Support NeoGAF

[Moral Panic]

Not unless someone manages to turn that benchmark into a game or something.

I assume if the driver detects it is a game then it will give it proper access to the 0.5GB segment. I highly doubt that segment runs slower than the ram in the xbox 360 as shown in those "benchmarks" otherwise we'd see a much worse bottlenecking.

Beyond satisfying our curiosity, though, I'm not sure what else to make of this information. Like the ROP issue, this limitation is already baked into the GTX 970's measured performance. Perhaps folks will find some instances where the GTX 970's memory allocation limits affect performance more dramatically than in Nvidia's examples above. If so, maybe we should worry about this limitation. If not, well, then it's all kind of academic.

 

[hoserx]

do you own NVidia stock or something

I own literally one share of nvidia stock, lol.

 

[Shao Kahn Brewing a Stew]

Where's my "Nvidia, fuck you!" gif when you need it?

 

[cyberheater]

Amazing.

I can't really believe that a vid card manufacturer would sell a card that can only access 3.5GB out of 4GB at full speed. That's pretty bad and if I was a GTX970 owner I'm be pretty pissed off.

 

[crazepharmacist]

Yeeeeah, this doesn't show anything other than you get stuttering when exceeding the 4GB of RAM on the card. Shocking!

Actually no, he was hovering right around 3700mb of vram usage in the second half of the vid.

 

[Gormenghast]

The way this is being tested in games now, supersampling and at very low framerates, probably means you are minimizing the effect.

If you have 60FPS and running at full 4Gb then the results would be very, very bad.

So this may be a problem for "future proofing", in all games that would use 4Gb AND run at 60fps this is going to be a real issue.

If Nvidia wrote drivers so that more than 3.5Gb is almost never used means that YOU CAN'T TEST THIS. If you force games to use more ram through supersampling you're merely hiding the memory bandwidth bottleneck behind another bottleneck. So it's not surprising that in those benchmarks you don't see differences.

 

[ntropy]

how come the hardware reviews didn't catch this?

 

[Xdrive05]

I sold my gtx 770 4GB to get one of these. Ugh. Will wait and see if there's any price impact or pack-ins that come from all this.

How soon until AMD releases their 300 line?

 

[FunkyMunkey]

Are the corporate apologists in this thread just being disingenuous on purpose or do you all really not understand how Nvidia fluffed that data in a dishonest way by using averages, and user tests are showing that this may actually be a problem?

And telling people to deal with it because the card is still great... what? If Nvidia purposely misled consumers about how much vram the card realistically can use for gaming, it's a problem. I don't care if I can play bf4 at 120fps right now, what happens in a year or two when more and more games require that 4GB vram even for 1080p? I need more tests

So this may be a problem for "future proofing", in all games that would use 4Gb AND run at 60fps this is going to be a real issue.

But but but future-proofing is dumb! The card is still great and no one should be mad because it can handle current games fine /s

 

[whatsinaname]

Personally; no, but there are a lot of people who already think 4.0 GB isn't enough to 'future proof' their systems. Those would've most certainly had second thoughts about buying the card had this been known before launch and I can't blame them.



Of course it is still the absolute best card available in its price range at the moment. This technical misrepresentation doesn't change any of the glowing reviews or real-world gaming benchmarks done so far and its immense popularity is good indication of the great value on offer here. Unless further investigation shows significant game performance limitations this doesn't diminish the 970 itself to a new owner, just Nvidia's reputation depending on your opinion on the way the handled the situation and possibly false advertising.

The only thing I would advise is to postpone purchasing the card until we're sure there isn't more to this issue than Nvidia is willing to admit.

That is what I plan to do. I will be in the market for a card in April, and this still looks like my best bet.

Both these companies have shit the bed so many times, if I went with my instinct of not wanting to buy anything from either of them, I would be left with no option.

 

[opticalmace]

I sold my gtx 770 4GB to get one of these. Ugh. Will wait and see if there's any price impact or pack-ins that come from all this.

How soon until AMD releases their 300 line?

I'm guessing the earliest would be march and the latest would be june. I think I've read rumours saying something like that...

 

[Toad King]

how come the hardware reviews didn't catch this?

Because maybe, just maybe, it doesn't effect 99% of applications in any meaningful way outside of benchmarks?

 

[Thrakier]

I remember a similiar discussion about the 660TI. Never had a problem with it.

 

[Gormenghast]

how come the hardware reviews didn't catch this?

Once again, because drivers try to hide the problem by not using the last 0.5Gb.

We would see a problem in benchmarks only if you have high fps and total ram usage.

 

[Xyber]

http://cdn.overclock.net/7/78/78ab3216_4vciohfw.png
And nvidia themselves admitthed that the memory is segmented to high and low performance memory.

According to other posts, people have seen the same results on Titans, 680 and other cards too.

I can only speak for myself and I have not experienced anything close to the stuttering linked in the video on the previous page. The only stuttering I can remember was in Watch Dogs and that was not because of my card.

I have no regrets buying this card on launch day, but I also don't have any plans to keep it for years. This was just a stopgap in my wait for the big Maxwell cards.

 

[Nachtmaer]

I'm guessing the earliest would be march and the latest would be june. I think I've read rumours saying something like that...

Yup. According to this it should arrive before Computex, so around June.

 

[Bricky]

That is what I plan to do. I will be in the market for a card in April, and this still looks like my best bet.

While the 970 remains the best price/performance card available today you'll have to keep an eye out for AMD announcements. There is a decent chance the R9 3XX series has hit the market by then and there is no way AMD isn't going to try and lure potential GTX 970 buyers to their brand by whatever means possible.

Nvidia isn't going to release anything surpassing it in the price/performance category except maybe for some high VRAM iterations.

 

[Skux]

Are the corporate apologists in this thread just being disingenuous on purpose or do you all really not understand how Nvidia fluffed that data in a dishonest way by using averages, and user tests are showing that this may actually be a problem?

Yeah this is bullshit and they need to be called out on it. We were advertised 4GB of VRAM - 4GB of uniform, consistently performing VRAM. Not 3.5GB of good VRAM and a trash 512MB chunk.

I was honestly looking at the 970 for my upgrade. Now this adds another check I have to make to ensure I actually get what is listed on the box.

 

[Gormenghast]

While the 970 remains the best price/performance card available today

In Europe the 290x costs exactly the same and even usually performs better. The HUGE downside it that it uses TWICE as much power as the 970, but otherwise the memory bandwidth is much better and without these issues.

 

[inpHilltr8r]

#ramgate

http://en.wikipedia.org/wiki/Ramsgate

 

[Moral Panic]

Are the corporate apologists in this thread just being disingenuous on purpose or do you all really not understand how Nvidia fluffed that data in a dishonest way by using averages, and user tests are showing that this may actually be a problem?

Err what? User tests have shown the opoosite: http://www.reddit.com/r/pcmasterrace/comments/2tfybe/investigating_the_970_vram_issue/

In conclusion: it really seems like that everything is fine - there was no performance drop in DA:I as the VRAM usage gradually rised from 3GB to 3,6 (according to my Nai's benchmark results, the "slow" area starts at 3200MiB, which is equal to 3355MB), Watch_Dogs ran at 4k with 4xmsaa at ~20-22 FPS, which is much better than I expected, and at 1440p with 8xmsaa at cinematic (~30-32) fps, with no stuttering that was present in 4k+4xmsaa due to VRAM being simply full.
And the most important part: GPU Usage was pretty much locked at 99% (except during sutters in w_d at 4k+4x), indicating that there was no bottlenecking in terms of memory bandwith suddenly becoming worse etc.

Nvidia's tests have shown that relative performance vs GTX 980 is maintained even when the ram usage goes above 3.5GB, even furmark shows 0 performance degradation going above 3.5GB. The only evidence that this memory configuration is a problem is from a single benchmark which accesses the memory in a different way to most games.

 

[AngryPuppy]

The people bitching about this have no life, seriously.

 

[THEF3AR]

AMD 300 series cards will feature "FULL 4GB HBM" on the box

 

[impact]

The people bitching about this have no life, seriously.

You're bitching about people bitching about products they paid for. You must be a fucking basement dweller then if they have no life LOL

So glad I didn't pick up a 970 yet. I'll just keep saving and go for a 980.

In Europe the 290x costs exactly the same and even usually performs better. The HUGE downside it that it uses TWICE as much power as the 970, but otherwise the memory bandwidth is much better and without these issues.

Well that and it's AMD.

 

[ref]

Well, I feel a little better knowing I didn't get SLI 970's instead of SLI 980's.

 

[LilJoka]

My results from the old thread:

Heres Shadow of Mordor from my system
i7 3770 4.2Ghz
Asus P8Z77I Deluxe
Samsung 2x4GB 2133Mhz CL10
GTX 970 1500/3880Mhz

The 1080p one was limited to 62fps by RTSS. The 2880p one was playable not stuttery.

Mean: 18.5ms
Variance: 90.16
Standard Deviation: 9.49531

Mean: 33.4ms
Variance: 159.5
Standard Deviation 12.6

More heavy VRAM usage

You can see the high frame times have big drops in GPU usage which is where it's having to use the 500MB section

You can also see how it tries to hang on to using the 3500MB section at the start.

 

[DSN2K]

I'd want a refund if I had purchased this. I fully expect AMD to take advantage of this regarding marketing their cards next time round.

 

[Krakn3Dfx]

The people bitching about this have no life, seriously.

Lol, what.

Huang is that you?

 

[Bricky]

In Europe the 290x costs exactly the same and even usually performs better. The HUGE downside it that it uses TWICE as much power as the 970, but otherwise the memory bandwidth is much better and without these issues.

Well, how much this bothers you is completely subjective but I would say that absolutely makes the 970 the better price/performance card. Power costs money too!

But both are good cards at a good price, let's not start a silly discussion over which deserves the imaginary 'price/performance award' more.

 

[EmptySpace]

so that's why it was "cheap". ahaha.

 

[Costia]

No I just haven't seen compelling evidence that this impacts real-world performance in a meaningful way. People saying the card is "gimped" or that it somehow isn't what they paid for are delusional. Especially insane are the people saying it should have been sold as a 3.5GB card.

Then look at it from a different perspective:
1)The card is advertised on NVIDIA's site as 4GB, 256bit. In their response NVIDIA admitted that it is not the case and it is actually 3.5GB high banwidth + 0.5GB low bandwidth
2) If you look carefully at NVIDIA's response regarding the low bandwidth of the last 0.5GB you might notice that it doesn't actually contain any data on the bandwidth of said memory region.

NVIDIA is not a game benchmarking company. They are a GPU designer/manufacturer and it is their duty to provide accurate specifications for their chips regardless of their opinion of the performance impact.
If they (and you) think the memory segmentation and reduced bandwidth doesn't have any real world impact - that's fine. But they still have to disclose it in their spec.

I don't think the lower bandwidth of the last 0.5GB will have any significant performance impact, but I still feel I have been lied to.
The 970 is still great as a 3.5+0.5 card, but it should have been advertised as such (and not as a 4gb 256bit) from the beginning.

 

[Jonm1010]

Man this whole thing puts me into a shitty position.

I just bought a 970 to put into my R1 alienware x51. It's still in the box. Got here two days ago.

Now I hear all this and wonder if I made a mistake. So much conflicting evidence right now.

But if I send it back and go for a 980 I will have to buy a new motherboard because the R1 motherboard can't take the 980 and so that is another cost tacked on.

...Ugh

 

[Zane]

Then look at it from a different perspective:
1)The card is advertised on NVIDIA's site as 4GB, 256bit. In their response NVIDIA admitted that it is not the case and it is actually 3.5GB high banwidth + 0.5GB low bandwidth
2) If you look carefully at NVIDIA's response regarding the low bandwidth of the last 0.5GB you might notice that it doesn't actually contain any data on the bandwidth of said memory region.

NVIDIA is not a game benchmarking company. They are a GPU designer/manufacturer and it is their duty to provide accurate specifications for their chips regardless of their opinion of the performance impact.
If they (and you) think the memory segmentation and reduced bandwidth doesn't have any real world impact - that's fine. But they still have to disclose it in their spec.

I don't think the lower bandwidth of the last 0.5GB will have any significant performance impact, but I still feel I have been lied to.
The 970 is still great as a 3.5+0.5 card, but it should have been advertised as such (and not as a 4gb 256bit) from the beginning.

The only numbers I actually care about are real-world performance numbers. I've never bought a piece of hardware based on the specs given by the manufacturer. They rarely mean anything.

 

[Dictator93]

Man this whole thing puts me into a shitty position.

I just bought a 970 to put into my R1 alienware x51. It's still in the box. Got here two days ago.

Now I hear all this and wonder if I made a mistake. So much conflicting evidence right now.

But if I send it back and go for a 980 I will have to buy a new motherboard because the R1 motherboard can't take the 980 and so that is another cost tacked on.

...Ugh

Why cant the mother board take a 980? That is more a question of PSU...

 

[Costia]

The only numbers I actually care about are real-world performance numbers. I've never bought a piece of hardware based on the specs given by the manufacturer. They rarely mean anything.

This is perfectly reasonable.
But it still doesn't mean that NVIDIA can write whatever they want in their specs. The spec still has to be correct, especially considering this is not a newly found bug but an intentional HW design decision.

 

[Easy_D]

If someone is legit done with Nvidia after this please send me your 970 for 100 dollars.

Spoiler

Please ;-;

 

[Zane]

This is perfectly reasonable.
But it still doesn't mean that NVIDIA can write whatever they want in their specs. The spec still has to be correct, especially considering this is not a newly found bug but an intentional HW design decision.

I think it's fine to be a little sour about that but I can't imagine being upset about it for more than a few hours, unless there's a significant real world performance impact and we haven't seen any decisive evidence of that yet. If it turns out that this is an issue that significantly hamstrings the 970 in comparison to other cards on the market, I will join everyone else in being upset. But not until then. Decades of experience with PC hardware have made me hard to convince.

 

[pestul]

Knowing Nvidia's scummy practices of years old (detecting .exe's/deliberate gimps/FX5900 fiasco), it wouldn't surprise me in the slightest if there was something hard-coded into their drivers for handling memory situations above 3500MB on the GTX970.. even appear as if it's using more in MSIAfterburner etc, but actually not. This is brutal. Now that it's out there, it will be tested to death to see if there really is something more to it at least.

I hope it is handled as gracefully as they state, but I seriously doubt it.

 

[jimmypop]

There's some cringeworthy manufactured outrage going on in these threads. The worst part is the people who clearly misrepresent or otherwise don't understand the technical specifics.

 

[Costia]

I think it's fine to be a little sour about that but I can't imagine being upset about it for more than a few hours, unless there's a significant real world performance impact and we haven't seen any decisive evidence of that yet. If it turns out that this is an issue that significantly hamstrings the 970 in comparison to other cards on the market, I will join everyone else in being upset. But not until then and decades of experience with PC hardware has made me hard to convince.

I take it more as warning for future purchases than anything else. I still think the 970 is a great card, but I will be a lot more careful in the future before buying a relatively new card made by NVIDIA.

 

[Zane]

I take it more as warning for future purchases than anything else. I still think the 970 is a great card, but I will be a lot more careful in the future before buying a relatively new card made by NVIDIA.

I think that's fine but to be fair, I think a lot of people were saying that after the FX 5900 and this is nowhere near that sort of disaster (also look how quick they bounced back after that. 6600 GT is still the GOAT).

 

[Rafterman]

So.. a paired down GPU performs slightly worse than a higher spec GPU? Would it be fair to say that people are just worried and concerned because a limitation is brought to their attention rather than an actual legitimate issue with the GPU?

Or am I being unfair and misrepresenting the issue? I haven't really kept up with this.

That's exactly what's going on. The 970 was seen as an amazing card this time last week, 99% of the people who are complaining had no idea this limitation even existed, and now today the card is shit, Nvidia is horrible, and the world is ending. People are going to feel really fucking silly when this time next year, and the year after that, the 970 is still performing in line with it's advertised power. The good is that will all of these chicken littles around anyone looking to buy an 970 used will probably get an excellent deal.

 

[Ronabo]

Just my luck that the first PC I build and the first GPU I buy since AGP slots were a thing ends up having an issue.

Oh well.

 

[Dryk]

I can't wait for someone to bump this thread in a few years time when someone releases a game that needs all 4Gb and it just doesn't work properly

 

[Dictator93]

There's some cringeworthy manufactured outrage going on in these threads. The worst part is the people who clearly misrepresent or otherwise don't understand the technical specifics.

People were falsely advertised to and bought the product thinking "x" was the case. It may not have a massive real world difference in every game scenario (but does in some, obviously), but there is still the fact that people thought one thing that was not true.

 

[LouisaFerre]

Yeah, as the owner of a 970 I'm really not comfortable with that. I'm still satisfied with the performance of the card but I'm not comfortable with Nvidia actively deciding to do this before just advertising the card as having 4GB of VRAM. It is technically true but I do feel like it's misleading, especially since they never said anything more specific until customers started to notice.

 

[chromatic9]

I'd like to see a video and some FCAT testing on Call of Duty AW with the settings shown in the OP since its still in the 40fps range when using.above 3.5GB.

All games I have that pass 3.5GB are already in the 10-20fps range and will stutter anyway.

 

[Momentary]

Now it's time for AMD to crush NVIDIA with that memory stuff that hasn't really been confirmed for their next architecture yet.

 

[espher]

I think it's fine to be a little sour about that but I can't imagine being upset about it for more than a few hours, unless there's a significant real world performance impact and we haven't seen any decisive evidence of that yet. If it turns out that this is an issue that significantly hamstrings the 970 in comparison to other cards on the market, I will join everyone else in being upset. But not until then. Decades of experience with PC hardware have made me hard to convince.

"Games/applications out today that the card can run fine run fine so the card is fine" is not exactly a convincing argument to me. If this was a known ahead of time I probably would still purchase the card since the value proposition is still there, but at least then I would know the risks/caveats.

If I buy a combo at a burger joint that is supposed to have a burger, fries, and a beverage, I'm going to be touch mad if I don't get the fries even of the burger and beverage fill me up. The fact that, in aggregate, they only rarely forget the fries in my combo, or that I had a coupon for the combo (so it was cheaper than normal) may temper my annoyance but they don't mean that the burger joint didn't fuck up or didn't sell me what was advertised.

Yeah, I know it's a bad analogy.

 

[AHA-Lambda]

The people bitching about this have no life, seriously.

Oh you'll go far here son.