Slashdot Mirror
Nvidia Faces Suit Over GTX970 Performance Claims
According to this story at PC World, Nvidia was hit with a class action lawsuit Thursday that claims it misled customers about the capabilities of the GTX 970, which was released in September.
Nvidia markets the chip as having 4GB of performance-boosting video RAM, but some users have complained the chip falters after using 3.5GB of that allocation.
The lawsuit says the remaining half gigabyte runs 80 percent slower than it's supposed to. That can cause images to stutter on a high resolution screen and some games to perform poorly, the suit says.
It was filed in the U.S. District Court for Northern California and names as defendants Nvidia and Giga-Byte Technology, which sells the GTX 970 in graphics cards.
Nvidia declined to comment on the lawsuit Friday and Giga-Byte couldn't immediately be reached.
This doesn't look very different from when Seagate was taken to court over mislabeling hard-drives sizes, using 1000000 bytes for a MB instead of the commonly used 1048576 bytes for a MB.
I fully expect they will lose this, lose some PR metric, and start to implement the age old skill of asterisks on packages and adverts.
Memory performance between the two segments (3.5 + 0.5 GB) of memory works in an XOR manner so that accessing the slow segment prevents access to the 3.5 GB segment. Also, the whole memory access issue is a distraction from the fact that Nvidia originally advertised that the 970 had 64 ROPs (when it really has 56) and that it has 2 MB of L2 cache (when it really has 1.75 MB).
First World Problems.
I think they will find it real difficult to prove that Nvidia was intentionally (Ntentionally?) misleading people with the advertisements. Nvidia's response and explanation for what happened seemed pretty detailed and made sense to me. These kind of lawsuits actually piss me off a little anyway, because the lawyers are the only ones who really benefit. Even if Nvidia is made to compensate people who purchased the card, it's unlikely going to amount to any more than a few dollars for each person. Except the lawyers, who will get their huge fees regardless.
Do the people who file these things actually think they are somehow taking the company to task or making the world a better place? They certainly can't be doing it for actual money or greed.
When the firm representing the class bills up a few million, the defendant agrees to paying the fees and to mail all class members a $5 discount coupon or some useless download.
09 F9 11 02 9D 74 E3 5B - D8 41 56 C5 63 56 88 C0 45 5F E1 04 22 CA 29 C4 93 3F 95 05 2B 79 2A B2
...and I say that as an owner of 2x970s. Every benchmark you see is as advertised. It actually has 4GB of RAM that can be reached at the advertised speed, just not all 4GB of it simultaneously. Nobody's come up with a "smoking gun" benchmark where framerates tank for gaming. The 780 Ti with 3GB RAM beats it in pretty much benchmarks - even at 3840x2160 in SLI - so it seems that the last 512MB don't make much of a difference at all, at least not in today's games. They'd better find some compelling examples of actual harm, because I still haven't seen it. I might be biased though, since I'm kinda hoping there won't be.
What is certain though is that nVidia screwed up big, because this really would have been a footnote if they'd just informed about it. It would have been known as a 4GB card that's really 3.5GB-ish. When I bought it I thought it had the same memory subsystem as the GTX980, like two GTX970 in SLI with 2x13 = 26/16ths the shaders will always perform better. Now that might not be true in a 3.5-4GB scenario but it's a maybe kinda thing. I've long since learned that you buy computers for what you want today, tomorrow.... maybe something entirely new comes around and you want to replace it anyway. Not that I see these two being out of date for a while, seriously kicking ass at 2x145W GPU + 88W CPU it's ~500W ass-kicking system.
Live today, because you never know what tomorrow brings
It's only confusing if you bend over backwards to make it confusing. Given speed X, the last 0.5 Gig runs at X - (X*.8) That is, speed minus 80% of speed or as one might say in English, 80% slower.
So yes, in this case it runs at 20% of the speed of the other 3.5 GB.
What other interpretation did you have in mind?
The last 0.5 GB is only connected by one memory lane, while the first 3.5 GB has many more of them.
It is just a bunch of whiny asshats who care about specs on paper rather than real world performance. The 970 is damn amazing. It makes the 980 nothing more than a overpriced luxury toy, and I say that as a 980 owner. Its performance is within 10-15% of a 980s and it is like half the price, what's not to like?
Also as for the memory thing this is actually a BONUS from nVidia, not a cripple. What I mean is in the past, they'd have just stuck 3.5GB on it and called it good. Then, if something needed more than 3.5GB, you go to system memory which is very slow 16GB/sec if you are running 16x PCIe 3 and much slower if you run less (like if you are doing SLI on a consumer board with PCIe 2 it would be 4GB/sec). However with this, you get another 512MB of RAM that is faster. Not as fast as the primary RAM, but much faster than hitting the system RAM over the PCIe bus. It won't perform as well as a 980 in those high memory situations, but it would perform better than if it just didn't have it at all.
I agree they should have noted it better, but really who gives a shit in reality? The 970 is the best "step down" card they've ever made compared to the highest end. Amazing value for the money and real world benchmarks from somewhere like HardOCP show it kills at modern games.
It's also funny how they act like nVidia did this to "harm" people for some business reason. If anything, they'd want to make the 970 look worse so people would be more likely to spend the near double to get a 980. However instead they made the 970 as close to the 980 as they could and I'm sure that ate in to the 980 market.
80% slower = 20% of the speed = 5 times slower (takes 5 times as long to do something) = 500% slower.
OR
80% slower = takes 80% longer to do something = 1/1.8 times slower = 0.55555555555 times slower = 44% slower.
See, ambiguous. There are 2 well defined definitions of 80% slower: gets 80% less done per unit time, OR takes 80% longer to get the same thing done. They have very different meanings. For one of them, 100% slower = gets nothing done ever, and the other means it takes takes twice as long (you have to wait an extra 100% of the original time). The same applies to 100% faster: one version means infinatly fast, and the other means twice as fast.
there's a whole industry devoted to suing "deep pockets". I once attended a presentation on this that explained to lawyers which companies they should target, how long they have to wait until they should gather plaintiffs for a class action law suit and how lucrative this is. There are even people financing these suits as an investment. I'm still not sure what is more unethical, lawyers leeching off of companies or companies lying to customers.
You're playing fast and loose with the terminology and adding extra steps that were never suggested to confuse yourself. KISS and it will all make sense.
No. The reason it affects the 970 is because it's missing some of the hardware that's in the 980.
I'm not the one who used the phrase in the first place. I'm just the one who somehow managed to understand a perfectly common phrase that really doesn't have a lot of ambiguity to it.
Perhaps if more people would take a moment to explain 'bi-weekly' (fortnightly), there wouldn't be so much confusion there either.
970 is only effected cause a shader core and ROP are disabled cause they are bad. the last 512mb ram shares a cluster so its slower cause that. Reviewers have test this and only time it became a problem is running super high rez with MSAA and scaling, pretty much to point probably running sub 40fps anyway.
I can't wait for my postcard-check in the mail for $0.21 explaining how the lawfirm got $34,500,000 of the $35,000,000 settlement and the remaining $500,000 gets divided among all GTX970 owners.
It does matter for CUDA and Opencl performance. So even though games might work around it (except on the mentioned high resolutions), it simply runs into problems if you are going to use it for computing on >3.5GB.
SHOCKING interview with Nvidia engineer about the 970 fiasco https://www.youtube.com/watch?...
There are plenty of examples of both usages of slower, showing that it is ambiguous across popular usage. For example, there's a whole 800% slower "genre" on Youtube where songs are played at 1/8 of their original speed. When popular usage is so sloppy, it's better to avoid the phrasing altogether in favor of something precise instead.
Maybe he works for Nvidia and that's how this mess got started?
So we take X - 800%X = -7X, hmmm I say to myself, spinning it backwards at 7x normal speed seems unlikely. They must not be using percentages right.
Note how the absurdity of the statement when interpreted the standard way tips me off that something is wrong. Doesn't that make more sense than pretending to be baffled when someone uses it correctly?
You may be on to something here!
Nope... the GF4MX was a terrible value, considerably worse than the Geforce 3,
it was a lot cheaper than the GF3, especially if you bought an OEM card.
Let's also not forget the considerable lack of honesty in marketing a product as the Geforce4
They didn't. It was the 4MX. It did in fact have some of the bits of the GF4 slapped onto its GF2 roots.
To compare with your truck analogy it would be like Ford bringing an "F450" to market which actually has a lower towing capacity than the F350 (note, I realize that Ford already has an F450, but the analogy still stands).
People get confused that diesels have less hauling capacity than gassers because of the heavy motor all the time, but nobody sues over it.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
No, its an ambiguous word and I've complained about it in life on numerous occasions.
- Michael T. Babcock (Yes, I blog)