Nvidia’s GTX970 has a rather serious memory allocation bug

970

[Update]

Nvidia has responded to the memory issue. 

NVidia’s GTX 970 is the current price-to-performance darling, offering incredible visual for incredible value. It seems, however, that it’s harbouring a dark secret. It’s a 4GB card, but it looks like a significant chunk of that VRAM doesn’t work.

According to a number of rather angry people on Reddit, Overlock and the Guru3D forums who’ve noticed their shiny, powerful GTX 970’s come to a screeching halt when maxing out their cards. Clever people, using VRAM benchmarking software, have discovered that when the last 500`700MB of VRAM gets accessed, memory performance drops significantly. Some users have even found that their cards go belly up when hitting 3GB.

Nvidiamemory

The issue doesn’t seem to affect the more expensive, more powerful 980 – so it’s not a Maxwell fault, but rather one specific to the 970. DRAM bandwidth drops from a peak of over 150GB/s, right down to 19.88 GB/s.

“Going beyond 3.5GB vram usage in games like Hitman Absolution, Call of Duty Advanced Warfare severely degrades the performance, as if the last 512mb is actually being swapped from the RAM,” says Reddit user nanogenesis in a thread on the subject.

The problems seems to be widespread – and Nvidia’s admitted that problem affects every single 970 (to varying degrees). They’re looking in to the issue – but unless it’s something that can be resolved via driver or firmware update, a recall may be on the cards.

“We are still looking into this and will have an update as soon as possible,” said NVidia community liaison ManuelG on the GeForce forums.

We asked our resident cabbage, Alessandro – who owns a GTX 970 – to test, and he was hit with the same results. You can download the benchmark to try it yourself, here.  If you require the extra library to run the benchmark, you can get it here.

GTX-970-Memory-Flaw

While the card is still a performance champ, it makes it hard to recommend right now, as it’ll never reach its full potential. If you have money burning a hole in your pocket and you’re aching to buy a new card, you may just prefer to opt for the 980, which isn’t affected by this. Or you could, loathe as I am to say it, wait to see what AMD has up its sleeve.

Check out our ASUS GTX 960 Review and ASUS GTX 970 Review.

 



Author: Geoffrey Tim

I'm old, grumpy and more than just a little cynical. One day, I found myself in possession of a NES, and a copy of Super Mario Bros 3. It was that game that made me realise that games were more than just toys to idly while away time - they were capable of being masterpieces. I'm here now, looking for more of those masterpieces. I am also the emperor of the backend

Share This Post On
  • Lord Chaos

    AMD FTW

    • Hammersteyn

      RED TEAM!

      • Blood Emperor Trevor

        RED SQUADRON. CHECKMATE, GREEN MOB.

        • Admiral Chief

          XD

        • Quo Vadis?

          That is a really cool looking spaceship!

          • Admiral Chief

            You DO know the significance of THAT spaceship right?

          • Quo Vadis?

            Zuma’s new prototype?

          • Admiral Chief

            Pffff, puh-lease, that muppet won’t know awesome even if it slapped him in the face

          • Quo Vadis?

            That’s what your girlfriend said!

          • Admiral Chief

            Luckily I’m married, so your argument is invalid

          • Quo Vadis?

            Married vs happily married? How thin is the ice I am treading on right now?

          • Admiral Chief

            …already sinking

          • Quo Vadis?

            Titanic sinking, or feather sinking?

          • ToshZA

            There’s a “yo mamma” joke in there somewhere.

          • Quo Vadis?

            The pirate is boiling already. Don’t poke the bear anymore LOL

          • ToshZA
          • Quo Vadis?

            aaaw, the Admiral is really a teddy bear! I KNEW IT!

          • ToshZA

            And I found this for @Admiral_Chief:disqus XD

            http://piratesbc.com/images/News/Frenchie.jpg

          • Admiral Chief

            Nope, not regal enough

          • ToshZA

            Bear > Squid man who’s only famous for shouting “It’s a trap” any day.

          • Admiral Chief

            I was referring to the image, MOAR REGAL IMAGE REQUIRED I SAY!

          • ToshZA

            Fine, fair point. *goes off to search*

          • ToshZA
          • Admiral Chief

            [laughs at successful distraction of ToshZA]

          • ToshZA

            Bastard XD

          • Admiral Chief

            So it is true, Jedi Mind trick does work on the weak willed 😛

          • ToshZA
          • Still not as cool as Gold Squadron. Then again they got retconned. Thanks Disney!

          • ToshZA

            You love trolling the Star Trek fan boys don’t you?

          • Blood Emperor Trevor

            lol – it’s just so easy with those fracking toaster-lovers.

          • HairyEwok

            Nothing wrong with some bacon toast ok. Actually i want some bacon toast, I’m going to call Wimpy.

          • Admiral Chief

            I’ve got bacon on my left over pizza, that is currently in my mouth…mmmmm

          • Quo Vadis?

            You utter bastard… Bacon could bring world peace if it was cheaper, I believe. That, and cute puppies and kitten pictures 😀

          • Admiral Chief

            I’ve not eaten puppies or kittens…oh wait you meant picture….no I have not eaten those pictures either

          • Quo Vadis?

            you blatant blue bull!

          • Admiral Chief

            Keep digging that hole

          • Quo Vadis?

            I am digging to China at this rate I’m going, huh Bakkies? 😀

          • Quo Vadis?

            Ag no 🙁
            If its star trek, then that spaceship isn’t so cool anymore

          • Ryanza
          • Quo Vadis?

            Too many space thingies now. Starting to hurt my eyes :'(

          • HairyEwok

            How do you like my spaceship. Bitching right? right?

          • Quo Vadis?

            Dude, no! are you a trekkie?

          • ToshZA

            Nah, he just likes playing space tennis.

          • Quo Vadis?

            Space tennis, or “spaced” tennis?

          • HairyEwok

            What’s a trekkie? Is it like a tekkie mixed with a tractor?

          • Ryanza
          • Ryanza
          • Blood Emperor Trevor

            That’s a nice exhaust port you have there, hate to see it… exposed.

          • HairyEwok

            Psh it’s not like anything can fit into that port so there’s no need to protect in in any way at all.

          • ToshZA

            That’s what she said?

          • Buddydudeguy

            Some one is young as fuk and/or never seen Star Wars…

          • Quo Vadis?

            Never seen any star wars or star trek. Not planning on watching them either 🙂

          • lokie67

            thats too bad, oh well its your loss….kids these days…

          • Quo Vadis?

            Never been interested in sci-fi movies tbh.

          • Buddydudeguy

            Obviously.Thanks so much for clearing that up. I wonder if you’re a too cool for school douche nozzle or just really young. I have to lean towards dumb ass going by your name, do you even have a TV?

          • Quo Vadis?

            Wow. So simply because I refuse to watch movies from a genre that I don’t particularly like, I am perceived as a “douche nozzle”. Here’s the thing. I do own TV’s. Three to be exact. “Quo Vadis?” is a colloquial term used in the philosophical sense, meaning “whereto now?” or “where are you going?”. In your case, I am guessing back of the unemployment line for foodstamps or in your parent’s basement. In relation to my age, I am well into my mid 30’s. I am educated, and most likely your next response would be “technical/ console moron”, as that seems to be the only chirp you like. Judging by your vicious demeanour to how another individual reacts to your entertainment tastes, I would have to place your intellectual/ emotional age at about 12 😉
            Now, keyboard Green Beret, be a useful participant to society and go find a job and contribute something. Best of luck! Get some clean clothes, wash your hair, and I think McDonalds might have an opening for you.

    • fag

      I don’t get this fanboy crap for AMD or Nvidia, I have a 970 only because it was the perfect card for my budget. Both companies hold sweet spots in various budgets, I would have gotten an AMD card if my budget was smaller.

      • Behrouz Sedigh

        Your Card is fine if doesn’t use more than 3200Mb.

  • Blood Emperor Trevor

    Wow, that’s a pretty big screw up. Hope it’s just a software/firmware-fixable thing, and not the actual hardware.

    • Lord Chaos

      Maybe they should have thought about a day 1 patch.

      • Ghost In The Rift

        If hardware starts with day 1 patches ill pack my PC in its wooden box and let it rest in the darkest corner in my house.

        • Lord Chaos

          Or just join team Red. 😀

          • Ghost In The Rift

            Or i can do that lol 😀

            Think i might just wait and see what team red is up to XD

          • Lord Chaos

            At least the new CEO seems to be getting things back in order.

        • ToshZA

          *ahem*

          Xbox One
          PlayStation 4

          • Lord Chaos

            *ahem* we were talking about actual hardware, not door stoppers. 😛

          • ToshZA

            LOL!!!!

          • Ryanza

            In America people uses the Xbox One to keep their houses from flying off when a tornado appears.

          • Admiral Chief

            I knew it! There was a reason behind the lack of homes being destroyed recently

          • Ghost In The Rift

            Its Friday, which means tomorrow i’m having a mean braai and you had to go and just break me;(

        • Uberutang

          Technically you do have day 1.. you need a driver to start.

          • Ghost In The Rift

            Which you normally get with a disk and aint 2 and up GB, i’m cool with the disk:-P

          • Uberutang

            yeah usually small 🙂 I never use the disk, since they are usually out of date. BETA ftw.

          • Ghost In The Rift

            lol true, 1 first use the disk then go Nvidias site and download the latest, always the long road lol

    • ToshZA

      Yeah if it’s hardware then that’s a major f%$# up for Nvidia and their customers. It might convince many to go AMD, depending on how Nvidia handle it. Costly, either way.

      At least a driver is a simple update – and it’s good practice to keep those up to date anyway.

  • Admiral Chief

    OF COURSE IT DOES!!!

    I wonder if OOBI is involved with the design process?…

    • Blood Emperor Trevor

      No, it’s a memory problem – which means Bethesda. 😀

      • Admiral Chief

        O NOES A DOOR!!!!

  • HAHAHAHAHAHAHAHAHAHAHAHAHA.

    On a serious note.

    HAHAHAHAHAHAHAHAHAHAHAHA!!!

    😛

  • ToshZA

    This just after the 960’s are released. Well, LOL.

    Nvidia has been getting so much right, it was inevitable that something would happen. Pity it happened to the card that IS Nvidia at the moment XD

  • Ryanza

    I’ve listened to the way the Nvidia people talk in interviews and such. I don’t trust them.

    Don’t Support Nvidia, unless you on a power diet and now VRAM diet.

  • Tarisma

    Ehhh I just bought one of these

    • ToshZA

      Eish

  • Uberutang

    Cool Recall means new card. Wont mind 🙂

  • geel slang

    That was a close call, almost bought one.

    • ToshZA

      Yup, this could be disastrous for Nvidia.

  • FSR

    *quietly strokes 980, sets all settings to max, pours mountain dew into doritos bag* not today my precious, not today.

  • Uberutang

    What app can we use to test this at home?

  • BDK
  • JL

    For pete sake, you need to run that Nai Benchmark with your IGPU being primary display and your GTX 970/980 in headless display mode. You guys do know what is headless display mode, right?
    Windows desktop compositing and web browsers DO reserve some portion of VRAM. Please don’t skew the result with improperly ran benchmark

    • Basil Nolan

      I wholeheartedly agree. I have 770 with 2GBs and decided to give a try to Nai’s Benchmark. Just as you said, if you don’t do the benchmark properly, the tool reports serious latency spikes in the last chunk. Ran the test again with everything turned off or lowered, and the last chunk had in fact the lowest latency.

      Please people, ensure this requirements are met, or you me be in for a bad impression.

      Thanks for the tip.

    • Windows doesn’t use 500MB of VRAM, though.

    • Anti-Pacheco

      I don’t have the slightest idea what you mean through “headless” mode. My GTX 980 benchmark show the same issues of 970 GTX screenshot in the article.

  • Echo

    Does the 960 have the same problem?

    • JL

      Nope, GTX 960 doesn’t have such issue.

  • Sidneys1

    Interesting, I get the same performance drop on my GTX 760.

  • Ele & Ela

    Anyone tested with different drivers version?

  • techytecherson

    Just tried this on my 980 and got the same:

    L2-Cache-Bandwidth of Chunk no. 26 (3328 MiByte to 3456 MiByte):504.29 GByte/s

    L2-Cache-Bandwidth of Chunk no. 27 (3456 MiByte to 3584 MiByte): 7.57 GByte/s

    L2-Cache-Bandwidth of Chunk no. 28 (3584 MiByte to 3712 MiByte): 5.94 GByte/s

    L2-Cache-Bandwidth of Chunk no. 29 (3712 MiByte to 3840 MiByte): 5.94 GByte/s

  • Jacob Evans

    This is really bad news. Last major fuck up I heard from Nvidia was when their Drivers killed the fans on GPU’s that was a good few years ago now.

    But fuck me this is pretty bad.

  • Uberutang

    MSI GAMING 4G

    :NVIDIA>Rec.exe

    ai’s Benchmark

    llocating Memory . . .

    hunk Size = 128 MiByte

    ress any key to continue . . .

    llocated 30 Chunks

    llocated 3840 MiByte

    enchmarking DRAM

    ress any key to continue . . .

    RAM-Bandwidth of Chunk no. 0 (0 MiByte to 128 MiByte): 155 GByte/s

    RAM-Bandwidth of Chunk no. 1 (128 MiByte to 256 MiByte): 146 GByte/s

    RAM-Bandwidth of Chunk no. 2 (256 MiByte to 384 MiByte): 153 GByte/s

    RAM-Bandwidth of Chunk no. 3 (384 MiByte to 512 MiByte): 156 GByte/s

    RAM-Bandwidth of Chunk no. 4 (512 MiByte to 640 MiByte): 153 GByte/s

    RAM-Bandwidth of Chunk no. 5 (640 MiByte to 768 MiByte): 155 GByte/s

    RAM-Bandwidth of Chunk no. 6 (768 MiByte to 896 MiByte): 153 GByte/s

    RAM-Bandwidth of Chunk no. 7 (896 MiByte to 1024 MiByte): 156 GByte/s

    RAM-Bandwidth of Chunk no. 8 (1024 MiByte to 1152 MiByte): 153 GByte/s

    RAM-Bandwidth of Chunk no. 9 (1152 MiByte to 1280 MiByte): 156 GByte/s

    RAM-Bandwidth of Chunk no. 10 (1280 MiByte to 1408 MiByte): 153 GByte/s

    RAM-Bandwidth of Chunk no. 11 (1408 MiByte to 1536 MiByte): 152 GByte/s

    RAM-Bandwidth of Chunk no. 12 (1536 MiByte to 1664 MiByte): 156 GByte/s

    RAM-Bandwidth of Chunk no. 13 (1664 MiByte to 1792 MiByte): 151 GByte/s

    RAM-Bandwidth of Chunk no. 14 (1792 MiByte to 1920 MiByte): 153 GByte/s

    RAM-Bandwidth of Chunk no. 15 (1920 MiByte to 2048 MiByte): 152 GByte/s

    RAM-Bandwidth of Chunk no. 16 (2048 MiByte to 2176 MiByte): 156 GByte/s

    RAM-Bandwidth of Chunk no. 17 (2176 MiByte to 2304 MiByte): 152 GByte/s

    RAM-Bandwidth of Chunk no. 18 (2304 MiByte to 2432 MiByte): 156 GByte/s

    RAM-Bandwidth of Chunk no. 19 (2432 MiByte to 2560 MiByte): 143 GByte/s

    RAM-Bandwidth of Chunk no. 20 (2560 MiByte to 2688 MiByte): 156 GByte/s

    RAM-Bandwidth of Chunk no. 21 (2688 MiByte to 2816 MiByte): 155 GByte/s

    RAM-Bandwidth of Chunk no. 22 (2816 MiByte to 2944 MiByte): 155 GByte/s

    RAM-Bandwidth of Chunk no. 23 (2944 MiByte to 3072 MiByte): 46 GByte/s

    RAM-Bandwidth of Chunk no. 24 (3072 MiByte to 3200 MiByte): 22 GByte/s

    RAM-Bandwidth of Chunk no. 25 (3200 MiByte to 3328 MiByte): 22 GByte/s

    RAM-Bandwidth of Chunk no. 26 (3328 MiByte to 3456 MiByte): 22 GByte/s

    RAM-Bandwidth of Chunk no. 27 (3456 MiByte to 3584 MiByte): 13 GByte/s

    RAM-Bandwidth of Chunk no. 28 (3584 MiByte to 3712 MiByte): 7 GByte/s

    RAM-Bandwidth of Chunk no. 29 (3712 MiByte to 3840 MiByte): 6 GByte/s

    ress any key to continue . . .

    enchmarking L2-Cache

    ress any key to continue . . .

    2-Cache-Bandwidth of Chunk no. 0 (0 MiByte to 128 MiByte): 418 GByte/s

    2-Cache-Bandwidth of Chunk no. 1 (128 MiByte to 256 MiByte): 422 GByte

    2-Cache-Bandwidth of Chunk no. 2 (256 MiByte to 384 MiByte): 422 GByte

    2-Cache-Bandwidth of Chunk no. 3 (384 MiByte to 512 MiByte): 418 GByte

    2-Cache-Bandwidth of Chunk no. 4 (512 MiByte to 640 MiByte): 422 GByte

    2-Cache-Bandwidth of Chunk no. 5 (640 MiByte to 768 MiByte): 417 GByte

    2-Cache-Bandwidth of Chunk no. 6 (768 MiByte to 896 MiByte): 422 GByte

    2-Cache-Bandwidth of Chunk no. 7 (896 MiByte to 1024 MiByte): 417 GByt

    2-Cache-Bandwidth of Chunk no. 8 (1024 MiByte to 1152 MiByte): 418 GBy

    2-Cache-Bandwidth of Chunk no. 9 (1152 MiByte to 1280 MiByte): 406 GBy

    2-Cache-Bandwidth of Chunk no. 10 (1280 MiByte to 1408 MiByte): 417 GB

    2-Cache-Bandwidth of Chunk no. 11 (1408 MiByte to 1536 MiByte): 422 GB

    2-Cache-Bandwidth of Chunk no. 12 (1536 MiByte to 1664 MiByte): 417 GB

    2-Cache-Bandwidth of Chunk no. 13 (1664 MiByte to 1792 MiByte): 402 GB

    2-Cache-Bandwidth of Chunk no. 14 (1792 MiByte to 1920 MiByte): 422 GB

    2-Cache-Bandwidth of Chunk no. 15 (1920 MiByte to 2048 MiByte): 418 GB

    2-Cache-Bandwidth of Chunk no. 16 (2048 MiByte to 2176 MiByte): 422 GB

    2-Cache-Bandwidth of Chunk no. 17 (2176 MiByte to 2304 MiByte): 417 GB

    2-Cache-Bandwidth of Chunk no. 18 (2304 MiByte to 2432 MiByte): 411 GB

    2-Cache-Bandwidth of Chunk no. 19 (2432 MiByte to 2560 MiByte): 422 GB

    2-Cache-Bandwidth of Chunk no. 20 (2560 MiByte to 2688 MiByte): 417 GB

    2-Cache-Bandwidth of Chunk no. 21 (2688 MiByte to 2816 MiByte): 411 GB

    2-Cache-Bandwidth of Chunk no. 22 (2816 MiByte to 2944 MiByte): 422 GB

    2-Cache-Bandwidth of Chunk no. 23 (2944 MiByte to 3072 MiByte): 136 GB

    2-Cache-Bandwidth of Chunk no. 24 (3072 MiByte to 3200 MiByte): 67 GB

    2-Cache-Bandwidth of Chunk no. 25 (3200 MiByte to 3328 MiByte): 66 GB

    2-Cache-Bandwidth of Chunk no. 26 (3328 MiByte to 3456 MiByte): 66 GB

    2-Cache-Bandwidth of Chunk no. 27 (3456 MiByte to 3584 MiByte): 15 GB

    2-Cache-Bandwidth of Chunk no. 28 (3584 MiByte to 3712 MiByte): 6 GB

    2-Cache-Bandwidth of Chunk no. 29 (3712 MiByte to 3840 MiByte): 6 GB

    ress any key to continue . . .

  • Luieburger

    I just bought one of these. Granted, I got it for $270. That’s cheap as hell.

    Still… come on nVidia. You’re better than that.

    I guess it’s time for me to get one of those AMD cards and go FreeSync.

  • Cakefish

    980M 4GB results (stock clocks, 344.75 driver):

    http://i60.tinypic.com/2rppdgp.jpg

  • Nigel Uno

    Fermi woodscrews, housefires, etc

  • Louisi9

    I was gonna get a 970 but instead chose to get a 780ti that was on sale. Glad i did now.

  • John

    Same problem with my GTX 780 Ti. Happens with the last 500MB.

  • John

    It actually happens after 2560MB, more than 300MB useless.

  • evolucion8

    THis is what happens when nVidia keep stripping their GPU from more logic to make them smaller and leaner at the expense of replacing that functionality on software on the driver stack. When the moved from Fermi to Kepler, they removed the hardware scheduler, hot clocks among other computing stuff, made it a leaner GPU, but also more reliant on driver tweaks as now the driver and the compiler are the ones that needs to reorganize the threads on a way to feed the Kepler cores by using static software scheduling which is good for games, but for games that uses DirectCompute and GPGPU, is not good cause algorithms aren’t always linear, now on Maxwell, they further stripped down the GPUs by removing additional logic to make the cores smaller, gimped the DP units (That is why the have the horrible 1/32th rate), and now seems that the driver is also the one that manages the VRAM, All nvidia is doing is stripping their GPUs and replace those on software to inflate their profits. That is why the nVidia driver is getting bigger, bloated and why suddenly the GTX 780 is losing against the R9 290 when it was launched, it was faster? Cause the GTX 780 and any Kepler/Maxwell architecture needs optimizations on a per app basis to run properly, and now that nVidia launched Maxwell, Kepler is no longer priority for nVidia. On the other hand, AMD’s Vector design focus on TLP instead of ILP unlike nVidia, and does not need lots of driver tricks to perform, as it is a self balancing design with a powerful hardware dynamic scheduler (That explains why it performs good on the faith that AMD drivers aren’t the best).

    • Anti-Pacheco

      Are you in drugs? Just saying.

      • evolucion8

        Don’t you have something smart to say or are u retarded?

      • evolucion8

        Can you post something smart? Just saying, don’t get intimidated by my knowledge, just learn from the masters. 😀

  • Neilas

    Those last chunks are reserved for the OS, you know Aero? I bet they didn’t even run it in headless mode with the IGPU running the display. Does anyone even know who or where this “benchmark” program came from? Also, moderate your damn comments section, its a giant spam thread until about half way down.

  • AMN3S1AC

    Thank you for posting this news…too many tech sites seem reluctant to acknowledge this issue.

  • It stinks.

    What a bunch of worthless FUD and speculation. Where’s the evidence that Nvidia has acknowledged the issue affects all 970 cards? “We’ll look into it” isn’t even an acknowledgement that it is an issue. No one from Nvidia has said anything about a recall either. Quit posting this horse shit you butthurt AMD fanboy.

    • OBXS

      here here

    • OBXS

      but it is a problem.

    • Anti-Pacheco

      I redid some tests after my 980 GTX show the same results of the “defective” 970 GTX VRAM, DSR from 4K and smothing all maxed in several games very much plying my VRAM up to 4 Gb and there’s no stutter at all in comparison to 1080 px gameplay (of course, lower FPS). I’m more into this random guy who coded this “util” is that same that infect a lot of computers with bloatware once or a big troll.

  • Based

    WOW WHAT THE FUCK.

    I just did the benchmark on my card, second I hit 3gb of Vram usage, it fucking drops.

    explains why Crysis3 and Elite have so much shuttering.

    FOR FUCKS SAKE.

    I AM PISSED>

  • OBXS

    Why the hell does the writer even care about buying AMD products? Why do you ‘LOATHE’ AMD? As long as it works good, and is cheap, buy it. If the 380X is better and more affordable than a 980, why does it matter? Why would you ‘loathe’ saving money and getting a better experience. The same goes for the other side, but often isn’t the case. It really shouldn’t matter, but fanboying over one brand, just because you pay twice the amount for maybe 1/10 better performance seems ridiculous.

  • az

    4gb version of nvidias 680gtx is BROKEN TOO!!!!
    Now I know why everything went to hell when the mem hit over 3.5gb!
    I WANT COMPENSATION

  • Battle_Tuba

    Is the benchmark “safe” It was just flagged on my AV. Im sure its not but I would like to be sure.

  • Bryan Del Rizzo

    It’s not a bug. The GeForce GTX 970 is equipped with 4GB of dedicated graphics memory. However the 970 has a different configuration of SMs than the 980, and fewer crossbar resources to the memory system. To optimally manage memory traffic in this configuration, we segment graphics memory into a 3.5GB section and a 0.5GB section. The GPU has higher priority access to the 3.5GB section. When a game needs less than 3.5GB of video memory per draw command then it will only access the first partition, and 3rd party applications that measure memory usage will report 3.5GB of memory in use on GTX 970, but may report more for GTX 980 if there is more memory used by other commands. When a game requires more than 3.5GB of memory then we
    use both segments. The best way to test that is to look at game performance. Compare a GTX 980 to a
    970 on a game that uses less than 3.5GB. Then turn up the settings so the game needs more than 3.5GB and compare 980 and 970 performance again. For example:

    3.5GB setting = 3456×1944

    GTX 980: 55fps (-24%)
    GTX 970: 45fps (-25%)

    In other words, on GTX 980, Shadows of Mordor drops about 24% on GTX 980 and 25% on GTX 970, a 1% difference.

  • Anti-Pacheco

    It’s not 970 only. Just run 980 GTX here and got the same issues for the last VRAM chunks.

  • Brian Blair

    I have used well over 3.5GB on my EVGA 970 SC ACX 2.0 and my performance has not dropped yet! Still get a perfect V-Synced 60FPS in every game I play!

Like this post? Share it with your friends! Hate it? Share it with your enemies!

We rely on good people like you to share our content. Spread the news to your friends, family or social media trolls!