KitGuru KitGuru.net – Tech News | Hardware News | Hardware Reviews | IOS | Mobile | Gaming | Graphics Cards
While only a small fraction of GeForce GTX 970 owners plan to return their graphics cards to stores following a scandal with the product's specs, there are still quite a lot of people, who got upset not only because the company originally published incorrect specifications, but because it decided to remain silent even after the mistake was uncovered. Apparently, those people have decided to take Nvidia Corp. to court because of it.
Andrew Ostrowski, individually and on behalf of all others similarly situated, this week filled a class-action suit against Nvidia in the U.S. district court for the northern district of California. The complaint accuses Nvidia and Gigabyte Technology of misleading advertising, unfair business practices, unlawful business practices, and deceptive business practices. The lawsuit seeks a jury trial as well as disgorgement, restitution, injunctive relief and all the other damages and reliefs permitted under California law.
Back in January it was discovered that Nvidia incorrectly declared the amount of raster operations pipelines (ROPs), actual memory bandwidth, capacity of L2 cache and the amount of high-speed onboard memory for its GeForce GTX 970 graphics adapter. Instead of 64 ROPs, the GPU features only 56 ROPs; actual usable memory bandwidth at present is less than 224GB/s; the L2 size is 1792KB, not 2048KB; and the amount of memory should be indicated as 3.5GB + 0.5GB because only 3.5GB can be accessed with maximum data-rate due to limitations of the cut-down GM204 architecture of the GeForce GTX 970.
“The defendants engaged in a scheme to mislead consumers nationwide about the characteristics, qualities and benefits of the GTX 970 by stating that the GTX 970 provides a true 4GB of VRAM, 64 ROPs, and 2048 KB of L2 cache capacity, when in fact it does not,” the lawsuit states. “Defendants’ marketing of the GTX 970 was intended to and did create the perception among purchasers that the product was, in fact, able to conform with the specifications as advertised.”
Nvidia has admitted that it incorrectly stated specifications of the GeForce GTX 970. The company said that all performance limitations associated with the specs cannot be cured with a driver update. However, the GPU developer did not promise any compensation to owners of such graphics cards. Moreover, Nvidia unofficially wants to distance itself from the scandal, according to a media report. By contrast, numerous retailers and graphics cards makers accept returns of the GeForce GTX 970 graphics adapters or provide partial refunds to their owners. Only between 1 and 5 per cent of owners have so far returned their GTX 970 add-in-boards.
Since the lawsuit is a class action one, it may be joined by other people, who believe that Nvidia and Gigabyte deceived them with the GeForce GTX 970 specifications.
Nvidia declined to comment on the lawsuit, reports PCWorld web-site. Gigabyte did not comment on the news-story.
Discuss on our Facebook page, HERE.
KitGuru Says: Usually proceedings like this one last for years, hence, do not expect any results any time soon. In fact, one of the reasons why the lawsuit was filled was the fact that Nvidia decided admit its mistakes, but offered nothing back to gamers. Keeping in mind that mobile Nvidia GeForce GTX 980M graphics processor is also affected by the same memory and ROP issues as the GeForce GTX 970, expect notebook gamers to slam Nvidia too…
yeh NVidia are not being honest and not sorting it out like they should but ive watched a lot of 4k reviews and all trustworthy reviewers that’s about 8-10 people all tried to get stuttering and they couldn’t in game except mordor by running completely ultra on the 970 which I wouldn’t do anyway cause its like 20fps lol. And that’s only 1 person who managed to viably show it, so its not a normal situation. All my games tomb raider 2013 metro, bf4, gta4, wolfenstien, farcray 2,3,4,blood something and max Payne 3 all run on ultra hd mode while I have my xbmc running on my 40 in hd tv and have social and overclocking stuff on a 3rd hd screen and still im getting (without v sync) 100-160 fps oc mostly with crisis 3 at 60fps SMOOTH and tomb raider with the hair on at 80-111 fps. Judging by the reviews I have seen this card also handles the 1440 resolution smoothly without that stuttering issue.
Tell me apart from the NVidia lie where does that make this .5 ram make any difference to me or my gaming? Id really like to know. PS when overclocking I have reached 4096 gig in benchmark tests and all I see is a increase of 2-4 fps in the benchmark lol no stuttering.
I understand people upset by betray I feel same and I had a choice to send the card back, but truth is there’s nothing out there to replace it for hd gaming, the x290x is close but it doesn’t quite do it for me. Turn down the power requirements and the noise then you’ve sold me. To all the moral people out there. Ask youself one question. Have you ever pirated a pc game before lol cause thats moral isn’t it.
Question: If the 970 came with all the ROPs and extra L2, would it cost the same amount? No way. There is no way the 970 is underperforming at the price they paid for it.
What Nvidia wants to do is make a 970Ti – a 980 with some cores disabled but full speed RAM in 1 partition. Offer that to the 970 owners as a replacement.
Then get returned 970s & re-brand them as 960Ti with the proper specs completely disable the last 512MB & resell it for $250. The last 512MB is slower then some system memory it’s of little to no benefit for any game that would use 3.5GB+ so it may as well be disabled.
–
Because the 970Ti would still be cut down abit it protects the 980s position.
The 970 owners get a better GPU for the money so they should be happy again.
People who want a still really good card can pick up the reworked 960TI for $250 which would be great. Giving Nvidia a good GPU at a price point that has more customers.
It’s win, win =)
If it runs at 22GB/s then it’s no faster than DDR3, just putting that there.
*sigh*
Just to be correct, that’s all. You carry on. I agree with you.
Yeah, and that’s the problem, if Nvidia would’ve just left the card alone and made it a 3.5gb card or even a 3gb card things would be fine. But adding in that 500mb super slow chunk and seperating it from the rest creates lag and mucks things up. And also, install the high res texture packs and some mods and you’ll start seeing stuttering in shadow of mordor, if not then you’re just blind to stuttering; which is honestly a good thing. Hell i wish i couldn’t see it.
4096 soz and yeh that actually is the amount you’ll see if your using all 4gig. And yeh when running and overclocking on certain benchmark’s I do see that 4096 its exactly 4gig so I don’t get why you think a 4 gig wouldn’t read as a 4gig chip?
for all those saying you cant hit 3.5 or higher on 1080p is wrong go turn dying light on with hbao+ 1/3 view distance medium shadows and boom u just hit 3.7gb vram usage…….i have have 2 970s……….and they do stutter …….but i got what i paid for ……….lower shadow res crank everything else up its all fine
The final 500meg is widely confirmed to be unusable by the consumer……..maybe you have a 980 or a different card.
I don’t mean to sound mean dude but you definatly have it wrong here some google search will prove that. The 970 all 970s have a 3.5gig ram at 196(or close to number, again im being lazy.) speed and a small.5gig ram at the slower I think 28 speed. To get the card to go over 3 gig ram is hard, but you can do this in some 4k games and also benchmarks. This clearly proves that the 970 does use the extra .5 gig ram ( the whole big issue is when using that .5 ram in 4k it causes stuttering.) when needed so that’s the 4 gig and the actual measurement of 4gig is 4096. I know this cause someone mentioned it last week lol. The fact the .5 gig runs really slowly compared with the .3.5gig is what people have a issue with and not the amount of memory on the card. What I have said here is true baring numbers being incorrect. I have used the extra .5 myself cause I have multiple measuring software up I can see whats being used and I can also state that , and only in benchmarking not real world, I never saw a stutter or a performance decrease when I was using the extra.5 ram. I spent about 20-30 hours testing cause im that sad lol. And when you say widely confirmed well I have seen my machine use it, I have seen other reviewers on you tube use it and well NVidia said it would be used when needed. Most people in the first week started shouting that the card didn’t have 4gig,, that’s was incorrect, that’s people jumping the gun before understanding the real issue. The card has a physical 4096 memory that’s plain fact and it does use it when needed that’s plain fact too. It doesn’t work in the way people were led to believe, This issue is not being replicated by many people. I don’t play in 4k and even if I did I wouldn’t be using my 970 on ultra settings cause even with a extra 2 gig of ram id still be playing with 20-30 fps and to me anything below 50fps is useless and unplayable. Again that’s only my standard, we are all different for sure. I hope you don’t take this negatively lol, I do sometimes get things wrong myself hell we all do.
Good. May Jen-Hsun Huang die in a fire for his lies. Micropenis faggot.
The card runs just fine using all 4GB of VRAM.
Reads from any physical location are the same speed(peak 28GB/s). There is no “slow” memory. You are talking about virtual memory, not that you know the difference. This gets into how GPU’s function. Their memory bandwidth and bus is a aggregate of reads from many physical locations. Due to the 970 having only 13 SMMs it can only output to 52 of the 56 ROPs. This is one reason for having 2 virtual pools. It actually improves performance by allowing output operations to complete faster.
28GB/s was common from videocards that are over 10 years old. I understand the speed of the .5GB perfectly, and that is the issue. it is too slow for modern demanding games. With most game engines (e.g., try working with unreal or the source engine. They focus on memory usage, and warning of excessive draw calls (especially the source engine). There is no warning or focus on memory bandwidth. It is always assumed that the GPU is getting all of the memory bandwidth that it needs, and thus not much of a concern for the game developer.
When the memory performance suddenly takes a 10 year step back in time, things slow down. The issue is not as big now, but if there is a game that is designed around fully using the 4GB of memory, then it will suffer, as in those cases, there will almost always be something in the scene that is stored on that last chunk of memory. (high end videocards from 2003, were pushing around 30GB/s on their VRAM)
Virtual memory does not take place on a videocard. basically its final level of caching, is system memory, which will perform very slowly as while gaming it is being heavily utilized by the game its self as well as other background applications. Furthermore, the PCI express bus will also be heavily utilized by the system doing its draw calls.
In its simplest form, game developers are not designing high end games that need such a powerful GPU, to also be able to work smoothly with slow video memory.
The key to the slowdown is determining which memory is actively being used. if the last chunk gets used for secondary caching (e.g., storing info on a random house you entered in the game, or effects for some QTE, then it will likely not impact your performance in much of the game. On the other hand,if the game actively using both chunks, then slowdowns will happen.
All of the memory chips run at 28GB/s. There are 8 of the chips and reads from any of the chips is 28GB/s.
Actual analogy would be a car advertised as a V8 and people flipping out when somebody pointed out that all cylinders don’t fire at the exact same time. They then say it is only a V2 and go to court.
Is a V8 really a V8?
You are incorrect.
All of the VRAM chips on the card are 28GB/s They are all read from at the same speed. The memory is only available as virtual memory just like your system memory.
As for the ROPs the 13 SMMs can only output to 52 ROPs anyways.
GTX 970 reference is 145 Watts TDP.
It’s the same chip, but 3 SMMs are cut on 970. Some of the chips had 3 broken SMMs and they can use those chips. Some chips have a broken L2 cache and they can use those chips after cutting the crossbar.
Every 200 series non-X AMD card is defective and was cut or disabled in the BIOS. That is a 290 is a defective 290x.
They are not defective as in opps we will fix that.. They are defective in that they saved them from the trash by disabling broken parts.
Many chips have good parts disabled so the match.
I have no idea which comment you read before replying.
Your comment Razor512 and yes I’m slow on response.
Texture input hasn’t been and issue with 2 pools of VRAM since it started with some 500 series cards. Output is faster this war due to the limited number of ROPs that can be outputted to.