So Nvidia lied about the GTX 970 for months

[sinty]

Here is a transcript of my convo with Nvidia

Status: Connected
Kenny (Listening)
Kenny: Hi, my name is Kenny. How may I help you?
Michael: Hello. I purchased a GTX 970 from Newegg on release day, I was just made aware of some problems with the GTX 970 series. Apparently,the specs aren't really what Nvidia said they were or something similar. There is talk of a recall on all 970's and I was told to contact Nvidia first before contacting Newegg about what you think of the situation. Not sure if I am supposed to return it or what the course of action should be for this.
Kenny: Please be informed that there isn't any recall for this model, It's just that there is a difference in the architecture of the GPU
Kenny: You can see the technical explanation from here : http://www.pcper.com/reviews/Graphics-Cards/NVIDIA-Discloses-Full-Memory-Structure-and-Limitations-GTX-970
Michael: Well, whatever the case is, many customers are very upset. I don't have any problems with the card, but if the card wasn't what it originally said it was, I would suspect that I should try to return it. How is Nvidia handling the situation for anyone who wants to return the card.
Kenny: There is an official statement from NVIDIA for this concern. Please visit our forum : https://forums.geforce.com/default/topic/803518/geforce-900-series/gtx-970-3-5gb-vram-issue/post/4432672/#4432672
Kenny: Don't worry, this card still have all the same specs
Michael: I've read the statement, which lead me here to ask about it further.
Kenny: OK
Kenny: It's just that the memory is in two segments 3.5 and .5 GB
Kenny: Which doesn't mean that the card has less specs
Michael: Due to so many customers being upset over it, I am not sure how to proceed. My question is will Nvidia accept a return or refund if I had purchased the card from Newegg on release day. I did not purchase extra warranty through Newegg though.
Kenny: I understand that you are dissatisfied with the card. Unfortunately NVIDIA can’t directly offer refunds on third party products. You need to contact to discuss refunds or exchanges
Michael: Contact who to discuss refunds, Do you mean Newegg?
Kenny: yes, if you think that the card has issues you can
Michael: Also, There is a big logo with an Nvidia sticker on my GTX 970. I thought this was an Nvidia product. I am quite shocked the Nividia GTX 970 gpu isn't an Nvidia product.
Kenny: The chipset is manufactured by NVIDIA
Michael : Seems like deceptive marketing to me. But, I won't argue further. Thank you for your help. I'll be sticking with AMD gpu's if this is how Nvidia works.
Kenny: If you have any further questions or comments regarding this matter, please feel free to write back to us.

 

[sinty]

Definitely not Neweggs fault, and if nothing is WRONG with the card per say I wouldn't want to make Newegg pay for that. But, the card itself was not marketing truthfully it seems and Nvidia should be the one who handles it. But, I am SHOCKED that Nvidia doesn't consider its own products actually their own. Nvidia products are 3rd Party products to Nvidia. That is insane.

Deceptive marketing 100%.

 

[sla70r]

100 percent Nvidia's fault. Wont stand by their own product..... yikes.

 

[junkeymonkey]

''Definitely not Neweggs fault, and if nothing is WRONG with the card per say''

well bad part is there the middle man and the one to deal with for refund /exchange on another card

or go tto the manufacture and deal with them and maybe at best just get a swap out card unless its evga and use there step up program to get a 980 if that's what he wanted to do

in the end you did not get what you paid for as far as this memory thing kinda goes [???] you bought on advertised specs not wrong specs .. right ?

 

[sinty]

Absolutely, I'm with ya. Newegg won't care. I am sure they will cite their policy and that because I've no warranty they can pretty much pee on me and I can't say anything about. Nvidia just proved they won't do anything. Im not at all complaining, as mentioned the card functions perfectly fine and is super powerful, overkill for what I want and need but I'm not cool with being mislead. This IS an Nvidia problem regardless of what they've said. They made it, they misadvertised it, therefore they should do the right thing and accept all returns no matter when you purchased the 970.

Question is what do they replace it with even if they did exuded good business ethics and accepted a replacement or w/e. Right now, this is bad business ethnics in play.

 

[junkeymonkey]

a thread over at g-force forums guys claim tiger and amazon it taking them back ??

https://forums.geforce.com/default/topic/806936/geforce-900-series/contact-your-retailers-some-are-making-exceptions-for-970-returns/

 

[youcanDUit]

Everyone - knock off the fighting or this thread will be closed and a few timeouts may be issued. Conjunctive criticism of a company or product is OK, but the personal attacks of GTX 970 owners vs. Radeon R9-290X owners is not, otherwise this will be viewed as just another NVIDIA vs. AMD thread.

people are fighting? meh. i bought the 970. stupid stupid move. my gts 450 crapped out on me after 5+ years. had i known i was going to be lied to i would've just kept my 450.

 

[youcanDUit]

Please sign it if you purchased a GTX 970.

https://www.change.org/p/nvidia-refund-for-gtx-970?after_sign_exp=member_sponsored_upsells#share

 

[vmem]

Definitely not Neweggs fault, and if nothing is WRONG with the card per say I wouldn't want to make Newegg pay for that. But, the card itself was not marketing truthfully it seems and Nvidia should be the one who handles it. But, I am SHOCKED that Nvidia doesn't consider its own products actually their own. Nvidia products are 3rd Party products to Nvidia. That is insane.

Deceptive marketing 100%.

yeah, that's scary. I purchased a 970 from Newegg as well. luckily I did purchase a warranty so technically I can return it had I not installed an EK waterblock over it. and I can't exactly return the waterblock...

That aside, I think all of this raises an interesting problem for Nvidia: vendors like Amazon and tiger are taking 970s back, so as a competitor, I wouldn't be too surprised if newegg starts doing so too (unless they want to lose loyal customers). All of these guys will then turn these "faulty" cards back to manufacturers such as ASUS, Gigabyte, MSI etc, who would in turn probably try to get a refund from Nvidia for faulty marketing.

Now if Nvidia keeps trying to sweep this under the rug and deny claims from either the major channel vendors or their board manufacturing partners for a product as popular as the 970, this could result in a nightmare for them in the future. (at minimum manufacturers would test Nvidia's products more extensively, maybe requesting engineering samples earlier, etc.) or vendors could decide to keep fewer boards in stock. But if Nvidia does take back the cards it could take a decent sized chunk out of their next quarterly earnings report (face it, the 970 was pretty popular over the holidays and was sold out everywhere)

 

[sinty]

There is a hell of a great debate on page two of junkeys link. If anyone is denied a refund, file a BBB complaint. A few hundred of those would soil Nvidias rating and they would be absolutely forced to fix the situation immediately. Nothing else will solve it, complaining anywhere else gets those upset nowhere at all. The BBB is the only place that companies actually care about. The petition wont do anything, but if you could get the same amount of users to file a BBB complaint, that would fix the issue within hours of Nividia execs finding out that THOUSANDS of complaints were filed against them through BBB's website.

I would say be nice and don't push the issue, but Nvidia has given the middle finger to everyone and specifically said too bad so sad for you. For that, I say drill them into the ground with complaints and everyone who owns a GTX 970 should file a complaint with the BBB against them. Awful business ethics happening now, and again if they were nice about it and said so sorry, we don't know what to do or how to help but will try...I'd be fine with it. But they took the opposite course, ignored and gave everyone the finger.

 

[junkeymonkey]

you know out side of this issue just look at all the 970 threads around -- fans don't work black-screen, coil whine , slow boot . no boot . and on and on .

I spoke out about this card way back like in sep./nov- and when I brought things up or what I thought were concerns all I got was how stupid I was and bashed or I was crazy ..

funny thing is how it was rushed out to make the x-mas rush that was a main clue to me over all then I read stuff like this from review sites

'' Furthermore, as we mentioned in our GTX 980 review, GTX 970 has been a pure virtual (no reference card) launch, which means all of NVIDIA’s partners are launching their custom cards right out of the gate. A lot of these have been recycled or otherwise only slightly modified GTX 700/600 series designs ''

not to say that the cards you got are that way now but that is just how sad it was to me like I buy a new card that's made of salvage / hand me down parts plus the cut back gpu that did not cut it for the real 980 cards ??

all along the writing was on the wall and some reading between the lines had me at buyer beware for the get go...

I had one in the cart and something kept telling me , don't do it and wait and see... well looks like it was a good thing I headed that little voice warning and removed from cart ...

 

[chenw]

has there been any X70 cards that were NOT made that way?

 

[vmem]

you know out side of this issue just look at all the 970 threads around -- fans don't work black-screen, coil whine , slow boot . no boot . and on and on .

I spoke out about this card way back like in sep./nov- and when I brought things up or what I thought were concerns all I got was how stupid I was and bashed or I was crazy ..

funny thing is how it was rushed out to make the x-mas rush that was a main clue to me over all then I read stuff like this from review sites

'' Furthermore, as we mentioned in our GTX 980 review, GTX 970 has been a pure virtual (no reference card) launch, which means all of NVIDIA’s partners are launching their custom cards right out of the gate. A lot of these have been recycled or otherwise only slightly modified GTX 700/600 series designs ''

not to say that the cards you got are that way now but that is just how sad it was to me like I buy a new card that's made of salvage / hand me down parts plus the cut back gpu that did not cut it for the real 980 cards ??

all along the writing was on the wall and some reading between the lines had me at buyer beware for the get go...

I had one in the cart and something kept telling me , don't do it and wait and see... well looks like it was a good thing I headed that little voice warning and removed from cart ...

This has been true of all the #2 and often #3 GPUs in the performance stack from both Nvidia and AMD though. looking back, the R9 290 is a cut down version of the R9 290x, and the 780 is kind of a crippled titan (or 780ti depending on how you see it) etc etc. and it is common practice to use PCBs for past products for the non-flagship GPU. manufacturers have been doing that for decades.

Look, I'm a fan of neither camp, and I recently switched from AMD (7970) to Nvidia for an upgrade mainly because AMD has been so slow to come out with their new GPUs. I'm not sure what your point is, that the gtx 970 is not the flagship card so people shouldn't buy lower tier products of those of us who do should expect less from them? I'd like to know what car you're driving 😛

 

[itzhalo]

Lets see what this driver has in store but if nothing changes, I'm not keeping this in my PC.

 

[junkeymonkey]

ya, I know but not like this 900 thing so I got a 7850 thing is it works as advertised right out of the box can the 970 so far say that ???

 

[getdamafiaonyou]

People like me are the ones who feel truly shafted, I bought 2 for my triple monitor setup. Was hugely disappointed in the performance I was getting, and ended up selling thing (I didn't purchase them at the same time so the 30 day newegg policy wouldn't have worked for me) All in all I took about a $200 hit on these cards. They, even a single one, worked wonderfully at 1080p, but at higher resolutions I wasn't getting what I deemed acceptable. So I sold them and moved on, then this information comes out and I feel I should be due my difference, but we all know that's not going to happen. I'm not fanboy one way or the other, but Nvidia's handling of this situation has certainly turned me off of them for a pretty good while.

 

[junkeymonkey]

also it seems to me if they want to do a allocated memory system then it should have its own memory chip for that and not ''rob'' from the memory dedicated to the graphics ?? if that makes any sense... so is it like its a 1 +1+1+ 1/2 of a full 1gb chip ?? seems like this is a your a beta tester for something they got planed for down the road for a future card to see how well it works out ???-- has any card done this with its memory before ??

 

[vmem]

@ funkeymonkey

if you look at images of the naked gtx 970 PCB, you'll notice that there are 8 memory chips, presumably each is 0.5GB. apparently the bandwidth for the last 0.5GB chip on the 970 is just gimped due to the way the chip was made 🙁

 

[junkeymonkey]

I wish all these thoughts came to mine in one post -- sorry

and to add to that look at this with the 960 cards with the 128 bus '' Don't be quick to write off the 128-bit memory bus width just yet as NVIDIA is backing it with a new lossless texture compression technology that reduces memory bandwidth usage, effectively making it a wider memory bus than it physically is.''

ok well cool but why don't they put this on theses 980/970 card to enhance the buss on them to make that even better on them also ???

or is it that you test this stuff all out for them now then in 6 months the real gtx 900 card show up with all this stuff working on them 100% with out issues as you see now then your looking to re buy one more time ?? you know the bigger better gtx 900 [ ti ]

 

[junkeymonkey]

@peeman
it was just for example as said you paied for a full 4 gb card not a 3.5 +.5 at a lesser rate -- sorry

 

[palladin9479]

also it seems to me if they want to do a allocated memory system then it should have its own memory chip for that and not ''rob'' from the memory dedicated to the graphics ?? if that makes any sense... so is it like its a 1 +1+1+ 1/2 of a full 1gb chip ?? seems like this is a your a beta tester for something they got planed for down the road for a future card to see how well it works out ???-- has any card done this with its memory before ??

How the heck are you not reading? Toms just did a break down of it, Anands did a really good technical breakdown describing exactly why everything is the way it is. And still you put out false and misleading information. I'm going to spell it out to you now so that you can't use ignorance as a defense anymore.

The GM204 chip has eight 32-bit GDDR5 memory channels. Each memory channel is connected to a ROP unit that contains 256KB of L2 cache and eight individual ROPs. Previously if there was a defect in one of those eight ROP's or in the associated L2 cache memory, the entire component would have to be disabled which also disables the associated 32-bit GDDR5 memory channel. Maxwell has the ability to selectively disable individual subcomponents, so instead nVidia only disables the L2 cache and connected eight ROP's but leaves the memory channel operational. Now since there is no cache or connection to the crossbar, that memory channel is then slaved to a nearby fully operational unit. Memory data is typically interleaved across all channels in 1KB blocks, so if you needed to read 4KB of data it would exist on four different 32-bit GDDR5 chips and would be read at the speed of x4 of a single chip. In the case of a fully operational GM204 chip, the 980, the entire eight 32-bit GDDR5 channels are used for an eight way interleave. For binned GM204 chips, the 970, only the fully independently operational seven channels are used for a seven way interleave. That last slaved 32-bit GDDR5 chip is kept separate as to not lower the performance of the rest and instead used as a victim cache since it's still faster then system memory. That is how and why you got your 3.5GB / 512MB segments, and why putting them all into a single 4GB eight way interleave segment would be disastrous for performance (two of the eight interleaves couldn't talk at the same time so every stripped access would have double the latency).

So ultimately what you are complaining about is why didn't nVidia give you a fully operational 980 when you bought a binned 970. Yes there was incorrect marketing material, and that is something that nVidia is going to have to deal with internally, I expect some middle manager is going to be asked to fall on his sword for this.

Now anymore deliberate spreading of incorrect technical information will be met with a vacation. You can have whatever personal opinion you want for whatever reason you want, just don't go around telling others lies.

 

[getdamafiaonyou]

also it seems to me if they want to do a allocated memory system then it should have its own memory chip for that and not ''rob'' from the memory dedicated to the graphics ?? if that makes any sense... so is it like its a 1 +1+1+ 1/2 of a full 1gb chip ?? seems like this is a your a beta tester for something they got planed for down the road for a future card to see how well it works out ???-- has any card done this with its memory before ??

How the heck are you not reading? Toms just did a break down of it, Anands did a really good technical breakdown describing exactly why everything is the way it is. And still you put out false and misleading information. I'm going to spell it out to you now so that you can't use ignorance as a defense anymore.

The GM204 chip has eight 32-bit GDDR5 memory channels. Each memory channel is connected to a ROP unit that contains 256KB of L2 cache and eight individual ROPs. Previously if there was a defect in one of those eight ROP's or in the associated L2 cache memory, the entire component would have to be disabled which also disables the associated 32-bit GDDR5 memory channel. Maxwell has the ability to selectively disable individual subcomponents, so instead nVidia only disables the L2 cache and connected eight ROP's but leaves the memory channel operational. Now since there is no cache or connection to the crossbar, that memory channel is then slaved to a nearby fully operational unit. Memory data is typically interleaved across all channels in 1KB blocks, so if you needed to read 4KB of data it would exist on four different 32-bit GDDR5 chips and would be read at the speed of x4 of a single chip. In the case of a fully operational GM204 chip, the 980, the entire eight 32-bit GDDR5 channels are used for an eight way interleave. For binned GM204 chips, the 970, only the fully independently operational seven channels are used for a seven way interleave. That last slaved 32-bit GDDR5 chip is kept separate as to not lower the performance of the rest and instead used as a victim cache since it's still faster then system memory. That is how and why you got your 3.5GB / 512MB segments, and why putting them all into a single 4GB eight way interleave segment would be disastrous for performance (two of the eight interleaves couldn't talk at the same time so every stripped access would have double the latency).

So ultimately what you are complaining about is why didn't nVidia give you a fully operational 980 when you bought a binned 970. Yes there was incorrect marketing material, and that is something that nVidia is going to have to deal with internally, I expect some middle manager is going to be asked to fall on his sword for this.

Now anymore deliberate spreading of incorrect technical information will be met with a vacation. You can have whatever personal opinion you want for whatever reason you want, just don't go around telling others lies.

Sorry to go on a side tangent here, but does that mean that when parts of the new maxwell series of cards go bad, they'll basically just under perform and not actually stop working?

 

[palladin9479]

also it seems to me if they want to do a allocated memory system then it should have its own memory chip for that and not ''rob'' from the memory dedicated to the graphics ?? if that makes any sense... so is it like its a 1 +1+1+ 1/2 of a full 1gb chip ?? seems like this is a your a beta tester for something they got planed for down the road for a future card to see how well it works out ???-- has any card done this with its memory before ??

How the heck are you not reading? Toms just did a break down of it, Anands did a really good technical breakdown describing exactly why everything is the way it is. And still you put out false and misleading information. I'm going to spell it out to you now so that you can't use ignorance as a defense anymore.

The GM204 chip has eight 32-bit GDDR5 memory channels. Each memory channel is connected to a ROP unit that contains 256KB of L2 cache and eight individual ROPs. Previously if there was a defect in one of those eight ROP's or in the associated L2 cache memory, the entire component would have to be disabled which also disables the associated 32-bit GDDR5 memory channel. Maxwell has the ability to selectively disable individual subcomponents, so instead nVidia only disables the L2 cache and connected eight ROP's but leaves the memory channel operational. Now since there is no cache or connection to the crossbar, that memory channel is then slaved to a nearby fully operational unit. Memory data is typically interleaved across all channels in 1KB blocks, so if you needed to read 4KB of data it would exist on four different 32-bit GDDR5 chips and would be read at the speed of x4 of a single chip. In the case of a fully operational GM204 chip, the 980, the entire eight 32-bit GDDR5 channels are used for an eight way interleave. For binned GM204 chips, the 970, only the fully independently operational seven channels are used for a seven way interleave. That last slaved 32-bit GDDR5 chip is kept separate as to not lower the performance of the rest and instead used as a victim cache since it's still faster then system memory. That is how and why you got your 3.5GB / 512MB segments, and why putting them all into a single 4GB eight way interleave segment would be disastrous for performance (two of the eight interleaves couldn't talk at the same time so every stripped access would have double the latency).

So ultimately what you are complaining about is why didn't nVidia give you a fully operational 980 when you bought a binned 970. Yes there was incorrect marketing material, and that is something that nVidia is going to have to deal with internally, I expect some middle manager is going to be asked to fall on his sword for this.

Now anymore deliberate spreading of incorrect technical information will be met with a vacation. You can have whatever personal opinion you want for whatever reason you want, just don't go around telling others lies.

Sorry to go on a side tangent here, but does that mean that when parts of the new maxwell series of cards go bad, they'll basically just under perform and not actually stop working?

Depends on the model and design decisions by the engineers. This all happens during the final QA phase of production, they test each chip and if it's perfect it's labeled "GTX 980", but if up to three SMM's and / or one ROP cluster are bad, then those are individually disabled and it's labeled "GTX 970". The 960 is a GM206 chip and made on a different assembly line. I could see a GTX 950 being made from GM206 with one or two SMM units disabled, but going any lower then four 32-bit GDDR5 channels would be really bad. That gets into the territory of a potential GTX 940. It also opens up the future possibility of a 960TI being a GM204 chip with two GDDR5 channels and four to six SMM's disabled. That is how products are placed, designing a die is expensive as f*ck, producing it even more expensive as you need an entire tool chain and production unit for it. If you can get multiple products to use the exact same die, then there is a huge cost savings as you don't have to throw away chips with only minor defects and you can appeal to another market segment without incurring the expense of a new design process.

 

[junkeymonkey]

look at what I said above -- you got the 980 a full working card -- you got the 970 with that alacated memory thing and then the 960 with the enhanced buss thing -- now they know the 980 works as expected but they pulled this 970 memory thing to see how you react to it and how well it works and now getting you feed back on that deal, and now will make adjustments for it [so they say as stated above ] then the 960 with the buss enhancement that's only on the 960 card and not the rest so them guys can give the feed back separately .. so as all the feed back comes in from you , and they work out all the bugs and issues that is on a individual card will be implemented on one new better card like for when they release the 980TI it will have the fully working allocated memory and the buss enhancement and will be the next kick ass card from NVidia [opinion]

 

[17seconds]

Well, at least the GTX 970's in retail have the exact same performance as those in the reviews. The advertised versus actual performance shouldn't really be in question, amateur YouTube videos notwithstanding. The only real complaint that I can see, then, would be the misleading architectural design notes. That is worthy of derision.

Now anyone thinking they'll just go and jump over to AMD, would probably have forgotten that AMD has a lengthy history of actually selling products that do not perform as advertised in reviews.

When the 7990 came out, Tom's Hardware discovered that there was a significant difference in the performance of the samples sent to reviewers and those cards available in retail to the consumer.
http://www.tomshardware.com/reviews/radeon-hd-7990-crossfire-overheat,3539-2.html

Similar findings were made about the R9-290, those sold in retail just weren't as fast and consistent as those sent to reviewers.
http://www.tomshardware.com/reviews/radeon-r9-290-review-benchmark,3659-2.html


The point is, the level of outrage doesn't seem to match up. In comparison to those debacles, the current outrage tends to seem a bit overreacted, justified to an extent, but maybe a bit much. In fact, I don't recall anyone starting a change.org petition or suggesting they should get their money back for buying a R9-290 or 7990 with performance claims that were proven to be false. For the GTX 970, I have yet to see a legitimate reviewer who has been able to show any serious significant impact on performance in those very rare instances of high VRAM usage. According to reviewers, your games perform exactly as advertised, and none have changed their recommendations. The same can't always be said about some of the alternatives.
http://www.pcper.com/reviews/Graphics-Cards/Frame-Rating-Looking-GTX-970-Memory-Performance/COD-Advanced-Warfare-and-Clos
http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/68595-gtx-970s-memory-explained-tested-2.html
http://www.guru3d.com/news-story/middle-earth-shadow-of-mordor-geforce-gtx-970-vram-stress-test.html