http://www.pcper.com/reviews/Graphics-Cards/NVIDIA-Discloses-Full-Memory-Structure-and-Limitations-GTX-970
>despite initial reviews and information from NVIDIA, the GTX 970 actually has fewer ROPs and less L2 cache than the GTX 980. NVIDIA says this was an error in the reviewer’s guide and a misunderstanding between the engineering team and the technical PR team on how the architecture itself functioned.
>>46277237
Just got to that part, but I shouldn't be surprised as it's nvidia we're talking about here.
>GM204 allows NVIDIA to expand that to a 256-bit 3.5GB/0.5GB memory configuration and offers performance advantages, obviously.
It really is a feature then, huh.
i'm kind of curious if this is a one time thing or they got away with shit like this in previous generations?
>>46277311
Do we need to list Nvidias lies/bullshit one more time? They do this crap a lot but they have a huge marketing/payoff budget.
>>46277152
TL;DR - wow, it's nothing
>>46277334
please post the list.
>Yfw you bought a GTX 970
>>46277351
>They as much admitted they lied and falsely advertised the product
>They even went out of their way to mislead reviewers with false info.
No nothing going, good old Nvidia at it again. Be sure to buy keep buying their defective and falsely advertised products since they so great!
Pick related.
>>46277166
can someone please tell me the source of that pic, its driving me nuts
>>46277237
aaaahahahahahaaaaaaaaaaaa
HAHAAAAAAAAAA
>>46277237
thread theme: https://www.youtube.com/watch?v=Qf63D4EQtV8
>>46277311
They got away with it with GTX 660 at least.
It had the same slow 0.5GB partition.
But it was the poorfag card so nobody cared much.
Can some of these GPU brands get sued for false advertising?
On the GTX 970 box, it says "4GB of GDDR5 memory". Yet, the graphics card only provides 3.5GB of actual GDDR5 memory speed with the remaining 0.5GB being much less than GDDR5 memory speed.
They clearly ripped off the consumer.
>>46277468
Until the day I actually see it affecting games by a significant amount, it will not bother me.
Find a benchmark and compare 2 cards that do and don't have this issue with the exact same configuration and find an issue for me. Please.
>>46277468
By legal technicality that wouldn't fly. It still has 4gb of gddr5. At least it wouldn't fly on that argument alone. There's still a case to be had though.
>>46277311
>>46277396
>nvidia lies
>>46277494
It may not bother you, but this will become a clear problem in the future and is already affecting high-end consumers right now.
Nvidia better have some strong damage control because this could be a class action lawsuit here.
>>46277494
Why buy a 970 then if a lesser mem'd card would suit your needs?
>>46277494
People starting looking for this when the 970 was stuttering. But don't worry stuttering is a feature according to Nvidia.
>>46277510
the 970 is actually about 184watts running at stock speed and some aftermarket 290's draw 210-215watts at stock speed
but over all thats generally accurate if the 980 was selling at 350$ that would be the actual card that nivida promised the 970 was
>>46277494
It's perfectly fine for you if it does not bother you, but it's not about how it still "works for me" and more about how they advertised higher specs than actually used for 4 months (ROP and L2 cache come to mind).
what you are refering to as the gtx970s 4gb vram is in fact 3.5gb/.5gb vram or as ive begain calling it 3.5gb + .5gb vram. the 970 is not a 4gb vram graphic card itself but rather another disfuntional product from Nvidia split into two separate memory partitions.
>>46277510
Still more efficient than amd, also what is thermal design power?
https://forums.geforce.com/default/topic/803518/geforce-900-series/gtx-970-3-5gb-vram-issue/105/
The rage is real
>Let's be blunt here: access to the 0.5GB of memory, on its own and in a vacuum, would occur at 1/7th of the speed of the 3.5GB pool of memory.
>There is 4GB of physical memory on the card and you can definitely access all 4GB of when the game and operating system determine it is necessary. But 1/8th of that memory can only be accessed in a slower manner than the other 7/8th
>>46277152
heh, i like how they released a full disclosure after everyone bought one
>>46277609
idiot design because if the just removed the .5gb partition the 3.5gb cache would refill itself with new textures at the proper speed
the card would literally run better if it was a 3.5gb card
>>46277642
uh ueah. but 4gbs markets better
>>46277651
Kek
>But 1/8th of that memory can only be accessed in a slower manner than the other 7/8th, even if that 1/8th is 4x faster than system memory over PCI Express. NVIDIA claims that the architecture is working exactly as intended and that with competent OS heuristics the performance difference should be negligible in real-world gaming scenarios
>v1.x: 4 GB/s (2.5 GT/s)
>v2.x: 8 GB/s (5 GT/s)
>v3.0: 15.75 GB/s (8 GT/s)
How the fuck can they get away with this shit?
>So the problem lies with the OS/Drivers sometimes using the incorrect pool. That means the 970 should still be fixable by drivers alone.
>>46277757
Because the company's name is Nvidia and not AMD
>>46277764
>mfw
>>46277757
They already have. Thank the fanboys.
You can already see them shouting about power efficiency to drown Nvidia lies.
They did the same thing with GTX 660 too.
The lies are here to stay.
>>46277764
What?
>>46277844
https://forums.geforce.com/default/topic/803518/geforce-900-series/gtx-970-3-5gb-vram-issue/post/4435059/#4435059
>>46277862
That guy is a retard.
>>46277888
Absolutely. Makes for good entertainment however
>>46277427
Bought both 660Ti and 970 at launch... Welp time to switch back to AMD
>>46277917
Oh, definitely.
>>46277430
i second that
>>46277428
>mfw all the people that bought the 970 last week hoping to get a free upgrade to 980 as a result of this.
get rekt m80s
Would a driver update that just limits the 970 VRAM to 3.5GB improve performance?
>>46277453
>https://www.youtube.com/watch?v=Qf63D4EQtV8
>>46277430
neogaf ;)
>>46277970
For everything that would otherwise forcibly use the remaining 500MB, I suppose it would. Not an expert on this topic though.
>>46277970
No. Better have a slow 512Mo than no more VRAM.
>>46277651
Excellent
>>46277970
Not performance as Nvidia defines it, but you'd never run into the stuttering caused by the slow vram partition.
>>46277958
>mfw literally noone did that though
because that would be utterly retarded. Do you really think anyone is going to get their card RMA'd or refunded by Nvidia? Jews gonna Jew.
>>46277970
This would be a bad idea, A better "band aid" solution would be allocate non gaming VRAM (such as windows/areo/browser/monitors/etc) into the "slow" part and everything gaming into the 3.5GB fast memory. Instead of the current that it's allocate everything until it hits 3.5GB
Time to get a refund
>>46277764
512MB of VRAM is in a different, smaller pool. This pool is accessible at 1/7th the bandwidth of the 3.5GB pool. They can't "fix" shit because that's how the GPU is made. Games will very likely be much less affected but it is what it is.
>>46277642
that's not how it works. it would have to start fetching shit from RAM which is a shitton slower than the 0.5gig cache.
http://www.ebay.com/itm/ASUS-Strix-GeForce-GTX-970-DCII-OC-Overclock-4GB-PCIe-STRIX-GTX970-DC2OC-4GD5-/171657039216?pt=LH_DefaultDomain_0&hash=item27f78e8d70
It's happening!
>>46278056
article says it's a 5-6% hit versus the theoretical situation where you have a '970' with enough rops to put the 4gigs all in one partition
>>46278099
>7 sold
AAAAAAHAHAHAHHAHAHAHAHAAHA
so basically nvidia advertises that there is 4gb vram, and there is 4gb of ram, whats the problem here?
>>46278125
It's 3.5GB of Fast /0.5GB of slow
Which is worse than 3.5GB fast only. >>46278135
>>46278125
>>46278099
I bought mine last month... If I knew that retards would sell their brand new 350€ GPU for those kind of price, I would have wait...
>>46277970
Probably. But then you have a 3.5GB card that is sold as a 4GB one, which isnt allowed (in Europe at least).
>>46277152
Im so happy for early-adopter-electronics-in-general-idiots. Ill buy myself a nice used GTX970 2months old, under warranty for 40% of retail price.
I can live with GPU with 3,6 GB VRAM for 200$
>>46278141
of course, but it is still 4gb is it not?
that's the available technology at that price point
if you wanted 4gb all fast then you would obviously opt for the more expensive 980
>>46278037
>implying Nvidia sells you cards directly
>>46278157
fraud post
>>46278101
No, that's what NVIDIA says 'real-world gaming performance' is. That memory benchmark is pretty much spot on, reading almost exactly 1/7th the bandwidth on the upper 512MB.
Even if NVIDIA's numbers aren't straight-out lies (they probably aren't) they still don't mention exactly how and what they measure. Is it 4-6% on average over multiple games? How does it affect frame time variance, not just FPS? What's the performance like in the worst case scenario?
>>46278150
shills got nothing on the facts
>>46278180
>next gen nvidia will do the same with all cards
What then?
>>46278109
even at 70 bucks no one wants it
wow
>>46278211
>shills
I'm actually just a sad customer. But if something is too good to be true, it probably isn't.
>>46278180
Or opt for an AMD GPU
>>46278099
>Goes from 7 to 8 sold since this being posted
Ahahaha, one of you faggots bought this shit, enjoy your extra binned gpu
>>46278244
sorry to hurt your feelings, it's a great card to use on 1080p single monitor for maybe a year, but after that the memory nerf probably starts to show, gotta crank down dem grafix
>>46278197
>>46278000
>>46278022
Did you idiots even read the article
>NVIDIA’s performance labs continue to work away at finding examples of this occurring and the consensus seems to be something in the 4-6% range. A GTX 970 without this memory pool division would run 4-6% faster than the GTX 970s selling today in high memory utilization scenarios. Obviously this is something we can’t accurately test though – we don’t have the ability to run a GTX 970 without a disabled L2/ROP cluster like NVIDIA can. All we can do is compare the difference in performance between a reference GTX 980 and a reference GTX 970 and measure the differences as best we can, and that is our goal for this week.
Basically when it hits over 3.5gb the entire card takes a 5% performance hit, not accounting the delayed frame times that makes the game a stuttery mess.
At this point it's clear a capped 3.5gb 970 will outperform a normal 970 at high resolution gaming.
>>46278283
>statement by NVIDIA's performance labs
You think we should still trust them after this mess?
>>46278283
>At this point it's clear a capped 3.5gb 970 will outperform a normal 970 at high resolution gaming.
Possibly, but not necessarily. The shit 512MB is still faster than hitting system RAM, it would only be faster if the shit 512MB are being used when the good 3.5GB aren't completely full.
>>46278283
Nvidia doesn't think less stuttering = improved performance.
Only average fps is performance.
>>46278318
>mixing 20gb/s ram and 170gb/s ram
>faster
They tried their best to hide this with that 3.5+0.5GB division.
>>46278299
I believe the issue was caused by overdifht from working with brand new technology.
Is it lying if it wasn't premeditated?
>>46278283
Nvidia gimped their cards on purpose and you trust their "performance labs"? Seriously?
>"NP guys, i wont eat the sheep, i swear" -Wolf
>>46278344
>overdifht
Oversight.
What the fuck SwiftKey
>>46278342
Faster than going over PCI-E to system RAM, not faster than a proper memory interface anon.
>>46278344
That would make it even worse, as that would imply that nvidia did not even know what they were doing in the first place.
Its easy, you buy from jews, you get jewed
Just get over it
>>46277237
>nvidia straight up lied about the 970's specs
i'm sure the nvidia shills will defend this
So, does the GTX 770 have this problem too?
Both should be just binned 780/980 chips, of course, but seems like gimping the memory is retarded.
![christian-bale-the-dark-knight[1].jpg](https://i.imgur.com/72uF7Oc.jpg "christian-bale-the-dark-knight[1].jpg")
>>46277166
>>46277311
>>46277744
>>46278188
>>46278211
>>46278244
>>46278277
>>46278299
>>46277888
>>46278000
Anyone got the Linus Torvalds .gif where he gives the finger to Nvidia?
>>46278364
It's still a terrible, terrible idea.
>>46277152
top.kek
>>46277609
Anandtech
>
"This in turn is why the 224GB/sec memory bandwidth number for the GTX 970 is technically correct and yet still not entirely useful as we move past the memory controllers, as it is not possible to actually get that much bandwidth at once on the read side.
>GTX 970 can read the 3.5GB segment at 196GB/sec (7GHz * 7 ports * 32-bits), or it can read the 512MB segment at 28GB/sec, but not both at once
>>46278398
dunno, but gtx 660 had a similar 1.5gb+0.5gb setup
but hey we like the lies
https://www.youtube.com/watch?v=Qf63D4EQtV8
>>46278415
>The only way Nvidia can make this right is to give me a full refund so I can go buy an R9 290x
posted in the thread on forums.geforce.com
>>46278454
this is fucking great
>>46278454
It's obvious amd is hiring patel and friends to spam the forums.
>>46278480
>amd
>having any money
Don't know about you guys but I use my card for gaming not running benchmarks
![1422205148713s[1].jpg](https://i.imgur.com/wku3GV2.jpg "1422205148713s[1].jpg")
>>46278495
>>46278502
>it's fine
>it's only a benchmark
>i'm not a sucker
>>46278502
I can't even tell if these posts are ironic or not anymore.
>>46278495
>One rupee has been deposited into your money-sack for this post, vakshishandehapish
>>46278540
That would explain the level of their reading comprehension.
>>46278099
>Transportation time need 25--35 days.
>Please wait.
Someone getting scammed
>>46278540
>vakshishandehapish
So the long and short of this is that the first 3.5 GiB of logical memory space are striped over seven GDDR5 controllers and the last 0.5 is not striped at all, so that the L2 cache with the disabled twin doesn't (usually) get fucked?
Yeah, this seems like fucking horrible design. If the last 0.5 GiB of RAM can't be meaningfully used AND Nvidia has gone out of their way to prevent it from being used, they should have just not populated the GDDR on the PCB location mapping to the disabled L2's controller.
This Majestic dude is rustling my jimmies
>>46278588
Pretty much. It would have been better as a 3.5GB card, but selling it as a 4GB card was just too sweet a deal.
Not one review site can induce stuttering.
AMD blown the fuck out.
>>46278527
Realistically it probably is 'fine' for most people.
That being said I know I won't be buying 970s to replace my old 7970s, I only hope something else comes out before Witcher 3.
>>46278588
They need to sell those binned chips somehow anon...
>still faster than the amd equivalent
>and has lower power consumption
what exactly is the issue
>>46278610
>review sites induce stuttering
>>46278610
Extremetech
http://www.extremetech.com/extreme/198223-investigating-the-gtx-970-does-nvidias-penultimate-gpu-have-a-memory-problem/2
>With that said, our 4K test did pick up a potential discrepancy in Shadows of Mordor. While the frame rates were equivalently positioned at both 4K and 1080p, the frame times weren’t. The graph below shows the 1% frame times for Shadows of Mordor, meaning the worst 1% times (in milliseconds).
>The 1% frame times in Shadows of Mordor are significantly worse on the GTX 970 than the GTX 980. This implies that yes, there are some scenarios in which stuttering can negatively impact frame rate and that the complaints of some users may not be without merit. However, the strength of this argument is partly attenuated by the frame rate itself — at an average of 33 FPS, the game doesn’t play particularly smoothly or well even on the GTX 980.
>>46278619
my guess is that radeon 300 series ships with witcher3
>>46278636
No issues, only features here.
I sure hope more companies start lying about their products and fucking the customers over.
It doesn't matter anyway, it's still good enough.
The card is still good, but the shit Nvidia pulled here leaves a taste of ass in the mouth.
>>46278680
told
>>46278680
so basically people complaining about it not being able to do something it isn't supposed to
>>46278699
well nvidia should be grilled for lying about it but still, this isn't the holocaust you guys are making it out to be
>>46278588
>>46278608
>>46278628
> marketing: 4GiB, 256b wide, 224 GB/s, (2048 kiB L2?)
> 99.?% of the time: 3.5GiB, 224b wide, 196 GB/s, 1892 kiB L2
> remaining 0.?%: +0.5GiB, 32b wide, 28 GB/s, 256 kiB L2
https://forums.geforce.com/default/topic/803518/geforce-900-series/gtx-970-3-5gb-vram-issue/post/4435196/#4435196
>guy posts how much he has loved Nvidia over the years for giving him proper VGA's
>guy asks if he should buy AMD and hang himself, or buy GTX 970 and still hang himself
>asking Nvidia to fix this issue, or else "https://www.youtube.com/watch?v=IVpOyKCNZYw"
>Nvidia's response is nuking his comment
>>46278743
what did they lie about?
the card has 4gb of vram
>>46277460
The difference was that reviewers knew about this and that most of them focused on that (besides other things) when reviewing it. It was not misleading, it was something that you could easily find out beforehand. The whole 970 situation is just plain bs, I really hope that they get fucked in a few courtcases in the EU, I doubt that it would fly in the U.S. though.
>>46278757
>what did they lie about?
Memory bandwidth and memory subsystem architecture.
>>46278415
Lmao
>Confirmed hardware problem
>Confirmed broken
>Confirmed the shills in damage control
>>46278757
>hurr durr last 0.5gb runs at 1/7 of the speed
You'd be saying that has 6GB of vram if they duct taped a 2GB DDR2 stick on it.
>>46278774
why didn't they just ship the card with 3.5GB?
>>46277562
>I would like to interject for a moment.
>>46278766
So even worse lies in this case?
Well, Nvidia will at least bring 4GB of memory to the GTX 960. Now the 960 will be a good buy, right guys?
guys?
http://techreport.com/news/27726/report-4gb-of-ram-coming-to-gtx-960-in-march
>>46278791
NOW WITH BRAND NEW RESERVE CACHE OF DDR2 MEMORY ON CARD FOR ULTRA-RESOLUTION TEXTURES
>>46278796
because of marketing and lies and inferior card
that would cost too much to fix to be able to use all 4gb
cheaper to lie and hope we don't find out
Rejoice everyone. This means ass-cheap 970s for everyone. It sure will be a good replacement for my 560ti.
>>46278836
>running a 960 for high resolutions
>still 128 bit memory bus
This is almost a worse scam.
>>46278796
Because they would have their new card look worse than the 11 month old amd cards.
>>46278849
>cheaper to lie and hope we don't find out
Well, my two last cards including the 970 has been Nvidia. Next will likely be AMD.
If only they can figure out how to do tesselation without grinding to a halt.
> PCPer discussion
> https://www.youtube.com/watch?v=b74MYv8ldXc
> we are here today to discuss the GeForce GTX 970 memory issue, which is an incredibly complex topic
>>46278874
pls amd let us love you
>>46278874
They already do for the past 3 gens
>>46278842
>>46278880
>on 980 the problem doesn't occur
yes it does
>>46278900
AMD cannot into tesselation.
>>46278180
>if you wanted 4gb all fast then you would obviously opt for the more expensive 980
Except when I bought a 970 this was not clearly the case. It was 1 month ago too so no "hurrr early adopt" bullshit.
ITT: Vidyagamefags complain about not being able to running BF4 at max twice at the same time on different 4k screens.
Fuck off fagets go back to /v/ and stop complaining already, it's already cheap as fuck.
>>46278874
>If only they can figure out how to do tesselation without grinding to a halt.
That hasn't been an issue since HD5000, and lets be honest now, initial Fermi was a much shittier product than not doing tesselation that well.
>>46278930
Yeah, these lies are acceptable because it's only a $360 card.
>nvidia lied
>the problem affects people pushing their cards to the limit, like 4k res.
>4k is useless and overpriced for at least 2 years
>you will have a new card by that time, or going to
I acknowledge this is shit and nvidia are assholes for doing this, but do we have to discuss this the whole week?
>i own a 970 myself occasionally playing linux steam games and d3 on wine
>no problems
>>46278970
>>46278970
What are you, poor?
If you can't afford an current max build, don't buy an pc? Or atleast don't complain about it.
>>46278967
Dude, the 660ti I have collecting dust beats the 290x on tesselation.
Yeah, a 128bit piece of shit beats AMDs 512bit monster.
"Good enough" for game levels doesn't deny the fact it's garbage once the levels get pumped up.
>>46278921
>>on 980 the problem doesn't occur
>yes it does
where has this been confirmed?
980 is in theory unbinned with no disabled L2/ROPs/etc. right?
>>46278930
Like we need this shit.
Is it still true the gigabyte revised version doesn't get this shit or was that a lie too?
>>46278430
I didn't even know that. I had a 660 and now I have a 970, I dun goofed twice in a row. Fuck.
>>46279010
Shills for both sides are gonna feast on this turd for a couple of months at least.
>>46279028
Same shit.
>>46279014
>if you don't do max build
>dont go pc
>>46278807
They weren't lies with the 660. The difference is this:
"Hey guys, this is a gimped card for X amount of money, this is how we designed the memory you might want to look at this and see if it still fits your needs" -660
"Hey guys, this is a great GPU for X amount of money." -970
They're not lying if they tell you that it is a gimped card, that's totally fine. The problem is that nobody knew that the 970 was a gimped card, not even reviewers.
>>46279018
I thought it was 192-bit, or at least my MSI 660ti PE is 192-bit.
Regardless, it does well for me and was hoping the 960 would be a major upgrade, but it isn't so i'm gonna hold out for another GPU generation.
I could do a better job at community steering that the M4jestic sock-puppet.
>>46279075
Well, that's nasty. No wonder about the outrage.
>>46279075
what is the 980?
>>46279081
You're right, it's 192 bit.
Doesn't excuse AMD 512 bit 2 1/2 gen later top of the range card gets an assraping from it on tesselation.
>>46279097
Working as intended with specs as written.
>>46279050
>mfw it's true
>AMD (NASDAQ) 2,63 $ +0,18 (+7,55 %)
>NVDA (NASDAQ) 20,57 $ -0,13 (-0,65 %)
Coincidence?
>>46279026
AFAIK yes. People just freak out because the nvidia "benchmark" shows both cards dropping 40-ish % in framerate, that's just because they needed to increase VRAM usage and cranked up the settings.
tl;dr: people are dumb, Nvidia are assholes, the 970 is a fucking bs GPU and the 980 is probably fine.
>>46277563
Well that ended a claim about a lie quickly.
>>46279118
That's weak. The investors know we'll take this and bend over again next gen. And ever bring our own lube.
>>46277651
The 970 should be Nvidia Green (tm).
Just for a more design impact.
>>46277862
Holy shit, that guys defends Nvidia for pages upon pages.
>>46279097
an overpriced "top tier" GPU. If you want to argue the case of "lol, should have bought a 980 if you wanted 4k" then fuck off. Nobody knew of this issue, people bought 970s because they wanted to add a second one for SLI later or they initially bought 2 970s, the whole situation is a mess.
>>46279050
>same shit
Damnit
>shills for both sides are gonna feast on this turd for a couple of months at least
May very well be the worst part. Every thread with a passing mention about graphics is gonna devolve into ENJOYING THAT 3.2? IMPLYING AMD NO DRIVERS LEL
>>46279149
He's really working for those shekels.
And the fanboys are eating the positive FUD like candy.
>>46279142
Nah, I'm going back to team red unless they fuck up spectacularly one or two years from now.
>>46279113
3rd party manufacturers can't fix shit that's embedded in the silicone.
>>46278725
>you aren't supposed to play the Way It's Meant to be Played (tm)
Why do they call it a 970?
Because when you see it, you turn 970° and walk away
>>46278099
>$170 shipping
>2 month shipping
>30 day return policy
Still not a bad deal if it's legit.
Probably half assed refurbs though.
Is there confirmation about if the updated Gigabyte 970s still have that problem?
>>46279169
Good man. You will be one of the few.
The rest are too fanboy.
>This is the reason for the bug but actually it's not a bug, it's a feature and perfectly fine, and to combat the claims of lag spikes here are some fps averages. Now shut up and buy our cards.
>>46279166
>He's really working for those shekels.
you're missing the point. Not that all consumer fanboys aren't idiots to varying degrees, but Nvidia somehow cultivates the most slavish of them all.
> he literally does it for free
>>46279158
so you pay way less for the 970 and expect 980 performance?
wat
>>46279175
Probably 8800GT in a 970 box if you ask me.
>>46279176
The problem is within the GM204, which Nvidia manufactures.
>>46279176
Gigabyte cannot fix this. They'd have to resolder a memory controller back onto the board.
All 970s are fucked.
>>46279174
the meme rises
>>46279191
Close, those are 9800GTs.
>>46279175
>Probably half assed refurbs though.
More likely several generations old card with some random fan and a bios tweak to make it look like a 970.
>>46279180
4850 > 5850 > 660ti > 970 > whatever is useful in 1-2 years.
Being a fanboy is retarded, you buy what is best performance for the buck you have available at the time.
>>46279189
>expect 980 performance
>expect full 4gb vram
>>46279191
>>46279206
>>46279208
Would probably be faster than a 970 too.
>>46278923
That benchmark is broken, the the other ones...
>>46278099
>Item Location: China
Uh . . . Why is my scam radar screaming louder than I had previously thought possible?
>>46279204
And died as quickly.
>>46279176
I just got one, how can I test if it has the issue?
>>46279189
No, you expect adequate performance no matter if the VRAM sits at 3.2GB or 3.7GB. Nobody expects the card to perform as well as a 980, people just want to use more than 3.5GB (with 2 970s in SLI, playing games at 4k, for example) without the GPU crapping itself. the last .5GB are basically useless, people got tricked into buying this PoS, this is not okay.
>>46279018
Well great, go play Crysis 3 12 more times.
The fact of the matter is that in every other metric the 290X slaps the shit out of nvidia's everything other than the 980.
>>46279225
Is it a 970? Yes? It has the issue. Nvidia thanks you for your loyalty.
>>46279217
Clearly 64x tesselation is broken, no game would ever use that level of tesselation unless paid by nvidia etc etc etc.
>>46279238
>Implying that it has an issue even though its working exactly as designed
>>46279225
If you have a 970 you have it. But you probably didn't "trigger" it, because nobody runs exclusevly benchmarks.
TL;DR you can safely ignore this
>>46279237
>>46279262
finally some sense, thanks
>>46279225
http://www.guru3d.com/news-story/does-the-geforce-gtx-970-have-a-memory-allocation-bug,11.html
link from article:
http://nl.guru3d.com/vRamBandWidthTest-guru3d.zip
You'll be happier not knowing.
>>46279014
So did they manage to cheap out somehow, or is this actually an "oops our engineers are retarded" situation? I was on the verge of ordering a GTX970 for a build I'm planning. I'm glad this news came out ahead of time.
So what do I buy now? Keep in mind it's mITX and heat may be an issue.
>>46279258
>"It's not a bug, it's a feature!"
>"Outright lying about specs isn't an issue!"
great call there nvidiot
>>46279245
No, that specific benchmark is broken as the 290x does 2x the tessellation performance of the 7970 in hardware and it is in effect in game and in the Microsoft DX tessellation benchmark.
>>46278135
man my car must be broken because when I put it straight into 5th gear and try to go it stalls too :(
>>46279282
They cheaped out and the marketing & legal departments decided it was still safe to sell as a full 4gb card.
>>46279273
well i have a 970 and i am not that sure about it. for now you can probably ignore it, but what about future stuff? don't want muh witcher to lag around
LEL AMDRONES, EVEN CRIPPLED, IT'S STILL BETTER THAN YOU SUB BINNED TRASH
>>46279295
>>46279295
>having a notork shitbox
nvidiots have such great taste in consumer products don't they
>>46278588
This architecture was premeditated scamware from the get-go.
Nvidia went out of their way to design L2/ROP/MC blocks that in theory could access the last 32b RAM module but designed the entire driver stack so that it won't unless it absolutely has to.
Maxwell is the first generation from Nvidia to allow this functionality, so it's not some magic oversight.
>>46277152
So does this mean 980's are the only option in the 900 series?
>>46279295
>implying easing gradually into the 3.5gb bug makes it magically work as a full 4gb card
>>46279322
amd = lsx
nvidia = rotary
>>46279329
Very yes.
>>46279329
if you have no need for your shekels I guess.
If you must go with the merchants, I suggest waiting until the R9 300 series comes out in a few months and encourages them to drop prices.
>>46279329
If you don't mind a magical limit of 3.5gb vram, it's still a good card. They're still fucking assholes for pulling this shit.
>>46277152
>Nvidia is the best we got amazing performance for only 30 dollars
>GPU has fatal flaws making it's performance worse than a 7970.
lol
So, when can we expect the new AMD cards again?
I really need a new one soon.
>>46279343
The people who bought 970 for SLI are completely fucked, so you forget to list that you just limited yourself to a single 970 or buy a new GPU
>>46279363
A couple of months. Maybe.
>>46279363
Q2 at the earliest.
>>46279363
Q2, sadly
this is fucking bullshit
>lied about # ROPs
>lied about memory access
>highest gpu sales in gaming history
>>46279363
May/June
>>46279423
thank you for your loyalty fanboys
>>46279375
Thats correct.
I also didn't mention the use of 3D software.
These people are the ones who are really fucked and i feel sorry for them.
I just get mad at all those shills who are treating it as an apocalypse. I don't want to defend nvidia here, i just want to settle this a bit down, because most ppl will have no problems until next-gen cards.
>>46279423
It's the same situation as with crippled online only DRM games.
Those sold like hotcakes too.
Trying to make sense of this "benchmark" (numbers are from nvidia, so, whatever)
This shows the 980 and 970 dropping in performance when going beyond 3.5 GB vram usage exactly the same (970 some 1-3% worse than the 980, so...fucking nothing)
It can't be just that 1-3%, nobody would have given a fuck or even noticed, right?
So are the "> 3.5GB" settings actually trying to use like 4.1 GB, to make the 980 show the same degradation when it runs out of memory?
Or is the 980 gimped, too (which doesn't make any sense if the 970's problem is supposed to be caused by the crappier memory bus/whatever))
>>46279423
Wow, lying about a product allows you to place it in a real price sweet spot. I'll have to tell my economics professor about this.
>>46279351
But will that make my room hot? I keep cards 3-5 years easily closer to 5. An extra hundred is worth it if I don't have to open a window in the winter.
>>46279355
Can you lock it at 3.5 somehow so I dont get stuttering or lag spikes if some shitty unoptimized game tries to go over?
Anyone heard if the 980ti will be a thing?
>>46279468
>crippled online only DRM games.
Not really, that shit is advertised.
>>46279474
It's bullshit with lipstick on to make it not look like an issue. It is an issue, but not an SKY IS FALLING issue.
Nvidia have just removed the 100+ page thread about this issue on their official forums
You can still access it via direct link
https://forums.geforce.com/default/topic/803518/geforce-900-series/gtx-970-3-5gb-vram-issue/114/
but it doesn't show up on the forum browser.
Incredible.
>>46279423
What are the odds someone is going to sue?
>people trust nvidia after housefires and woodscrews
>kids comparing the 290x anything to the monster that was the 480
toplal
>>46279530
>it's true
>>46279499
The drivers sort of soft lock it now, that's why you have to walk over corpses to force games to use more.
>>46279530
It's been removed and put back and removed again now it seems.
>>46279474
That 1-3% is an average. In game its a spike and causes stuttering.
can I just physically tear off the 0.5gb partition and make my 970 faster?
>>46279560
Average frame rates is lipstick on the dog turd that is massive spikes.
>>46279474
Those are not the benchmark numbers. Those are Nvidia PR department numbers, you can safely ignore them.
These are the benchmark numbers. They have been confirmed by Nvidia engineers to be legit, and that last 0.5GB vram is exactly 1/7 of the speed of the fast section.
They focus on average fps in the PR piece because it hides the stuttering.
>>46279474
>So are the "> 3.5GB" settings actually trying to use like 4.1 GB, to make the 980 show the same degradation when it runs out of memory?
>Or is the 980 gimped, too (which doesn't make any sense if the 970's problem is supposed to be caused by the crappier memory bus/whatever))
980 is in theory free of the 970's bullshit, but it's still debatable how much real games are bandwidth limited across all 3.5-4 GiB workloads or whether the benchmarks are bullshit.
You could in theory still have shader-limited tests that still touch ~4 GiB memory, or you could have the driver dancing around keep frequently used textures or whatever below the 3.5 GiB death line.
The bigger issue is that the ROP count, L2 size, and memory bus width have all effectively been lies for the 970.
>>46279538
>People trust nVidia after the 8400/8600 packaging defects
>Nvidia = apple
>Amd = android
