NVIDIA Responds to GTX 970 3.5GB Memory Issue

[Cainn]

Well, the more I think about it... I'm pretty sure it was a REALLY poorly made map I went on.. I don't have that issue on any other maps I have installed.

Good to hear. 

[Cainn]

Can't they release a temporary driver to limit the VRAM usage until they figure this shit out?

Figure what out exactly? The way i read it was.. Once you go over 3.5 gigs of vram usage you're performance is going to degrade. That's just how that card is going to handle the situation. Period.

[kpreg]

Figure what out exactly? The way i read it was.. Once you go over 3.5 gigs of vram usage you're performance is going to degrade. That's just how that card is going to handle the situation. Period.

That's total shit. If that's true, the 970 will be my last card from nvidia. I would like to see an option to limit the VRAM to only the good modules.

[Misanthrope]

8 Pages in and I don't feel like reading all of it, but I see some people saying 3% difference. If you have 200 frames in 2 seconds and only a handful drop, that will still be a noticeable dip in performance and micro-stutter. 

Is there any reasonable argument to justify this or not really?

[MageTank]

8 Pages in and I don't feel like reading all of it, but I see some people saying 3% difference. If you have 200 frames in 2 seconds and only a handful drop, that will still be a noticeable dip in performance and micro-stutter. 

Is there any reasonable argument to justify this or not really?

 

I only said 3% referring to what Nvidia themselves stated. I do not own a 970, so i cannot specifically say what effect going over 3.5GB causes. I do hear that performance is tanking to an unplayable degree when passing that threshold, so i am inclined to believe the majority of people that provide evidence to their claim. 

 

Sadly, until we get more than a random forum post on Page 47 of a random thread on the Nvidia forum, we will not know much of anything. I would like to see Nvidia showing their means of testing this, and showing that the card will perform at its rated speed throughout its effective memory count. Might be a while before we see that happen though.

[Misanthrope]

I only said 3% referring to what Nvidia themselves stated. I do not own a 970, so i cannot specifically say what effect going over 3.5GB causes. I do hear that performance is tanking to an unplayable degree when passing that threshold, so i am inclined to believe the majority of people that provide evidence to their claim. 

 

Sadly, until we get more than a random forum post on Page 47 of a random thread on the Nvidia forum, we will not know much of anything. I would like to see Nvidia showing their means of testing this, and showing that the card will perform at its rated speed throughout its effective memory count. Might be a while before we see that happen though.

If I understand their statement correctly, this is not something they can correct at all.

[MageTank]

If I understand their statement correctly, this is not something they can correct at all.

It seems so. It almost looks like a blanket post to me. I still stand at my original point that i do not think this was an intentional side effect, because if it was to be limited to 3.5GB of memory, then i am pretty sure they would just put 3.5GB of memory on it (Pretty sure EVGA did that with thsoe 465's or whatever they were, the not-quite 470's). However, if i am wrong, and Nvidia did intentionally cut corners, and cause this 4GB of Vram to only perform its rated speed so long as it is under 3.5GB usage, then i can see them running into some problems with customers, and perhaps even a legal issue depending on if someone can prove that this was an intentional limitation.

 

I am not a lawyer, but advertising specific memory capacity and memory speed, and then not being able to perform that capacity at that speed MIGHT fall under false advertisement, depending on the country and legal definition of false advertisement. I am not calling it false advertisement myself, but just saying that it could potentially come to this if Nvidia admits that this limitation was an intended design.

 

If Nvidia can prove that this design is not having a negative effect, and that the issues people are having goes beyond the GPU itself, then they will be fine. Just gotta wait and see for now.

[mr moose]

I don't think Nvidia was aware of this issue before. If the bug was more dire it would have been found way earlier and rectified. They take their pride in the brand image very seriously (maybe to a fault) and wouldn't try to get away with cheating.

 

I'm not going to shun their graphics cards just because of this (other than the 970 if the bug remains unfixed) Despite having way more respect for the generous red team.

 

To say the very least, this is sure to cause raised eyebrows.

 

To be honest I would be surprised if they didn't know about it.  I believe they do a lot of testing in the R+D stages and should have known  it would do this.  However, I don't think they were expecting it to be an issue in any games, and it's still possible that it is a game/driver issue.  

 

The thing that stands out to me the most is that there have been plenty of reviews showing what this GPU can do under many different settings and across many different games. These results are not about to change so it's not like people have bought a card that won't perform the way it reviewed.   So my next question is does a game have any control over how much ram is used or is that solely the domain of the driver/load condition?

[Albatross]

They should switch the name to "3.5GB" then, because that's what people get, not 4gb.

[kpreg]

They should switch the name to "3.5GB" then, because that's what people get, not 4gb.

Well, errrrr, yes, in a way.

4GB is there, but access to the remaining memory (~.5GB) is painfully slow, enough to ruin gameplay.

Maybe This:

"3.5GB + 512MB*

*Slow access, reliable performance not guaranteed"

[Albatross]

Well, errrrr, yes, in a way.

4GB is there, but access to the remaining memory (~.5GB) is painfully slow, enough to ruin gameplay.

Maybe This:

"3.5GB + 512MB*

*Slow access, reliable performance not guaranteed"

 

That would do, as long as they are accurate with the sell description.

[MageTank]

Should let me CEO Nvidia for a day. I will turn all of the current GTX 970's into GTX 960 Ti's, and use the following slogan:

 

"GTX 960 Ti: The 960 that the original 960 should have been!". 

 

All jokes aside, i hope this issue is atleast properly addressed and laid to rest soon. I plan on buying a 970 soon, regardless of how this situation blows over, but i would still like the piece of mind in knowing that no further complications will arise from this situation. The ITX 970's that are starting to show their heads are looking rather nice, so i need to make a move sooner or later.

[Sanctuary2808]

Bring on the new AMD cards Tis a shame though, I was quite looking forward to trying an Nvidia card as my 2nd ever card...

[johnnysasaki]

8 Pages in and I don't feel like reading all of it, but I see some people saying 3% difference. If you have 200 frames in 2 seconds and only a handful drop, that will still be a noticeable dip in performance and micro-stutter. 

Is there any reasonable argument to justify this or not really?

There's a 3% difference (as the graph comparing gtx970 and 980 shows) but in the average frame rate. They should have posted complete results, with min. and max. fps and also testing methodology.

[Tam3n]

I remember seeing some charts in launch reviews that showed 970 keeping up nicely little behind 980 at 1080p and 1440p, but its performance suddenly bottoming out at 4k while 980 scaled as expected.

[panosk82]

I think what nvidia did was a fraud...

[Mitrailleuse]

I'm happy i decided not to buy the 970...

[nikohd]

Weiird..

[panosk82]

As it turns out it is a decision from nVidia that the ram runs this way.

 

 

nice try linus...

[Verne]

Holy crap. People are acting as if the world was ending. The GTX 970 is still an amazing card and 3.5 gigs + 0.5 gigs still equals 4 gigs.

[nikohd]

Holy crap. People are acting as if the world was ending. The GTX 970 is still an amazing card and 3.5 gigs + 0.5 gigs still equals 4 gigs.

There are people who always based it off on benchmarks and not on real world performance. Sad story bro

[Majestic]

 

 

 

Yes but increasing the graphical options and fiddling with settings to achieve 4GB of VRAM usage is going to hurt performance anyway. You would need neutral way of increasing memory usage, without hurting performance to test the theory properly. Along-side an FCAT analysis of the high memory usage situation, because that would reveal any stuttering. Also, Nai's program that people have been using is clearly bugged because I doubt every single card other than the 970 that people are testing with also have this same problem. So far these are the cards testing with Nai's benchmark that also show the same problem as the 970; 580, 680, 670, 760, 770, 780 Ti, Titan. I'm sure if I searched around I could find more included in that list, but these are just some from off the top of my head that I saw pictures of the benchmark being done. Which means, Nai's benchmark is not accurate.

 

 

Well yes, but then there is Star Citizen. Which just immediatly fills up the pool like all cryengine 3 games. Even at normal settings 1080p. As soon as i start walking around the hanger my frametimes are all over the place when it hits over 3.5GB.

 

I bought the card for Star Citizen/Arena Commander mostly. And this game is, to me, unplayable with the frametimes it has because of this issue.

[Khvarrioiren]

There's something fun brewing here in Finland, seeing that the GTX 970's might end up being illegal to sell here.

[osbios]

GPU drivers normally swap "unused" data to system ram when they are under memory pressure.

My hypothesis:
With ~20 GiB/s the last 0.5 GiB only seems to be useful for swapping.

But the driver is advertising 4 GiB to applications. If you then get a application that tries to use the full amount of VRam you get this big lag spikes where the GPU has to copy big chunks of memory between swap memory and VRam.

 

The same phenomena you can see if you use BF4 + Mantel with <= 2 GiB VRam.

 

In this case you could solve a lot of problems if the driver tells the true "usable" VRam size.

[Majestic]

Welp, Guess i don't need to ponder on what to do with my card. It died anyway.

RIP ASUS STRIX 24-01-2015 - 25-01-2015. You lived a good life. Such wonderful quality products these days.

 

Back to the trusty HD4600. Atleast that thing works without issues.