GTX 970 flaw

Status
Not open for further replies.
its not about SMM or ROP or even 3,5gb of vram

its about lying and misinformation in other words misleading the consumers about the real card specs. Fail of communication between the techs and nvidia PR teams . Say what ?

that is not OK and nvidia should answer for it
 
VonGriffin said:
its not about SMM or ROP or even 3,5gb of vram

its about lying and misinformation in other words misleading the consumers about the real card specs. Fail of communication between the techs and nvidia PR teams . Say what ?

that is not OK and nvidia should answer for it
Click to expand...

Nvidia already answered for it, they explained what was going on architecturally on the cards. If you want more than that, they aren't going to give it to you. Dozens of sites reviewed and recommended these cards, the performance information provided by those groups hasn't changed.

Also Nvidia got on top of this pretty quickly, it has only been a few days since this bubbled up and they have already provided all the information. Get over it.
 
ocellaris said:
Nvidia already answered for it, they explained what was going on architecturally on the cards. If you want more than that, they aren't going to give it to you. Dozens of sites reviewed and recommended these cards, the performance information provided by those groups hasn't changed.

Also Nvidia got on top of this pretty quickly, it has only been a few days since this bubbled up and they have already provided all the information. Get over it.
Click to expand...

Read about this this morning on twitter, and their explanation makes sense. It is an architectural issue in how the memory is placed on the board (physical traces and location) and addressed by the GPU logically due to the physical placement.
 
schizrade said:
Read about this this morning on twitter, and their explanation makes sense. It is an architectural issue in how the memory is placed on the board (physical traces and location) and addressed by the GPU logically due to the physical placement.
Click to expand...

Ummm... no.
 
I don't know much about the issue (I don't buy Nvidia products) but seeing this table on AnandTech made me die a little inside.

HE17cZl.jpg


"Slow segment"... SMH.
 
Nvidia already answered for it,
Click to expand...

well i understand that in the US people might get over things like those but in the EU we do have laws that forbid this kind of costumer misinformation

56 ROPs and 1792 KB of L2 cache or 64 ROPs and 2048 KB of L2 cache as advertised is not just a techical hindsight, personally i might just return my two cards
 
ocellaris said:
Nvidia already answered for it, they explained what was going on architecturally on the cards. If you want more than that, they aren't going to give it to you. Dozens of sites reviewed and recommended these cards, the performance information provided by those groups hasn't changed.

Also Nvidia got on top of this pretty quickly, it has only been a few days since this bubbled up and they have already provided all the information. Get over it.
Click to expand...

How did they answer for it?

Saying 'sorry we lied' is answering for something? LOL
 
ocellaris said:
Nvidia already answered for it, they explained what was going on architecturally on the cards. If you want more than that, they aren't going to give it to you. Dozens of sites reviewed and recommended these cards, the performance information provided by those groups hasn't changed.

Also Nvidia got on top of this pretty quickly, it has only been a few days since this bubbled up and they have already provided all the information. Get over it.
Click to expand...

Some people just love to make a mountain out of a mole hill!
I love my 970, no problems at all, and that includes coil whine!
 
its rather amazing aint it, amd or nivdia, ANY company that sells products that turn to be technically different than advertised is liable

Some people just love to make a mountain out of a mole hill!
I love my 970, no problems at all, and that includes coil whine!
Click to expand...


you have been trained well
 
evilsofa said:
No, they won't. Graphics performance of the PS4 is about equivalent to a GTX 750 Ti. The only case in which those games use more than 4GB of VRAM on the PC is when you have them running at settings current consoles can't even dream of.



Who's saying it's lost? You think it doesn't work at all?

"When a game needs less than 3.5GB of video memory per draw command then it will only access the first partition, and 3rd party applications that measure memory usage will report 3.5GB of memory in use on GTX 970, but may report more for GTX 980 if there is more memory used by other commands. When a game requires more than 3.5GB of memory then we use both segments."

The reason a lot of people in the witch hunt think their GTX 970 will only use 3.5GB of VRAM is that they have no idea how high the game settings have to be to use that much VRAM.



It doesn't cause stuttering or glitching. Anyone who thinks that has failed to understand the problem.

It also doesn't cause the extreme slowdown you see in NAI's benchmark. See EdZ's post below.



It's easy to solve. Attempt to reproduce the problem on your card. See what settings you need to run between 3.5GB and 4GB of VRAM on the games you're playing now. You'll figure out who is bullshitting who then.
Click to expand...

What about the missing ROP's and cache? They're blaming the marketing dept., but they don't make up the specs. They release what they are told.
 
LordEC911 said:
Ummm... no.
Click to expand...

Not much of a reader or technical mind I take it?

It is pretty simple, the 970 one less memory port to access frame buffer than the full part (980), which means one would be flooded the whole time, causing bottlenecks. Their solution was to limit the access on that last memory port to the final 500mb to a lower speed so as to not cause flooding on that last memory port. It is a physical limitation of the 970 binned part, and seems that the performance hit on benchmarks by non-random, spyware infected PCs (trusted sites) is minimal.

Dunno, maybe I am a paid NVidia shill. ;)
 
schizrade said:
Not much of a reader or technical mind I take it?

It is pretty simple, the 970 one less memory port to access frame buffer than the full part (980), which means one would be flooded the whole time, causing bottlenecks. Their solution was to limit the access on that last memory port to the final 500mb to a lower speed so as to not cause flooding on that last memory port. It is a physical limitation of the 970 binned part, and seems that the performance hit on benchmarks by non-random, spyware infected PCs (trusted sites) is minimal.

Dunno, maybe I am a paid NVidia shill. ;)
Click to expand...

The check is in the mail.


You have to ask yourself why you choose to defend a big business company that's been caught lying.
 
schizrade said:
Not much of a reader or technical mind I take it?

It is pretty simple, the 970 one less memory port to access frame buffer than the full part (980), which means one would be flooded the whole time, causing bottlenecks. Their solution was to limit the access on that last memory port to the final 500mb to a lower speed so as to not cause flooding on that last memory port. It is a physical limitation of the 970 binned part, and seems that the performance hit on benchmarks by non-random, spyware infected PCs (trusted sites) is minimal.

Dunno, maybe I am a paid NVidia shill. ;)
Click to expand...

You were talking about physical traces on the board to the memory.......
 
VonGriffin said:
well i understand that in the US people might get over things like those but in the EU we do have laws that forbid this kind of costumer misinformation
Click to expand...

We do in the US as well but it is EXTREMELY rare for companies to actually have it used against them. Same with lots of laws that target anyone more powerful than the Average Joe.
 
evilsofa said:
They don't matter. When you see forum posters outraged about disabled SMMs or the number of ROPs, that's your cue that they don't understand the problem and are just blurting out buzzwords:

"Before people complain about the ROP count difference as a performance bottleneck, keep in mind that the 13 SMMs in the GTX 970 can only output 52 pixels/clock and the seven segments of 8 ROPs each (56 total) can handle 56 pixels/clock. The SMMs are the bottleneck, not the ROPs."

"Second to that, it turns out the disabled SMMs have nothing to do with the performance issues experienced or the memory system complications."

http://www.pcper.com/reviews/Graphi...Full-Memory-Structure-and-Limitations-GTX-970

Accusations that Nvidia lied about the number of ROPs is a red herring. If the 970 had 64 ROPs like the 980 does, it wouldn't change anything about this issue; it would still be bottlenecked by the disabled SMMs.
Click to expand...

Dave65 said:
Some people just love to make a mountain out of a mole hill!
I love my 970, no problems at all, and that includes coil whine!
Click to expand...

I own two 970s myself. I don't game at 4K, have never run into this vram wall, but I'm still mad as fuck about this whole ordeal.

You guys forget something: sometimes it's the PRINCIPLE that matters. You let this one slip, next time who knows what nVidia will pull.
 
Coil whine, ROPS, L2 cache, 0.5GB of useless VRAM, SLI voltage discrepancies, MFAA taking forever to get released, DSR not working on SLI, ect. ect.

No one should be down playing this garbage.
 
Pretty annoyed at this and I bought a pair of them in SLI, and in most cases I can't even get full performance out of these cards since most games already have hit the 3GB wall with constant stuttering and low GPU usages resulting in low framerates. Already listed one on eBay for sale and I don't want to deal with this bullshit anymore.
 
LordEC911 said:
You were talking about physical traces on the board to the memory.......
Click to expand...

What do you thing that last 512 mb of ram is connected to the single memory bus on the die with? Transwarp tubes? Unicorn tear trails?

I am not defending their shit bag marketing. They screwed something up there, but the memory thing has been massively overblown.
 
Actually I saw on Overclock.net a good possible explanation for why not everyone is having the issue (sorry could easily find the explanation to give credit, it is blowing up over there). He thought that it depends on which SMX were disabled to make the 970. And also why it seems everyone is figuring it out now= A lot of people got 4K monitors during after Christmas and now they are hitting that Vram wall whereas they likely had a lower res screen before and weren't.
 
n=1 said:
I own two 970s myself. I don't game at 4K, have never run into this vram wall, but I'm still mad as fuck about this whole ordeal.

You guys forget something: sometimes it's the PRINCIPLE that matters. You let this one slip, next time who knows what nVidia will pull.
Click to expand...

This.

If people that own 970's and are still whining like babies about what and what not definitely don't deserve to play with hardware in the first place.

Second, yes, they should have come clean. It's as simple as that, it's a matter of principle, and not a matter of periphrasing BS.

To the European guy who thinks that companys can get away with this in the USA, no they can't. I doubt that one would sue Nvidia for misinformation just because they couldn't max out witcher 2 or some BS.


It's not a matter of what way they used whatever they did with the memory, it's a matter of coming clean, especially if it effects us customers.
 
n=1 said:
I own two 970s myself. I don't game at 4K, have never run into this vram wall, but I'm still mad as fuck about this whole ordeal.

You guys forget something: sometimes it's the PRINCIPLE that matters. You let this one slip, next time who knows what nVidia will pull.
Click to expand...

You mean like 290X review samples that are faster than retail cards?

http://techreport.com/news/25609/up...9-290x-cards-may-be-slower-than-press-samples

I guess you should just switch to Intel graphics on principle.
 
PRIME1 said:
You mean like 290X review samples that are faster than retail cards?
Click to expand...
Wasn't that an issue with fan curves in the driver? And they immediately fixed it with a simple driver update. It could have even been an accident... AMD has lots of those. And of course, AMD actually acknowleged it was a problem and didn't attempt to make lame excuses about it.

The VRAM limitations were done intentionally by Nvidia, and it seems to be a hardware problem which means no driver update can fix it.
I don't think AMD is violating any ethics principles by, let's say, screwing up their drivers which is a pretty common occurrence.

I'm sure AMD has done some shady things in recent years, remember the texture compression thing a few years ago? Somewhere around 2010 maybe. That's a good one because they did it on purpose to boost their benchmark scores.

edit: Got it
http://archive.benchmarkreviews.com/index.php?option=com_content&task=view&id=12845&Itemid=47

Still a software issue, though. Which means it can actually be fixed.
 
TaintedSquirrel said:
The VRAM limitations were done intentionally by Nvidia, and it seems to be a hardware problem which means no driver update can fix it.
Click to expand...

Physical limitation of the physical silicon. You cannot run full bore on one pathway vs. 2. 980 has 8, 970 has 7, 3 pairs and an outlier, so one has to get the performance chop.
 
To PRIME1

If nvidia thought this was such a great feature, why didn't they tell anyone about it?

Why keep it a secret?

Why when people starting asking questions, did nvidia respond by saying we have to look into it (knowing exactly why the ram preformed the way it did)?

Nothing seem fishy to you?
 
trick0502 said:
Why when people starting asking questions, did nvidia respond by saying we have to look into it (knowing exactly why the ram preformed the way it did)?
Click to expand...
The PR guy answering all the spam emails doesn't know how the cards are designed.
He has to climb out of his office chair and walk a few floors down to drag the engineers away from their Cheetos and cat videos to make a write-up about what's causing the issue.

Of course it's going to take a few days to get detailed info released. There's nothing fishy about it.
Although if you want to complain about the actual info they released, that's a different story.
 
PRIME1 said:
You mean like 290X review samples that are faster than retail cards?

http://techreport.com/news/25609/up...9-290x-cards-may-be-slower-than-press-samples

I guess you should just switch to Intel graphics on principle.
Click to expand...

I guess you didn't follow the story because the card was simply thermal throttling to all hell due to the absolutely shitastic AMD reference blower. It was a fixable issue. The 970's issue is unfixable, in addition to the product itself being misrepresented.

And also, two wrongs don't make a right.
 
schizrade said:
What do you thing that last 512 mb of ram is connected to the single memory bus on the die with? Transwarp tubes? Unicorn tear trails?

I am not defending their shit bag marketing. They screwed something up there, but the memory thing has been massively overblown.
Click to expand...

All the memory is connected to the package with traces. I'm not sure why you are mentioning it like it is an issue...

durquavian said:
He thought that it depends on which SMX were disabled to make the 970.
Click to expand...
The point of having the SMMs connected to a crossbar is so that it doesn't matter.
 
Last edited:
And also, two wrongs don't make a right.
Click to expand...

I love the deflection and how all the forum shills are on the clock 24/7 lol!
They will not stop trying to make this a non-issue when everyone knows that computer gamers have ALWAYS expected the memory advertised to work at full speed and be equally accessible by their GPU.When buying a GPU many of us look at VRAM as a possible limiting factor when considering future games or resolutions. As a 7970 owner I know I would have never considered the GTX970 if I had always known it was really a 3.5GB GPU. I want a full 4GB upgrade and not something gimped.
 
You and me both brother. My original plan was either 2x 970s until GM200/390X dropped, or 2x 980s to hold me over until 16nm in 2016.

Well I cheaped out and went for the 970s at the time because I just couldn't justify the 980s to myself. But so far the GM200 has been a disappointment from the leaked benchmarks, not to mention the ridiculous rumored MSRP of $1350 on the Titan X/Titan II. And then 390X won't be released until June, and now this fucking 3.5GB vram bullshit.

I really would've been better off had I just swallowed the pill and got 2x 980s instead. *sigh*
 
Putting aside Nvidia lying and now trying to cover their butts, what gets me is what a stupid design the 970 is. Fragmented memory royally blows, I can only guess Nvidia is doing this because it makes it easier for them to die harvest. But just stop with this crap Nvidia, are you really that desperate to save a few dollars per GPU?

This kind of cost cuttings shows no respect for the customer at all.
 
n=1 said:
You guys forget something: sometimes it's the PRINCIPLE that matters. You let this one slip, next time who knows what nVidia will pull.
Click to expand...

Exactly. Full disclosure is the key here, everyone would have understood the need to cut the chip down a bit. They're only correcting the specs now because someone found out and twisted their arm over it.

Same thing with AMD's sneaky optimization to HDR rendering surfaces by using reduced precision. Sure, the "end result" is faster and very close to the unoptimized render, but the wrongdoing is purely in how they represented the change. The community called-out AMD over the issue and now we have the option to disable the optimized render. I leave it enabled because I can't tell the difference, but I certainly don't want my intelligence to be insulted by sneaking around the truth like that.

Software is one thing, but misrepresenting hardware is a big no-no.

nVidia had the opportunity to push the segmented memory as a selling point. Slap a 3.5 GB sticker on it and have another sticker that says 512MB SideCache or some crap. The benchmarks are what matter in the end, but it DOES matter how you got those numbers.

§kynet;1041387531 said:
Putting aside Nvidia lying and now trying to cover their butts, what gets me is what a stupid design the 970 is. Fragmented memory royally blows, I can only guess Nvidia is doing this because it makes it easier for them to die harvest. But just stop with this crap Nvidia, are you really that desperate to save a few dollars per GPU?

This kind of cost cuttings shows no respect for the customer at all.
Click to expand...

I could be wrong, but I think its a design limitation of how the memory controllers are peered with the compute cores. The memory controllers that would have been directly wired to their own compute cores are now only accessible to the remaining compute cores through a shared bus that is much slower. EDIT: anand has a graphic up showing the disabled L2 cache chunk between the compute cores and the memory controller as the culprit. So the crossbar between the compute cores isn't the cause, its the path through the disabled L2 to the last .5GB segment.

http://images.anandtech.com/doci/8935/GM204_arch.jpg

If this is accurate, there is one L2 chunk handling both of the memory controllers for the final two .5GB segments. I am no engineer, but could this impact performance with more than just the final .5GB segment? The last L2 would seem to be responsible for the entire 1GB segment. I suppose if the last .5GB is effectively avoided, then the L2 should see no additional impact. However, the moment that cache is responsible for both segments, I can imagine the performance of the entire 1GB segment might be impacted. Or perhaps the final .5GB segment can operate without the L2 entirely? I guess Anand makes it seem like the L2 isn't used with the final segment, but I'm not so sure. He did say that the crossbar cannot read from both segments at the same time, so the main partition should ideally be writing while the small one is reading.

Anand said:
In concept, the larger the percentage of the time the crossbar is reading the 512MB segment, the lower the effective read memory bandwidth would be from the 3.5GB segment.
Click to expand...

At any rate, it sounds like a mess. It might have been a better idea to scrap that last memory lane entirely. Maybe nVidia will come up with a driver option to disable that last segment so at least people know their real ceiling.
 
Last edited:
Gundamit people, you still HAVE 4gigs of VRAM, it DID NOT just dissapear on your 970's

The "problem" is that Nvidia "forgot/lied" about specs indicating that accessing the last 500megs is technically "slower" then the first 3.5gigs vs the 980 which has all SMM pathways enabled as a result of a "higher quality" die from the manufacturing process :rolleyes:

Or did everyone convinently "forget" that all 970 dies were "flawed/imperfect" 980 dies and were rebadged for 970 model usage?

Up until last week, not ONE OF YOU was bitching about a sudden "slow down" in your games with the 970. Now its the placebo effect in full action.
 
I'll put good money up that Nvidia introduces driver "optimizations" for the 970 that shuffle ram around to minimize the potential slow-down.

All of that bruhaha aside, R9 290s are back down under $250. This yet again makes a 970 purchase tenuous on the bang-for-the-buck scale unless you're an Nvidian who would have paid the extra $$$ anyways just because its green.
 
Up until last week, not ONE OF YOU was bitching about a sudden "slow down" in your games with the 970.
Click to expand...

look at it this way: if people didnt notice the vram difference would nvidia ever come up clean about it by itself?

Nvidia droped the ball here: we know it, media knows it, thy know it, thy never did something like this before and without an outcry thy will do it again
 
The 970 still performs very well for the money while using very little electricity / heat, yes it is not ideal but you can always sell the 970 by the time more than about 5% of games actually need more than 3.5gb ram and get a different card, I feel a bit sorry for people who have bought 970 SLI though because the ram amount will make more difference there. Its not good what nvidia have done and is another reason I will probably go back to AMD again, shame because I was just starting to like nvidia again and now they do this.... (also the ridiculous price for gsync vs freesync)
 
VonGriffin said:
well i understand that in the US people might get over things like those but in the EU we do have laws that forbid this kind of costumer misinformation

56 ROPs and 1792 KB of L2 cache or 64 ROPs and 2048 KB of L2 cache as advertised is not just a techical hindsight, personally i might just return my two cards
Click to expand...

This.

64 ROPs and 2048 KB of L2 cache is what consumers were told multiple times. 56 ROPs and 1792 KB of L2 cach was what was actually shipped.

What the apologists forget (rather conveniently, I might add) is that there are consumer protection laws in place to protect consumers against companies making false claims.
 
Status
Not open for further replies.
Back
Top