NVIDIA header image 2

NVIDIA GeForce RTX 3070 GPU only uses 14Gbps GDDR6 Memory, 16Gbps reserved for future SKUs

Nvidia recently announced its Ampere lineup of Flagship GPUs, the RTX 3090, 3080 and 3070. And while there were some specs highlighted, there was chatter before on the web that Nvidia might use 16Gbps GDDR6 memory for the RTX 3070 GPU. But that doesn’t appear to be the case.

We already know the GeForce RTX 3070 only uses the GDDR6 memory type, the non-X variant from Micron, and now we have some confirmation on the memory clock speed as well, thanks to ASUS and Zotac.

According to updated specification pages for the ZOTAC GAMING GeForce RTX 3070 Twin Edge and ASUS Dual GeForce RTX 3070, the GeForce RTX 3070 uses the older, and slower 14Gbps GDDR6  memory.

More importantly, the use of 14Gbps GDDR6 across a 256-bit wide memory bus gives the RTX 3070 448 GB/s of memory bandwidth, which is the same as the Turing RTX 2070 SKU. This mirrors the memory speeds of its preceding models (the GeForce RTX 2070 and GeForce RTX 2070 SUPER), but it is still less than the GeForce RTX 2080 SUPER’s 15.5Gbps GDDR6 memory.  My guess is that Nvidia is reserving the sweeter/faster 16Gbps GDDR6 modules for the rumored upcoming GeForce RTX 3070 Ti/SUPER SKUs.

There is a high chance we might see a new Ti or a SUPER variant SKU next year, depending on how AMD’s RDNA2 cards turn out. The new Zotac and ASUS product page entries give more credence to this rumor.

Nvidia might be saving the higher 16Gbps GDDR6 module for faster models, like the Ti or Super SKUs. Nvidia currently uses an even faster 19.5Gbps and 19Gbps GDDR6X memory for the RTX 3090 and RTX 3080 GPUs, both from Micron Technology.

The GeForce RTX 3070 launches this October, with prices starting at USD $499 for the Founders Edition.

NVIDIA GeForce RTX 3070 GPU memory 14Gbps-1NVIDIA GeForce RTX 3070 GPU memory 14Gbps-2

44 thoughts on “NVIDIA GeForce RTX 3070 GPU only uses 14Gbps GDDR6 Memory, 16Gbps reserved for future SKUs”

  1. Makes sense. Nice article !! I knew these Super cards are going to offer even more better performance jump in this Ampere lineup, so there is no point in buying any of these now.

    “cause next year the Ti/super cards will offer more performance for lesser Money.

    1. By that logic you might as well hold off forever becouse every year there are new upgrades to be had for a lesser price.

      1. Oh, obviously what you wrote is correct, but I didn’t mean to say keep holding off the upgrade till next the few years. I was only referring particular to the Ampere lineup.

        Of course, the next lineup after Ampere, is going to even bigger and faster, but that’s the future. I mean these Ti/Super cards will hit early next year, so we don’t have to wait much too long.

        But the next-gen lineup after Ampere is not going to come out at least next year, maybe after full 2 years. So that’s a far cry.

        1. Afcourse if you feel waiting a few more months is worth it then go right ahead, i just feel the gains will be minimal at best.

          For me i know the 3080 will be more then enough, but then again maybe im not the best person to listen to becouse i originally bought a regular 2070…though i havent regretted it at all.

          1. Well I’m not fully sure, as it was a guess from my side. but I expect them to come between the first and second quarter next year, Q1 and Q2.

            After how many months did the previous gen Turing GPUs got this Super variant ? This will give us some idea about the time frame.

          2. I’d have thought Q3 or more likely Q4 next year to be more realistic. Time will obviously tell.

          3. After how many months did the previous gen Turing GPUs got this Super variant ?

            This will give us some idea about the time frame of these cards.

          4. It doesn’t matter really, the point still stands if you are an early adopter of anything nvidia you are the earliest fool. 3 to 6 months time they will refresh everything 50x over and then there will be a new super duper variant more powerful for less. And the cycle goes on and on and on, the consumer never learns. I use to be that consumer who would buy nvidia’s BESTEST CARD and in 3 months, TI more powerful 35% cheaper. i’d say nvidia is very disrespectful but after decades of this I think it’s the consumers that are DUMBA$$E$$.

            When it comes to Nvidia and gaming, i wouldn’t early adopt a baby by them. I’d wait 6 months to a year and see if the babies 3rd leg finally grows out.

          5. Yeah and every release cycle these clowns who call themselves gamers acts as if they don’t know how this works and how NVIDIA operates.

          6. After how many months did the previous gen Turing GPUs got this Super
            variant ? This will give us some idea about the time frame.

  2. Most people are excited about the RTX 3070, but I honestly don’t think it’s worth saving $200 vs the performance jump the RTX 3080 offers.

    40% price increase for 48% more CUDA cores, 70% more memory bandwidth, 25% more memory capacity, and probably 50%-60% more performance at a minimum?

    Give me the 3080 every damn time.

    1. What if we don’t need that much extra performance ? Why spend even 200 USD more, if a lesser priced card offers a much better value for money for someone ?

      This means you are claiming that the 3070 is not a faster/better GPU, only the 3090 and 3080 are heaps faster and better.

      1. Obviously the 3070 is a better offer than what high end Turing offered. But the extra performance of the 3080 also means you can go longer before having to upgrade. It’s a better value proposition in my opinion.

        I’d also argue that most people who can afford a $500 GPU can afford a $700, even if it means saving up for a few more weeks/months.

      2. the point is that the x80 card is actually offering more performance per dollar than the x70 which has NEVER been the case before. really the 3070 has no business being more than $450 when you consider what the 3080 is giving you for $700. and i hate to break it to you but 8gb of vram is going to be an issue in some upcoming games. hell its already an issue in Wolfenstein Youngblood now if you try using max settings and raytracing even at just 1440p. you will have to turn down texture streaming one notch and put DLSS on balanced or the game will hitch and lock up. with more than 8gb there is no issue.

  3. I mean on every Turing card you can easily set +1000 Memory clock on MSI Afterburner and get those 16Gbps on the card, but yeah little to none more OC headroom after that. If this is true then OCing the memory would be a must, but seems it comes short on VRAM quantity aswell… for being faster than a 2080 Ti feels a little short on memory department.

    Pretty sure a Ti/Super version is on the way

      1. I’m pretty sure 3060 will have the same memory setup as 3070.

        Probably not going to be all that much slower too but might be like 400$.

        If it ends up being worse then I think I can’t imagine 5700 successor having less then 8gb.

  4. I will wait for the AMD cards. Than nexus gaming benchmarks and maybe i will buy something for chrissmas for my kid…. Maybe.

  5. To all those willing to upgrade to the 30 series…hold your horses till the super and ti variants are released. Your patience will be fruitful.

    1. Well, you can always wait for something better. I’m going to get a RTX 3080 when Zen3 becomes available. Should be a good upgrade coming from a 6700K with GTX 1080.

    2. the waiting game can be an endless one. the tip for not having buyer remorse is once you’re decided with your purchase then just be satisfied with it. you still going to get that performance increase. new parts coming out cheaper and faster? it happen all the time. for example i got my GTX460 exactly a week before nvidia launch GTX560Ti to the market. but since it is a significant upgrade over my older GTS250 i’m still happy with it. and then i got my 2500K roughly a week or two before ivy bridge comes out. it ended become one of my best purchase ever. use that CPU for almost 7 years due to it’s overcloking capabilities.

  6. Something is not right here. How it is possible to change GPU from 6.5 TFLOPS to 20.3 TFLOPS without providing more data? Is this real 20.3 TFLOPS or Nvidia just changed way they count cores/tflops?

    GTX 2070 (6.5 TFLOPS – 7.5 TFLOPS)
    – 2560 processing units at variable speed 1410-1620 Mhz
    448 GB/s of data bandwidth

    GTX 2070 Super (8.2 TFLOPS – 9 TFLOPS)
    – 2560 processing units at variable speed 1605-1770 Mhz
    448 GB/s of data bandwidth

    GTX 3070 (20.3 TFLOPS)
    – 5888 processing units
    448 GB/s of data bandwidth

    2560 cores in 2070 series = 13,600 million transistors
    5888 cores in 3070 series = 17,400 million transistors
    It is 2.3x more cores but only 27% more transistors?

    Those numbers are strange – why new cores use 40% less transistors?

  7. I’ll still buy the 3080 10GB Version. We are talking here about ULTRA 4K Texture Settings. You’ll be fine till RTX 5080. If not, the difference between ULTRA Setting and VERY HIGH setting isn’t that much. Ultra somestimes even causes stuttering. Or its just uncompressed will take vastly more VRAM for minimal quality gain.

    BUT for the RTX 3070. Yes, wait for the higher VRAM version. Otherwise you may be “forced” tos et to very high earlier than the RTX 3080.

    But probaly your GPU will lack the rawpower to that time. Its very well the case now with my 1080.

    Also, is the higher price worth the additional Ultra Settings for 1-2 Years for like a handful of games?

Leave a Reply

Your email address will not be published. Required fields are marked *