Official Nvidia GTX 970 Discussion Thread

[Sammael]

The main thing I'm confused over is how no one at nvidia caught the misinformation.

 

 

I get how people "in the know" at nvidia had no qualms with the 970 being shown as a 4GB card, sure the last 500 MB has slower access than the rest of the memory and sure those two separate sections of memory CAN'T be accessed at the same time (according to anandtech), but there IS 4GB of memory on the card that can be used and filled with stuff.

 

 

What I DON'T get is how NO ONE caught the statements the 970 having 2MB of l2 cache instead of 1.75 or saying the 970 had 64 ROPs when it really has 56.

 

I can understand the marketing department making assumptions about it having that if not specified clearly, but did NO ONE that was writing drivers with intimate knowledge of the cards or ANYONE with deeper engineering knowledge raise their eyebrows when that incorrect information was being plastered on all the review sites?

 

Ryan and the guy at Anandtech think it perfectly plausible that the people in the know genuinely had no clue that the information that went out was incorrect.  But HOW?  Do the engineers not look at reviews of their own work?  Are they such COSMICALLY cloistered human beings that they never peruse the sites talking about the hardware they slaved over for YEARS of their lives?  THOSE people who ought to have known what was off just never EVER bothered to check sites and see the incorrect info?

 

 

I find that incredibly implausible.  I don't get how NO ONE at nvidia caught the wrong info.  If that was genuinely true, then it almost seems more plausible that NO ONE actually knew the cards did not have what they were said to have, not JUST the marketing department, but some of the engineers that were writing the material as well.  THEN when they inevitably saw the incorrect information plastered on different sites, they would not get red flags because THEY would be under the same misinformed information.  But even that is frightening to consider, that they did not really know what was going on with their cards.

 

 

I don't get how no one caught anything after so many months, someone please explain that to me.  I'm slow.

[MEC-777]

maybe they will give some free games for this.

and everyone will be hapy

 

No need. We all know how the cards perform and that hasn't changed. 

 

It's not like Nvidia did what Ubisoft did with AC4, claiming to deliver greatness and delivering crap instead.

[spartaman64]

They bought a cheaper card that wasn't a flagship card and were okay with it. Now everyone knows WHY it isn't a flagship card and they're getting real mad that it doesn't do 4K well (no benchmark said it would) and they're getting upset that it stutters when it should, because its NOT A HIGH POWERED FLAGSHIP WITH A FULL SIZED CHIP

but its stuttering because of the .5GB of slow ram and nvidia didnt tell anyone that it was 3.5GB of regular ram then .5GB of slow ram so people thought they were getting a full 4GB card. sure it is still a great performing card but how can we trust nvidia in the future now. if they say the 980 ti has 5000 cuda cores how do we know they dont have 1000 full cuda cores then 4000 cuda cores running at 1mhz

[NAP51DMustang]

Yup, this "revelation" suddenly made your GPU worthless, eh? Your GPU was already second tier trash to Nvidia, never to be a flagship card. Now we know precisely how Nvidia made it a 2nd tier card and everyone is getting their jimmies rustled. 

amazing. 

It's not about the quality of the GPU, its about the fact Nvidia completely lied to the public about a card and priced it based on those lies. It's called principle and if you don't stand for it then companies will do as they please.

[Sammael]

Except its still a 4GB card, did you even READ what Nvidia released on this issue? 

 

 

If I sold you a car with 4 tires, but one of the tires had hollowed out rubber on the interior so it was not easy to tell it was degraded and I passed if off as no different than any other, would you feel that there was a misrepresentation of what you bought?

 

 

It's clear that the access to the 4GB of memory is not identical, otherwise there would not be such a massive priority to allocate everything to the first 3.5GB section of memory.  And the reports on anandtech that you can't access both sections of memory at the same time.

 

http://www.anandtech.com/show/8935/geforce-gtx-970-correcting-the-specs-exploring-memory-allocation/2

 

"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 on the read side. 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 not both at once; it is a true XOR situation. Furthermore because the 512MB segment cannot be read at the same time as the 3.5GB segment, reading this segment blocks accessing the 3.5GB segment for that cycle, further reducing the effective memory bandwidth of the card. The larger the percentage of the time the crossbar is reading the 512MB segment, the lower the effective memory bandwidth from the 3.5GB segment."

 

So yes, you DO have access to 4GB of video ram on the card, but apparently the 970 can't read both the 3.5GB and 500MB section at the same time.  If people knew that worked differently, many might have either spent more to get the 980, or just gone with a reduced cost 290 custom card.   Or hell, even waited.  There are people who had nvidia 780s an got the 970s SPECIFICALLY because they had 4GB of memory and wanted to future proof their rigs for games that were utilizing more vram, particularly for 4k sli setups.

 

IF they had known and did not want to spend the cash on two 980s, they could easily have gone with a 295x2 instead of the 970s.  This memory disclosure has consequences, even if general performance is still excellent.  So was performance of already owned 780s vs the "upgrade" of a 970.

[Lynx]

No need. We all know how the cards perform and that hasn't changed. 

 

It's not like Nvidia did what Ubisoft did with AC4, claiming to deliver greatness and delivering crap instead.

yeah, but false advertisement is false advertisement

[Crion]

What I DON'T get is how NO ONE caught the statements the 970 having 2MB of l2 cache instead of 1.75 or saying the 970 had 64 ROPs when it really has 56.

 

I can understand the marketing department making assumptions about it having that if not specified clearly, but did NO ONE that was writing drivers with intimate knowledge of the cards or ANYONE with deeper engineering knowledge raise their eyebrows when that incorrect information was being plastered on all the review sites?

 

Ryan and the guy at Anandtech think it perfectly plausible that the people in the know genuinely had no clue that the information that went out was incorrect.  But HOW?  Do the engineers not look at reviews of their own work?  Are they such COSMICALLY cloistered human beings that they never peruse the sites talking about the hardware they slaved over for YEARS of their lives?  THOSE people who ought to have known what was off just never EVER bothered to check sites and see the incorrect info?

 

Most likely, the engineers don't read the reviewers guides which had the incorrect information leading to this "scandal".  I think people that are crying foul over this are either AMD fanbois or were completely delusional/did not pay attention to what reviews actually said about the card and are just whining. 

 

Everyone should have known going into this launch that the 970 was going to have stuff disabled due to it not being the flagship. 

 

EDIT: Oh, and I just looked over all the official nvidia docs on wayback and the incorrect information that was in the reviewers guides does not appear so the "false advertisement" claim probably wouldn't stand.

[spartaman64]

one thing nvidia needs to do immediately is tell manufacturers to update their info and put a disclaimer that only 3.5GB of the ram can run at full speed with .5GB of slower ram 

[SpaghettiCarbonara]

Just a few things.

1. I think it was Linus who said that you won't actually see any FPS drops, just stuttering and frame-pacing issues at higher resolutions. That's a big problem IMO.
2. Why the fuck are people bringing AMD into this!? AMD had nothing to do with this! If you're one of these people, you're just trying to ease the heat on Nvidia.

 

I saw this comment on one of PC Per's videos by EmmaCross94, I thought it was good.

 

 

While I don't own a high-end graphics cards, I can understand their customer's frustrations. Perhaps some of them are overreacting, perhaps not.

When Nvidia advertises this card as having 4GB of memory, they are giving the impression that the 4GB is a package of hardware that is all the same, a whole. If part of it is different from the rest, they should disclose this, no matter how small the difference is.

I don't care if it was a "mistake". Your average joe on the street can get away with a mistake because he's an individual human being, people make mistakes. Nvidia is not an individual, they are an international company. They should have known about this, so they were misleading. They have not made this mistake before, even when they were significantly smaller as a company than they are now.

Nvidia has been doing many shady things recently. Gameworks, Maxwell power consumption and now this. Why does anyone still trust them?

[spartaman64]

If I sold you a car with 4 tires, but one of the tires had hollowed out rubber on the interior so it was not easy to tell it was degraded and I passed if off as no different than any other, would you feel that there was a misrepresentation of what you bought?

 

 

It's clear that the access to the 4GB of memory is not identical, otherwise there would not be such a massive priority to allocate everything to the first 3.5GB section of memory.  And the reports on anandtech that you can't access both sections of memory at the same time.

 

http://www.anandtech.com/show/8935/geforce-gtx-970-correcting-the-specs-exploring-memory-allocation/2

 

"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 on the read side. 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 not both at once; it is a true XOR situation. Furthermore because the 512MB segment cannot be read at the same time as the 3.5GB segment, reading this segment blocks accessing the 3.5GB segment for that cycle, further reducing the effective memory bandwidth of the card. The larger the percentage of the time the crossbar is reading the 512MB segment, the lower the effective memory bandwidth from the 3.5GB segment."

 

So yes, you DO have access to 4GB of video ram on the card, but apparently the 970 can't read both the 3.5GB and 500MB section at the same time.  If people knew that worked differently, many might have either spent more to get the 980, or just gone with a reduced cost 290 custom card.   Or hell, even waited.  There are people who had nvidia 780s an got the 970s SPECIFICALLY because they had 4GB of memory and wanted to future proof their rigs for games that were utilizing more vram, particularly for 4k sli setups.

 

IF they had known and did not want to spend the cash on two 980s, they could easily have gone with a 295x2 instead of the 970s.  This memory disclosure has consequences, even if general performance is still excellent.  So was performance of already owned 780s vs the "upgrade" of a 970.

holy shit i thought it was a tiny difference of maybe 10GB/sec 50GB/sec at the most but that is massive

[EmmaLong]

This is a case of not doing your research, thorough research, before you make a purchase. If you were anticipating and planning to go 4K in the near future, then you should have looked more closely at all the 4k benchmarks thus far and then decided on a GPU. If you bought your 970 right at release, then I have no sympathy for you. You should have waited to see how they perform several months after release. If you'd done your homework you'd have realized the R9 290/290X are still better (overall) for 4K. You could have grabbed a 290X on sale and had a card that would give you decent 4k performance (better than a 970) and then added another later on or grab a 380X once they're released.

That's a new low, blaming the 970 users for Nvidia's false advertising. Shameful.

[NAP51DMustang]

That's a new low, blaming the 970 users for Nvidia's false advertising. Shameful.

They're anti-consumer fanboys. As long as Nvidia is doing well they don't care how they are treated.

[spartaman64]

also im surprised none of the manufacturers like asus and evga caught this. they must know about this as they make some of the cards. did they not see how this could be an issue if the consumers dont know that .5GB of the ram is way slower

[Snickerzz]

I love my 970... Though i still think i should be compensated somehow.. I bought the card because it was basically an underclocked 980. Guess ill wait 5 years for my $20 from the class action lawsuit

[Sammael]

I'm just going to leave this here.  When the 970 needs to access the vram located on the 500MB section, your effective memory bandwidth does exactly what this gif I found in a comment on pcper shows.

 

[deathjester]

So Nvidia lies. The bench marking sites/review sites are a joke and get advertising money and sponsorships to cherrycoat BS or not run a story at all. Tom's Hardware has nothing on this. PC Gamer has nothing on this. IGN? Nope. They have a article on the GTX 960 though, just like PC Gamer (where was that r9 285 article PC Gamer). I posted a link on PC Gamer and it was immediately removed and modded.

 

Yet people in here want to blame the consumer for not knowing what the media would not tell them, when we had several media sites selling SLI GTX 970 as the ultimate 4k solution and better than R9 290/x's and those articles were linked repeatedly in this very news section?

 

L O L. 

 

Oh and you also want to tell people as consumers what they can and can't be mad over, and then attack them like Faa did, to the point where they prob just said screw this forum.

 

I used a min itx Gigabyte GTX 970 in a build for my sister. I posted screenshots, pictures etc. I am not doing cartwheels over this news and I am very disappointed in Nvidia, MORE disapointed in our laughable benchmark/tech review sites (there are some exceptions but not many) who have shown repeated bias and I am laughing my butt off over people telling consumers what they can and can't be upset over. 

 

LTT forums has resorted more and more to fanboyism/blind bias and favoritism and it is becoming more of a joke.

 

PC Master Race my ass. More like fanboys, salesman and idiots.

[NAP51DMustang]

So Nvidia lies. The bench marking sites/review sites are a joke and get advertising money and sponsorships to cherrycoat BS or not run a story at all. Tom's Hardware has nothing on this. PC Gamer has nothing on this. IGN? Nope. They have a article on the GTX 960 though, just like PC Gamer (where was that r9 285 article PC Gamer). I posted a link on PC Gamer and it was immediately removed and modded.

 

Yet people in here want to blame the consumer for not knowing what the media would not tell them, when we had several media sites selling SLI GTX 970 as the ultimate 4k solution and better than R9 290/x's and those articles were linked repeatedly in this very news section?

 

L O L. 

 

Oh and you also want to tell people as consumers what they can and can't be mad over, and then attack them like Faa did, to the point where they prob just said screw this forum.

 

I used a min itx Gigabyte GTX 970 in a build for my sister. I posted screenshots, pictures etc. I am not doing cartwheels over this news and I am very disappointed in Nvidia, MORE disapointed in our laughable benchmark/tech review sites (there are some exceptions but not many) who have shown repeated bias and I am laughing my butt off over people telling consumers what they can and can't be upset over. 

 

LTT forums has resorted more and more to fanboyism/blind bias and favoritism and it is becoming more of a joke.

 

PC Master Race my ass. More like fanboys, salesman and idiots.

Fanboys kill all the good in the world.

 

Gaben be praised.

[mr moose]

That's a new low, blaming the 970 users for Nvidia's false advertising. Shameful.

 

I don't know what thread you're reading, No one's blaming the consumers for false advertising.  In fact not one person here has said the false advertising is o.k, we've all said that sucks.

 

What people are saying is that you can't blame Nvidia because the card doesn't perform better than the reviews.  Everyone knew (or should have) how the card performed before they purchased. that hasn't changed. 

[Zi6iX]

thats nvidia's nature, titan z for only 2999$ anyone?

[ACatWithThumbs]

Was about to buy a 970 right before the news of the memory issue hit now I'll just wait for Pascal.
I'm more upset about the false ROPs numbers than I am about the memory because that's straight up lying about specs.
Also how come that nobody noticed the wrong ROPs shouldn't GPU Z show the ROPs?

[That Norwegian Guy]

Always thought there was something sketchy about this card when compared to the GTX 980.

[Snickerzz]

I'm just going to leave this here.  When the 970 needs to access the vram located on the 500MB section, your effective memory bandwidth does exactly what this gif I found in a comment on pcper shows.

Honestly i kinda just hope nvidia releases a driver to lock the vram to 3.5 so that doesnt happen..

[Dabombinable]

So, Nvidia lied about the true specifications of the GTX 970, but it still performs better than my GTX 650ti 2GB, and the previous Quadro NVS 110M in my laptop. Yeah, I'm going to say that I'm not bothered at all as it is a massive step up from what I've been using in the last 16 years (Nvidia Geforce 256 32 MB>Diamond Stealth SIII S540 32MB>AMD Mobility Radeon 9000igp 128MB shared>Nvidia Quandro NVS 110M 128MB shared, 128-256MB shared>Nvidia GTX 650ti 2GB>Nvidia GTX 970)

[Sammael]

Honestly i kinda just hope nvidia releases a driver to lock the vram to 3.5 so that doesnt happen..

 

 

To be fair, as long as it's not sticking around in that space for too long, it's probably not a big deal, but if the card/game needs to address more than 3.5GB of vram... yeah, the user needs to just tone down the specs, or better yet have the geforce experience auto reduce the settings to make sure that is not an issue...

[wyattzx]

I honestly wouldn't mind if NVIDIA said that there was some misinformation, and that the card actually performs worse than an Intel 3000 integrated chip. Why? Because the benchmarks and hands-on performance are what sells the card like hotcakes. Not the "amount of cache" or "number of sweatshop workers not injured in production." No. People see the performance and price, and they buy it.