lol, man the GiF seems too much attractive in the current situation of NVIDIA. and for 970 users, they can bear it. at all

Well, I've got a 1440P 144hz monitor... so i suppose i'll notice the problem before many others will. though so far I haven't noticed any big issues.

FluffmeisterHey I guess Nvidia are ruthless and AMD are inept, one is successful the other just can't seem to make any money.

AMD tried ruthless some time back - unfortunately they tried ruthless with the wrong market segment and ruthless turned to inept almost immediately. After Randy Allen telling anyone who'd listen that Barcelona had a 40% performance advantageover Intel's server parts,AMD doubled down with the false benchmarks and non-existent processors. They've pretty much steered clear of ruthless ever since, reverting back to its strengths (aside from some very dubious Bulldozer PR).

TRWOVThis isn't as bad as the bumpgate and nothing serious came out of that (they paid 200mill to settle but that's it).

Nvidia's actual total charge was $475.9 million(claims and warranty replacements).

TRWOVPlus they can always force the card to just use 3.5GB by drivers (which I suppose it already does).

Likely if they rewrite the driver they'd look to use the 0.5GB partition for Windows/ancillary processes, and prioritize the larger portion for game/app resources, assuming that it is deemed a significant enough issue. From my reading, maxxing out the vRAM still doesn't alleviate the bottleneck from the reduced SM count. A large majority of users seem unfazed due to not experiencing the issue first-hand, and virtually every graphics reviewer seems to view it as a non-story - which leaves a very few (judging by those posting results) affected, those with an affection for synthetic vRAM loading programs, and a vociferous mob of people with no real vested interest.

FluffmeisterHere we go... I'm on Nvidia's payroll now? It's a waste of time discussing anything here frankly.

Then why are you defending NVIDIA if something fishy is clearly going on around GTX 970 specifically?

darkangel0504

Ahahahaha, you're killing me :D Should be GB/s though, but oh well, we get it :D

yeah, Vram throttled or un usable after 3.5 which turned out to be huge throttling for memory bandwidth. and that is killing for NVIDIA users espeially 4K - 1440p users.

RejZoRThen why are you defending NVIDIA if something fishy is clearly going on around GTX 970 specifically?

I'm really not, just pointing out I geniuninly haven't experienced any slow downs in actual real world gaming scenarios, in fact I've been finding myself "going out of my way" pulling extreme DSR res and the like to try and see what happens.

The Nyeah, Vram throttled or un usable after 3.5 which turned out to be huge throttling for memory bandwidth. and that is killing for NVIDIA users espeially 4K - 1440p users.

Fine here @ 1440P

Don't benchmarks in GPU reviews prove these cards work perfectly fine in modern games from 1080p upwards to 4K? They're tested on plenty of resolutions, and W1zzard tests them in plenty of games. Not to mention I run VRAM usage figures in my benchmarks, and I've not come across any particular excessive performance bugs.

I mean, going out of your way to utterly cripple a card, and then proclaiming that the card is crippled seems to be a slightly awkward situation to be in. How many people with 970's are playing games and complaining about it?

darkangel0504

love this meme. which movie is it from?? :)

BiggieShadyI wonder why other Maxwell and Kepler GPUs (like GTX980M, GTX970M or GTX660) with their asymmetric memory configurations aren't affected?

Actually, GTX 660 had a problem too, that only worked like it should on the 3GB version. IIRC, the 660 had its memory in 1.5GB sections. Only 3Gig versions got the full, seamless use of all the memory. the 2 GB versions, which was the vast majority sold, had a 1.5GB portion and a .5GB portion, and reportedly there was some lag as the half gig section was addressed and confined to that half a gig. But no issues reported in the 3GB versions due to symetrical layout.

FluffmeisterHey I guess Nvidia are ruthless and AMD are inept, one is successful the other just can't seem to make any money.

If your implying Nv should give me a free upgrade to a 980, I agree.... I agree with you all!

tell you the truth dude i hope they do.

if not that then those affected should be compensated in some way. i mean while i do not own a 970 myself i have built systems with them for others and talked a few into buying them too. now 3 out of 4 of those friends have seen this issue and i feel like a cunt....

rtwjunkieActually, GTX 660 had a problem too, the 2 GB versions, which was the vast majority sold, had a 1.5GB portion and a .5GB portion, and reportedly there was some lag as the half gig section was addressed and confined to that half a gig.

And I'm guessing Nvidia did nothing about that? They are still selling those cards...

Anybody here smart enough about video card design, who knows why this happens? I'm guessing it's the way the chip is cut down or some parts disabled, and there is no other viable way to do it. Would this be typical of any reduced chip, or rare?

Nvidia could have just sold this with reduced specs I guess, but that would have hurt their marketing against the R9 290 and 290x.

They say picture speaks more than words ...

Looks like they lied ... did I say lied ... I mean miscommunicated the information about missing L2 cache

BiggieShadyThey say picture speaks more than words ...

Looks like they lied ... did I say lied ... I mean miscommunicated the information about missing L2 cache

LOL, they were misquoted.

The thing here is that nVidia has changed the internals of their GPUs since Fermi and thus SM count is the main variable for GPU performance; pixel fillrate isn't ROP dependent for example (both 970 and 980 feature 64 ROPs but 970 has 20% less fillrate at the same speed):

www.anandtech.com/show/8568/the-geforce-gtx-970-review-feat-evga/13

Now we realize that SM count has an effect on VRAM allocation too.

TRWOVNow we realize that SM count has an effect on VRAM allocation too.

It doesn't. Crossbar is used so it wouldn't have an effect.
You could always bin the SM-s without changing memory bus width, what's new is L2 cache is now also binnable. If this was a kepler both memory controllers would be gone also with the other cache module. It would be 192 bit bus card.

Steevowww.anandtech.com/show/8935/geforce-gtx-970-correcting-the-specs-exploring-memory-allocation/4

What to the who now? A specs correction? So it's down to 56 ROPs now? :eek:

Well, now the lower pixel fillrate makes sense, I thought it was due to the changes in the Fermi family and thus the limitation wasn't in the ROPs anymore... so this means that previous GPUs could have exhibited this same behavior just that people didn't notice/cared?? :confused: I recall rtwjunkie saying that 660s also behaved this way. (edit:yes, some posts above)

TRWOVWhat to the who now? A specs correction? So it's down to 56 ROPs now? :eek:

Well, now the lower pixel fillrate makes sense, I thought it was due to the changes in the Fermi family and thus the limitation wasn't in the ROPs anymore... so this means that previous GPUs could have exhibited this same behavior just that people didn't notice/cared?? :confused: I recall rtwjunkie saying that 660s also behaved this way. (edit:yes, some posts above)

True, it had the issue, but not regarded as a problem, more of an efficiency thing, with the VRAM being more efficient with two 1.5GB sections vs a 1.5 and a .5. Remember when it came out, very few games used that much VRAM. 2 was a lot, and 3 was viewed as something a mid-tier chip couldn't use.

Now that they have stated the cards actually have less L2 and ROP's despite advertising otherwise, will people still be defending these liers? Seeing how they are now correcting their shit I have serious doubts on the sincerity of the memory allocation explanation from their side...

No matter how good GTX 970 performs, you still didn't get the advertised goods. End of story.

FordGT90ConceptIt is what Visual Studio uses to denote a platform-neutral binary.

But I wasn't talking about Visual Studio! And I'm pretty sure that the compiler option was called "All CPUs"... In LLVM, I mean. It always used to produce 32-bit binaries no matter what. It's not Microsoft's thing (and is not translated into any sort of intermediate language, but rather to processor instructions directly). I mean, it is, the base toolset (although I always used it in conjunction with at least something), but not the particular tool chain that came into my mind when I saw those screenshots. Don't really know why, maybe because it was so popular back in the days of GPGPU craze.

And why would you write a GPU benchmark using .NET, anyway?.. Like, that's going to be a huge overhead (and the only two "true" .NET parts are going to be the data points and the CLI itself, I think). The rest (the logic) is still going to remain unmanaged, which really makes no sense when you're writing such a small tool.

Funny funny stuff ... engineering department miscommunicated specs to marketing department ... and the whole company is these two departments. This had to be one person responsible that still uses pen and paper and has really bad handwriting ... or they may blame it on autocorrect

VulkanBrosOh oh... Friday I ordered a ASUS GTX 970 Strix..... Damn always unlucky......

Hey VulkanBros,
Just checking out the forums, and stumbled upon this thread regarding the GTX 970 vram issue. I work for NVIDIA out here in Santa Clara, CA.

With all of the questions and concerns going around, let me just jump in and say that while the GTX 970 is just as amazing today as it was when the card launched, we've obviously did not communicated thoroughly as a company.

We understand why GTX 970 owners have concerns regarding the misinformed specs and that we didn't properly explain the memory architecture. We never intended to deceive anyone but despite our best intentions many of you recieved wrong information/specs that impacted your purchasing decision.

The GTX 970 is still an amazing GPU in my opinion and still deserves the praise it has received throughout the community and reviews since launch.

But, with that said, you and others may feel different. You might feel mislead and left out of options. I'd like to inform you that you now have an option.
If any of you feel the need to return the GTX 970 that you have purchased, knowing what you know about the performance in your system, you should return it. Get a refund or an exchange. You should do what will give you the best gaming experience possible and if you need help to get that done let me know, we'll help.

@Brian@NVIDIA can you explain the first release about the issue? at first it seemed to me that there was really no memory issue if the 980 was receiving the same performance hit in the games shown over 3.5gb.
this brings me to thoughts about a article I read and the reviewer shared that when he recommends gpu's he takes vram into little consideration.. someone on the nvidia forum said shadow of mordor can eat over 3.5gb at 1080p :wtf: I would like to see what happens on a 3gb gpu with the same settings and res.. ya know?
could just closing off the slower part of the vram solve most of it and make a game respond differently?

on a side note.. I think its great that someone from the source is here even if its not the best of circumstances. I have long thought the major players for technology need full time representatives on forums such as nvidia, amd, intel, microsoft, samsung.. etc.. people like it when someone from like msi, gigabyte, asus.. etc.. pops in but its not the same.

Brian@NVIDIAI'd like to inform you that you now have an option.
If any of you feel the need to return the GTX 970 that you have purchased, knowing what you know about the performance in your system, you should return it. Get a refund or an exchange.

It would be great if any owner of GTX 970 could do that, but as an NVIDIA representative how can you speak for all different card manufacturers about their product exchange/refund policies?