If you own a 970GTX you may wa…
Daz 3D Forums > General > The Commons>If you own a 970GTX you may wa…
If you own a 970GTX you may want to watch this..
Ghosty12
Posts: 2,077
As the title suggests if you own a 970GTX you may want to watch 
seems that Nvidia are in hot water over the 970GTX seems they were advertising them as a 4GB card when they were found to be a 3.5GB card..
Post edited by Ghosty12 on
Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2025 Daz Productions Inc. All Rights Reserved.
Comments
yeah, well they've been exaggerating a lot of specs and performance it seems.
Which has one wondering about their new cards..
...I remember in the year before the release of the Maxwell 970/980 the talk was of 8 GB of memory. Seemed totally plausible at the time as the Titan X doubled the memory of the original Titan from 6 GB to 12 GB (I try to forget about the Titan-Z which was advertised as 12 GB but was actually two 6 GB Titans permanently linked together and cost more than just getting two individual cards).
yeah. tbh, I was already aware of that default when I bought it.
Yeah, I knew that going in. ;)
Still pretty good for the price!
Interesting I wonder if that means our gtx 980ti cards are not what they say they are?
From what I have read elsewhere the fact that only 3.5GB is fast memory is of greater concern for gamers than renderers, so since I do not use the card for 3D games, I have not suffered this particular issue. Naturally whilst a big cheque from nVidia coming my way would be nice, it is not something I feel I deserve.
Pfft, reminds me why I first started seriously thinking about my Crimson switch to Green...
I should have left a while ago when I realized that my 6950 was a cheaper version of the 6970 (Basically the same card when flashed) and more recently with my Sapphire r9 290x vs the stock version (The stock vers better performance vs the Sapphire) not to mention the r9 390x is simply a 290x oc'd with barely a difference in performance as well!
Saw the Fury-X comparison (The highest performance GPU AMD has right now) vs the 1080 and was completely obliterated, (Even the 1060 smacked it up a bit) with nearly the same price even though the 1080's the considerably newer card!...
The choice was obvious, and the best time to make the switch!...In SPITE of crossfiring two weak cards (RX 480) to compare/compete with the 1080! Which is a crappy investment because two 4gb GPUs do not equal 8gb, never mind that one might fail...
The best time to crossfire is late in the GPU's life by buying two, as they're much reduced in price by then!...Good news, I got a Gigabyte version instead of MSI/Asus as they sold out due to popularity! Dodged a bullet there!
You could easily check that yourself before you throw that accusation to the wind.
This situation is one that normally gets covered in "Engineering Econ" in University. Many times with electronic circuits, there are needs for specific "bookkeeping" areas where (for lack of a more suitable general term) variables are stored. With MMUs (memory Management Units) there is a finite amount of memory that can be managed for the cost of the unit. As the area size increases, the costs and additional support circuitry also increase. So in the case of the 970, nVidia had selected a 4GB MMU (at the time the next step with a compatible unit was 8GB) to handle the memory of the card, but along the way the firmware's memory use footprint exceeded the memory that was economically feasible as a separate unit. So the choice becomes whether to add in a new smaller MMU and memory subsystem for the firmware (for those that haven't made the connection yet: the CUDA scheduler and state machine) with the associated costs and board space, or to allocate a chunk of the current MMU's map for the purpose. Going the second route solves 2 problems: 1) there are no extra costs for the additional MMU and support circuitry, 2) the unused part of the chunk can be reallocated for use as VRAM if necessary.
From an engineering standpoint this makes perfect sense. From a marketing one, it doesn't. So, they've announced that a new 4GB card is going to be available, but now engineering has decided that it needs a chunk of the allocated address space, AND the memory timings necessary are significantly slower than the High-Performance VRAM. BUT a good portion of that slower memory CAN be used for processing -- so Marketing says: "Well, there *IS* 4GB of memory on the card and we never specified that all of it would be high-performance, AND we are using a 4GB MMU. So we can still say that there is 4GB of memory on the card." The Engineers walk away shaking their heads.
I can go into significantly more detail on the electronics part, but I don't think most would be interested.
Kendall
this was discussed in great detail on these forums when the card was introduced by nvidia prior to the lawsuit ruling.
The card does indeed have 4GB RAM, just not for you to use for what you would like.
so $30 coming your way if you bought this 3.5GB lie of lies. Hopefully you still have your receipts. $30 of Daz content or seven and half Big Macs (before tax)!
3.5 available when they advertised it as 4
How is this any Different from a "one terabyte" External USB Drive ,sold on Amazon, that really only has 992 Gigabytes of actual available storage???
I did Mike. I have 2 EVGA GeForce GTX 980 Ti Super Clocked Gaming ACX 2.0 6GB GDDR5 384bit PCI-E Graphic Card
Nivida has stated its a different architecture than the GTX 970'sc & looking at my windows rating experience for my gpu's. For 2 cards it says it had available 12027 MB Vram with the 2 combined cards.
But my husband has a EVGA GeForce GTX 970 04G-P4-2974-KR 4GB SC GAMING w/ACX 2. Graphics Card and his windows Vram rating states he has 4098 mb of ram not 3.5gigs that it should according to the article. So he check his today after i showed him this thread. So if we go by what our pc is rating our Vram its says its what the cards are rated for. . But according to the Video and the article that followed in the link they say the ratings is not accurate. so it is a fair question for me to ask without being a "accusation"
According to what I've read, 3.75GB is actually fully available with the final 256MB pretty much constantly used. It is possible to use the full 4GB, but in order to do so you must effectively turn off CUDA completely, resulting in a FrameBuffer card. The 512MB of memory in dispute is a much, much slower RAM that is timed to be accessed by the firmware. The address space can be allocated to the framebuffer, but doing so causes the whole memory bus to stall while and when the slower memory is accessed.
The class action is likely to have to play lots of semantics games to come out on the winning side in this I'm thinking.
From nVidia:
"“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.
Read more: http://wccftech.com/nvidia-geforce-gtx-970-memory-issue-fully-explained/#ixzz4Gl4SQtPq"
Kendall
http://hwbench.com/vgas/geforce-gtx-1080-vs-radeon-r9-fury-x
Most of the better performance is due to having about twice the memory and twice the speed on that bus of the nVidia card.
I'm lucky too as I've been waiting to buy a new computer until the next generation of video cards after DirectX 12 comes out as these video cards are very expensive. Complicating matters there is Vulkan and Metal to consider too. And it is very fortunate that AMD has open sourced their ProRenderer as soon one will be able to choose the best card performance per dollar spent.
I read earlier this year that the next big AMD video card is going to have a smaller die process than 16nm, I believe it was 10nm or 12nm more likely but I don't remember the amount of RAM it was going to have. That is supposed to come about this winter.
For my computing interests I really only have interest in this technology when it filters down to tablets and phones at a functionally equivalent level meaning that Intel mobile CPU/GPU chipsets and ARM CPU/GPU chipsets although I was under the impression that both nVidia and AMD are trying to get CPU/GPU chipsets designed and manufactured that could compete with Intel and ARM.
I received a check for $10 from Red Bull for false advertising but am not eligible for the nVidia refund which demonstrates the profit margin these companies make on their products. I've also received refunds for other class action suits and 100% refunds from businesses engaged in deceptive businesses practices via class action suits. All unsolicited by me and I was found by the businesses that had to pay the claims. Extremely expensive. People dismiss such law suits as fabricated by greedy lawyers looking for a payday but lawyers have been one of the best safety inventions ever. A novice computer user using DAZ Studio and seeing their average scene size renders were using 4GB RAM would be easily deceived into thinking a 4GB RAM video card from nVidia would be a vidia card with 4GB RAM they could use completely via DAZ Studio or a video game. It was nVidia that came to the consumers and told them that, so others discovering that was a lie and not being bothered by that lie enough to file for a claim from that class action settlement does not make that lie acceptable or legal.
HD and SSD manufacturers list preformatted and formatted space in the consumer ads and it's hard to avoid seeing both those values if you read the ad, that was not the case with nVidia and that video card.
Last March I had a new computer with an EVGA 4GB GTX970 custom built by CyperpowerPC. Now I'm wondering: 1) Am I eligible for the $30 refund and 2) If so, how do I get it?
As a conclusion: what nvidia card do you recommand for DS 4.9 after this? I am ready to spend about 350 to 500 if required. Oh, I don't play games, I only need it for IRAy and OctaneRender.
A 4Gb is pretty much minimum and I really wish I had gotten a card with more memory like 6GB or even better 8GB. So, for now that leaves the GTX980 or Titan series; but you mifgt want to wait until IIrqay is up updated to ru n on the 10xx cards.
You have to answer that yourself. How much video card memory do you need to load your DAZ scenes since I suppose that is what you are doing. It is important that you get the usuable amount of memory they are promising you on the box for you to be able to make the correct buying choice.
So if you search
https://smile.amazon.com/s/ref=nb_sb_noss_2?url=search-alias=aps&field-keywords=nVidia+Video+card
it will let you choose how much video RAM they are advertising and the GPU processor and then you can narrow down to you price range.
Then you can ask her in the forums if the chose card actually has that much video RAM available for DAZ Studio scenes to be rendered.
So taking the high end of your price range gets you 8GB video RAM for the MSI GeForce GTX 1070 Aero for $419 and free shipping:
https://smile.amazon.com/MSI-GeForce-GTX-1070-AERO/dp/B01GUAJMRU/ref=sr_1_4?s=pc&ie=UTF8&qid=1470683617&sr=1-4&keywords=nVidia+Video+card&refinements=p_n_feature_four_browse-bin:6066318011
So now you can ask the vidio card experts here or elsewhere if this card actually delivers what the ad specs says it has and can do.
Yes; but the GTX 1070 doesn't yet support IRAY for rendering in DAZ
Some of the issue seems to be that folks are taking this issue for a specific card and assuming it is / might be / could be the case with every other Nvidia card. I have not seen any traffic that suggests it's a general Nvidia issue. I haven't loaded my GTX980TI enough to know. I can say that I ran my GTX 780 right up to 3072 MB (per gpu-z) before Studio dropped back to cpu rendering.
I also thought it was primarily a gaming issue. I don't get the level of outrage/hurt, especially from a set that typically pops for $50 to $60 for a game. Folks would still have bought the 3.5 GB card for $30 less since the next choice was the GTX 980 at a lot more money or a GTX960 with much less memory and performance. So how did this really affect anyone's decision?
Oh sorry, I thought I read in these forums that was soon to come in September.
At any rate, thanks, I'm glad I deferred the final answer to more knowlegeble folk.
Yeah, at this point all (prospective) owners of Pascal cards are waiting for the Iray update. They just released the SDK, so it'll take a bit of time for it to percolate down to various products.
FWIW, the 970 *was* -- and probably still *is* -- the most popular graphics card on Steam for quite a while despite the memory issue. Most people wouldn't hit the 3.5GB limit when gaming at 1080p, and I doubt that those of us who casually render in Iray will approach the 3.5GB limit either. The most I've hit on my GTX 960 was 3.4GB and that's with the card driving the monitor as well...
Well I CPU render and I hit 12GB RAM in use by DAZ Studio making not too complex scenes (well compared to the real world) - scenes like loading the Full Carnival Preset and two Genesis 3 characters won't CPU render without exceeding 12GB RAM by DAZ Studio. That's one reason I'm not too anxious to go out and buy a 4GB or 8GB nVidia or AMD video card.
...same issue here. a lot of my scenes end up dumping to Virtual Memory which is even slower than CPU. I have 12 GB (well, really 11 GB after Windows and system utilities) and constantly exceed physical memory using Iray on larger scenes. When I built this system 3DL was all we had and the machine was a beast. Iray turns it into a kitten.
Sadly my system wouldn't support even the Maxwell cards very well, let alone Pascal so it would mean building a new one from scratch.
It's probably worth noting, for accuracy, that Iray is only impossible on a reasonable machine if you render busy scenes. If you render scenes that actually fit in 3.5 GB, then you don't need so much.
(One of my hopes with procedural shaders is that it can nicely step around these problems.)
This is something that has been known since Iray released. We advised several times (more than a few dozen times myself) along with the other mods for people to watch the limits of the and be advised that you only get 3.5GB on the GTX 970
I bought a laptop with a supposedly 6 gig 970m card in it that has been showing at 2gb (in both Speccy and Belarc Advisor) ever since a graphics driver update.
I guess I should've added the caveat of texture compression and decimation :)
...yeah, but I don't do portraits.