDDR6 configurations, a figure the company says is more than double what's provided by current “LPDDR5” memory systems.

The company says it has been sampling GDDR6 memory chips to its partners so they can get a start on designing controllers and accessory components for the new memory type. Rambus, every Pentium 4 owner's favorite memory IP company, is the first company with a GDDR6 PHY ready for integration into other products.

Micron is showing off its brand-new +6 Chips of Graphics Remembering in its booth at the ISC High Performance show in Frankfurt this week. The company didn't speculate on when gamers and data scientists might expect to lay hands on products containing the new enchantment, but we wouldn't be surprised if something will hit the PC market before the end of the year. That said, Nvidia CEO Jensen Huang has said that the company's next-gen GeForce announcement won't be for “a long time,” so pinning the tail on the launch donkey could be a fool's errand.

In any case, SK Hynix and Samsung have already announced mass production of GDDR6 memory, so Micron's relatively late ramp seems unlikely to impede any consumer products designed around this RAM that were already in the pipe. We've all heard the rumors that Nvidia will broach the topic of a new generation of GeForce cards later in August, and that may be where we first see this RAM as part of a shipping product. Only time will tell.

Comments closed
    • ronch
    • 1 year ago

    Ok cool. Now about those exorbitant RAM prices…

    • techguy
    • 1 year ago

    14Gbps GDDR6 isn’t fast enough to allow GTX 1180 to equal or surpass GTX 1080 Ti in the bandwidth department so we can throw those earlier rumors about “512GB/s bandwidth” right out the door as it would require 16Gbps GDDR6.

    14Gbps x 256-bit bus = 448GB/s bandwidth
    11Gbps x 352-bit bus = 484GB/s bandwidth

      • BobbinThreadbare
      • 1 year ago

      Possibly video card companies will use HBM exclusively for high end cards with GDDR6 for mid range and probably keep using DDR3 for low end

        • techguy
        • 1 year ago

        Zero chance Nvidia uses HBM2 on a desktop card not named Titan.

      • Srsly_Bro
      • 1 year ago

      Made up specs for an unreleased part for specs needed on an unreleased part.

      You forgot the disclaimer. Generally needed when transitioning between make-believe land and IRL.

        • techguy
        • 1 year ago

        The purpose of the comment was to indicate the previously “leaked specs” are in fact, bogus. Not sure why you would decry this action as though it is spurious.

          • Srsly_Bro
          • 1 year ago

          Always good to be transparent with your info, bro.

      • Waco
      • 1 year ago

      There’s not much stopping Nvidia from running them out of spec for higher speeds…but I agree, it’s unlikely the leaked specs are anything other than guesswork.

      • smilingcrow
      • 1 year ago

      Samsung have stated they have up to 18Gbps and there’s SK Hynix so who knows what speeds will be released initially for high end cards.

        • techguy
        • 1 year ago

        If Nvidia single-sources RAM from Samsung they can get what they need in order to meet the alleged leaked specs. NV hasn’t single-sourced RAM for a consumer graphics card in quite some time, to my knowledge though so I would be surprised to see this happen. More likely they will use 14Gbps chips to keep costs down and make sure there is a steady supply.

          • smilingcrow
          • 1 year ago

          You may well be right but it also depends on when they release their next cards.
          14Gbps is just the starting point and that is due to increase so if the cards are delayed 16Mbps may be available from multiple vendors.

            • Srsly_Bro
            • 1 year ago

            Nvidia released refreshed versions with faster RAM this gen. I don’t know why Nvidia wouldn’t go it again.

            • smilingcrow
            • 1 year ago

            That’s very conservative of you not to speculate on what they won’t do as speculation is generally the order of the day.

          • K-L-Waster
          • 1 year ago

          Alternatively, Nvidia could (and probably would) go with the higher priced component and pass the cost + markup on to the consumer. They aren’t shy about putting large price tags on their cards.

        • Srsly_Bro
        • 1 year ago

        [url<]https://videocardz.com/press-release/micron-begins-volume-production-of-gddr6-high-performance-memory[/url<] The press release mentions 20 Gbps. Not sure if TR saw the same press release or failed to include that part.

          • Waco
          • 1 year ago

          No, it doesn’t. It says 20 [i<]GB/s[/i<].

            • Srsly_Bro
            • 1 year ago

            You’re right. I failed at comprehension on that one. thanks, bro. +1

        • freebird
        • 1 year ago

        Yes, but it being manufactured on 10nm … SK Hynix pretty sure 16nm & Micron started 16nm GDDR5 only in 2017… and claims 1X nm for GDDR6, which I doubt will be 10nm.

        In addition, SK Hynix 12 & 14 Gbps is listed at 1.35v, but Micron is listing 10 & 12 Gbps at 1.25v and 12,13 & 14 Gbps at 1.35v; so the main stream will probably be using 12-14 Gbps @ 1.35v

        Also Samsung will be the only vendor with 18Gbps GDDR6 which means it will be VERY expensive compared to the 12-14 Gbps versions available from all 3 vendors in 2018/2019.

        Ergo, only high-end cards with high end prices will use anything faster than 14 Gbps GDDR6.

    • blastdoor
    • 1 year ago

    [quote<]Rambus, every Pentium 4 owner's favorite memory IP company[/quote<] Nice 😉 Those guys really do deserve to be punched in the arm at every opportunity for at least 100 years as punishment for their sins.

      • Growler
      • 1 year ago

      After the Pentium 4 debacle, I’m surprised they didn’t end up rebranding.

      • BobbinThreadbare
      • 1 year ago

      otoh the rest of the memory companies did collude against them

      basically all corporations are bad, no need to differentiate

        • chuckula
        • 1 year ago

        If RAMBus had played its cards right it could have gone down as one of those OMG heroic companies that fanboys always remember as being daring innovators whose magically great technology was sabotaged by “the man”.

        At least it’s still in business.

          • Wirko
          • 1 year ago

          They were the first company specialized in patent trolling. Isn’t that innovation? Others just followed.

            • Waco
            • 1 year ago

            Too bad they didn’t patent that! 😛

            • jihadjoe
            • 1 year ago

            In hindsight, there would probably be a lot less patent trolls if the original patent troll patented the model.

            • Waco
            • 1 year ago

            …that’s the joke. 😛

        • blastdoor
        • 1 year ago

        Well…. yes and no. I agree that the corporation exists to extract rents from employees and customers on behalf of shareholders and that this can lead to many bad things.

        However, the people running these corporations do vary in the extent to which the pursuit of self interest is enlightened versus unenlightened.

        Rambus was particularly unenlightened. They were trying to force people to use an inferior product and at the same time prevent them from using a better product. If they had just tried to charge a reasonable royalty for DDR instead of simultaneously trying to kill DDR and force the adoption of RDRAM, I think they would be remembered more fondly.

        If anybody could make the RAM mafia look like good guys, it was Rambus.

    • Hsew
    • 1 year ago

    And to think, 10 years ago to the day, AMD launched the HD 4870, the first consumer product to use GDDR5.

      • Waco
      • 1 year ago

      Wow. I didn’t realize it’d been that long!

      • cygnus1
      • 1 year ago

      (☉_☉)

      sweet baby jesus I feel old now.

        • Srsly_Bro
        • 1 year ago

        How do you look tho? Amirite?

        • maxxcool
        • 1 year ago

        you not old until you remember ordering a 1MB EDO ram chip and stuffing it into a expansion socket on a S3 VIRGE Card

          • psuedonymous
          • 1 year ago

          You’re not old unless you remember anticipating the new AGA chipset to replace ECS!

            • derFunkenstein
            • 1 year ago

            You’re not old unless you remember using an abacus not because you wanted to, but because electronic calculators didn’t exist.

      • cynan
      • 1 year ago

      Sadly, at least performance-wise, the HD 4800 series probably doesn’t quite “kick more ass than a pair of donkeys in an MMA cage match” anymore.

        • DoomGuy64
        • 1 year ago

        IMO, both the peak and fall of radeon, because after AMD bought ATI, the drivers and performance dropped off a cliff, not to mention developers being strong armed to not support dx10.1, and the later dx11 vliw series had serious balance issues.

        Since then, it’s taken until the 480 for AMD to get back to 4800 competitive levels.

      • Goty
      • 1 year ago

      Yikes. One of those is still pulling duty in one of my parents’ desktops.

        • cygnus1
        • 1 year ago

        I wonder if putting something newer in there could reduce power consumption/heat output

          • Goty
          • 1 year ago

          Probably, but it rarely gets turned on anyhow. My parents are firmly on the tablet/laptop train for their daily computing needs despite having what are still quite powerful desktops (the machine with the 4870 is a Haswell i7 and was used mostly for Lightroom/Photoshop work.)

        • Usacomp2k3
        • 1 year ago

        A 4850 is my most powerful dGPU that I own.

          • BobbinThreadbare
          • 1 year ago

          at this point you should probably upgrade to like a 1030 or something just for power savings

            • Kretschmer
            • 1 year ago

            How many decades of power savings would you require to “pay off” the cost of a 1030? If it ain’t broke…

            • Anonymous Coward
            • 1 year ago

            Not sure if the comment was meant seriously, but thats definitely a question you’d have to answer if you wanted maximum greenishness. I suspect you could use an old and inefficient card a good deal before the net harm to the environment was greater than the cost of a replacement component.

            • DPete27
            • 1 year ago

            HD 4850 TDP = 110W
            GT1030 TDP = 30W

            The GT1030 most likely outperforms the HD 4850

            Sure, 80W of power savings, but thermals/noise can be a big factor also…..if GT1030 levels of performance is all you need. That’s probably not much better than Intel HD630 graphics in 8th gen processors, and certainly worse than Vega8 inside the AMD Ryzen 3 2200G. Not to mention what the HD4850 is hooked to on the CPU side in terms of power usage and performance compared to the 2200G.

            • BobbinThreadbare
            • 1 year ago

            my comment was only meant half-seriously, but a new CPU, motherboard and RAM would take a lot longer to pay off from power savings

          • freebird
          • 1 year ago

          OUCH, dude… seriously OUCH. 5850s can be had for probably $10 or $20 on ebay. I probably have 2 or 3 lying around.

      • wingless
      • 1 year ago

      And I bought a 512GB version and regretted not getting the 1GB for years…

        • ermo
        • 1 year ago

        Same here, just for the HD4850.

        That card is currently enjoying life in an old Gigabyte EP45-DS3R motherboard sporting a Q9400, 4x2GB DDR2 RAM and a refurbished SSD. This is all orchestrated by macOS High Sierra which, by the magic of Clover, was tricked into believing this is an old iMac 10,1.

        I think I actually completed Portal 2 on that box if I’m not mistaken…

    • chuckula
    • 1 year ago

    This is a necessary but insufficient condition for newer GPUs so we will still be waiting a while for something from Nvidia. I personally think that some type of product will launch this year but maybe not during the summer like some of the rumors were indicating.

    Incidentally, Nvidia pulled its Hot Chips talk about its next-gen GPU, so that might be an indicator that we will be waiting longer for these chips.

    [FYI: [quote<]Micron is showing off its brand-new +6 Chips of Graphics Remembering [/quote<] I'm guessing that's "Rendering" instead of "Remembering" although considering it's about memory maybe "Remembering" is the right word.]

      • sweatshopking
      • 1 year ago

      Yeah, without serious competition there is no reason to release new products. AMD hasn’t caught them yet, so they don’t need to move on new products.

        • chuckula
        • 1 year ago

        Whether Nvidia wants to release a product or not, it won’t be able to if the GDDR6 isn’t available*. That is short of dumping GDDR6 and going back to GDDR5 which probably isn’t practical to do.

        * Does this sound like AMD with Vega waiting on HBM2? Sure, except DDR6 looks like its cheaper & easier for memory manufacturers to produce for the consumer market.

        • Srsly_Bro
        • 1 year ago

        They do if the sales of current gen starts to decrease and subsequently revenue.

          • derFunkenstein
          • 1 year ago

          just like Intel, Nvidia’s biggest competition is its last generation of products.

        • freebird
        • 1 year ago

        Yeah, and that is what Intel thought last year…

      • Waco
      • 1 year ago

      I laughed when reading it, so I think we should reference everything in D&D-esque terms. Made my Monday morning just that little bit brighter. 🙂

Pin It on Pinterest

Share This