Gaming enthusiasts have been griping for months that Nvidia's GeForce GTX 970 graphics chip doesn't operate up to snuff, and now someone has taken the company to court over it.
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.
Responding to the issue last month, Nvidia acknowledged that the GTX 970 uses a different memory subsystem design than its higher-end GTX 980, but it said that difference has a negligible impact on performance.
Remember the rules guys. Be civil, be smart, be beautiful inside and outside.
NVIDIA knew that they was in the wrong from the start, after a while they tried to kinder cover it up.
NVIDIA employee on reddit :
Quote:
"We posted stats. We didn't properly explain the memory architecture. We messed up. We never intended to deceive anyone but our best intentions be damned you all got information that you thought you could trust and made decisions based on that.
We screwed up.
But, with that said, you feel different. You feel deceived and maybe you even feel like you don't have any options. You do.
If you don't want the card now, knowing what you know, 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.
"
Really? There are 4GB of GDDR5 memory on the card. It is clocked at 7000MHz effectively. That's all they advertised. There is no case here. Anand ran benches. Performance difference between this and the 980 could be attributed to cores and only cores, even when 3.5-4GB were in use.
Really? There are 4GB of GDDR5 memory on the card. It is clocked at 7000MHz effectively. That's all they advertised. There is no case here. Anand ran benches. Performance difference between this and the 980 could be attributed to cores and only cores, even when 3.5-4GB were in use.
Really? There are 4GB of GDDR5 memory on the card. It is clocked at 7000MHz effectively. That's all they advertised. There is no case here. Anand ran benches. Performance difference between this and the 980 could be attributed to cores and only cores, even when 3.5-4GB were in use.
Read up. Seems you are correct on the ROPs point, but they never advertised bandwidth best I can tell.
Quote:
To get straight to the point then, NVIDIA's original publication of the ROP/memory controller subsystem was wrong; GTX 970 has a 256-bit memory bus, but 1 of the 4 ROP/memory controller partitions was partially disabled, not fully enabled like we were originally told. As a result GTX 970 only has 56 of 64 ROPs and 1.75MB of 2MB of L2 cache enabled. The memory controllers themselves remain unchanged, with all four controllers active and driving 4GB of VRAM over a combined 256-bit memory bus.
If there were grounds for a lawsuit, it would be 192-bit GK106 and I think GK104 GPUs using 2GB of VRAM. Those actually had a memory controller disabled, and the bus width actually dropped on a hardware level from 192-bit to 64-bit for the last 512MiB. For the 970, it's just a memory partition that can still be accessed. The ROP count and L2 cache might have some hope of being taken seriously, but there were zero false claims with regards to memory.
EDIT: And even if it does turn out that the bus width is more accurately 32-bit plus 224-bit, well, that's still 256-bit according to marketers. How else can a dual-GPU have 768-bit or 1024-bit buses yet not increase bandwidth? It's misleading, but it isn't technically incorrect.
Really? There are 4GB of GDDR5 memory on the card. It is clocked at 7000MHz effectively. That's all they advertised. There is no case here. Anand ran benches. Performance difference between this and the 980 could be attributed to cores and only cores, even when 3.5-4GB were in use.
There's the loss of bandwidth (the fast memory on 970 only has 7/8'ths of the bandwidth of the fast memory on 980) as well as ROP's and Cache which were not disclosed. Also, i don't know if you've been keeping up with news, but many OCN members have demonstrated 970 refusing to allocate over 3.5GB of RAM even when required, resulting in heavy usage of the paging file on the system, stuttering and even crashing.
Quote:
For the 970, it's just a memory partition that can still be accessed.
It's only accessed in niche situations because accessing that pool prevents you from accessing the fast portion of the RAM for that cycle. Instead of an entire pool of 224GB/s, you can either access 196GB/s 3.5GB or 28GB/s 0.5GB. They're mutually exclusive, and in many games the driver will decide never to use that slow RAM at all.
But it doesnt. Imagine if NV didnt get caught with this scandal, they would of done it again;
"GTX-990 8GB OF VRAM but really its just 4 lol You wont know notice the difference unless you run skyrim with mods or any game that is heavy on VRAM. If you are going to benchmarks then make sure its does not require much VRAM." - some advertisement guy
Read up. Seems you are correct on the ROPs point, but they never advertised bandwidth best I can tell.
If there were grounds for a lawsuit, it would be 192-bit GK106 and I think GK104 GPUs using 2GB of VRAM. Those actually had a memory controller disabled, and the bus width actually dropped on a hardware level from 192-bit to 64-bit for the last 512MiB. For the 970, it's just a memory partition that can still be accessed. The ROP count and L2 cache might have some hope of being taken seriously, but there were zero false claims with regards to memory.
They advertised 64 ROP's, 2MB L2 and 4GB of VRAM at 224GB/s. Whole specs in the reviewers guide - which EVERYBODY got - and partial specs on the main site.
we got 56 ROP's, 1.75MB L2 and 3.5GB of VRAM at 196GB/s - the last 0.5GB at 28GB/s, but only readable on cycles where the main VRAM pool is not being read, aka completely unusable most of the time
Honestly, that's the first time I've seen that number officially. Either it hasn't come up or I'm extremely oblivious.
Article I linked to goes on to say that even that is technically correct:
Quote:
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 when doing a pure read or a pure write. In the case of pure reads for example, 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 it cannot read from both at once; it is a true XOR situation. The same is also true for writes, as only one segment can be written to at a time.
Its easy to win. Since an official one of the heads in NV has apologized and confessed that it was mistake its a easy win. If he didnt say "sorry" then it would of been a different story.
They should've advertised it as a 3.5GB cards with 0.5GB "overflow" cache.
This is sad, really. What they did is better than just cutting out that section of the memory controller entirely, but after this fiasco, they won't ever do that again.
It should have been just advertised at 3.5GB VRAM even though they come with 4096 MB of GDDR5. Nobody would have sued for getting more than what was advertised
They should've advertised it as a 3.5GB cards with 0.5GB "overflow" cache.
This is sad, really. What they did is better than just cutting out that section of the memory controller entirely, but after this fiasco, they won't ever do that again.
They should've advertised it as a 3.5GB cards with 0.5GB "overflow" cache.
This is sad, really. What they did is better than just cutting out that section of the memory controller entirely, but after this fiasco, they won't ever do that again.
But is it really better if people are having performance issues? The issue seems more prevalent in SLI setups, but still. They should have just reported it how it actually is and saved themselves tons of headache. Clear conscience and all.
The advertising of 4GB is correct (there is indeed 4GB of ram on the board), just like dual-card with double memory but only half ''used''. The thing is that the specs didn't match what one would expect. On your 8GB on the dual gpu card you get the rated memory speed while on the 970 you never get the advertised speed.
I hope they learn their lesson. I called it the VRAMgate and a mod deleted my post... Still wondering why. Looks like we will hear from that for a while.
Its easy to win. Since an official one of the heads in NV has apologized and confessed that it was mistake its a easy win. If he didnt say "sorry" then it would of been a different story.
Its probably going be settled out of court with NVIDIA giving people a refund. It depends on how ideological the plantiffs are if they are going to go for class action or not.
They probably won't release cutdown versions of stuff without getting rid of memory controllers instead of passing the cost to us. For that, there probably wont be any cut down versions of GM206.
Pulling stuff like what? Apparently there was a miscommunication between marketing and engineering and the wrong stuff was published. The lawsuit is correct with it being grounds of false advertisement, but I don't really think its more than that. I don't get why OCN looks into everything so cynically that everything NVIDIA does is part of some grand conspiracy theory.
Status
Not open for further replies.
You have insufficient privileges to reply here.
Related Threads
?
?
?
?
?
Ask a question
Ask a question
Overclock.net
27.8M posts
541.2K members
Since 2004
A forum community dedicated to overclocking enthusiasts and testing the limits of computing. Come join the discussion about computing, builds, collections, displays, models, styles, scales, specifications, reviews, accessories, classifieds, and more!