Petitioning for 970 Refund

[Nenu]

Why do you spread false information?
I have posted a link in this very thread debunking that claim:

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



Please stop posting false information as a "argument".

Interesting.

The memory block diagram (provided by Alben) shows 2 links from the crossbar for each block.
Except the last one where there is only one cache element for 2x 0.5GB memory blocks.
It wont be possible to get full bandwidth ever.
1/8 of the bandwidth capability of the crossbar is unavailable.

The following statement made by Alben from NVidia is misleading because the full bandwidth should be higher than 224GB/s.

To those wondering how peak bandwidth would remain at 224 GB/s despite the division of memory controllers on the GTX 970, Alben stated that it can reach that speed only when memory is being accessed in both pools.

Another important point.
That single crossbar link and its cache are shared with the last 0.5GB of the "fast" area"
The whole last GB of ram can cause stutter if over utlilised.
This might be why its difficult to push it to 3.5GB.

 

[Factum]

Interesting.

The memory block diagram (provided by Alben) shows 2 links from the crossbar for each block.
Except the last one where there is only one cache element for 2x 0.5GB memory blocks.
It wont be possible to get full bandwidth ever.
1/8 of the bandwidth capability of the crossbar is unavailable.

The following statement made by Alben from NVidia is misleading because the full bandwidth should be higher than 224GB/s.



Another important point.
That single crossbar link and its cache are shared with the last 0.5GB of the "fast" area"
The whole last GB of ram can cause stutter if over utlilised.
This might be why its difficult to push it to 3.5GB.

So you will stop with the false atgument from now on, now that you are up-to-date and don't use false information?

 

[Nenu]

So you will stop with the false atgument from now on, now that you are up-to-date and don't use false information?

You are missing the point.
The cards are advertised as having a 256 bit memory interface.
The available bandwidth is quoted as 224GB/s not the expected 256GB/s from 256 bits.
Its a 224 bit memory interface from the bandwidth figures and is demonstrated to be 224 bit on the block diagram NVidia provided.

I'm not the one misinterpreting the problem.

 

[Factum]

You are missing the point.
The cards are advertised as having a 256 bit memory interface.
The available bandwidth is quoted as 224GB/s not the expected 256GB/s from 256 bits.
Its a 224 bit memory interface from the bandwidth figures and is demonstrated to be 224 bit on the block diagram NVidia provided.

I'm not the one misinterpreting the problem.

I am not missing the point, just 3 pages back you posted this:

Nothing theoretical about them, they are the capabilities of the hardware.
Also, the 2 blocks of memory cannot be accessed simultaneously. Read the thread.
You have no case.

I am asking if you are done with that false argument (which you used to bash another post with) so we can look at your other claims?

 

[Nenu]

You pointed out there was an update that cleared that point up, understanding is key.
Even so, there is no guarantee you have the max available bandwidth when performing a memory access and reduced available cache and bandwidth for the last GB has been demonstrated to cause problems which alone are enough to warrant a refund.

And it takes nothing away from a 256bit memory interface being advertised when the actual interface to the memory is 224bit.

 

[trick0502]

Fact: the last 0.5gb operate at 1/7 the speed of the first 3.5gb.

Could you imagine if Intel or amd released a cpu with 4 cores at 3ghz. Then two months later said oops, marketing messed up the specs. It is 4 cores, but the 4th core can only operate at 1/4 the speed of the other 3 cores. No one would be mad right?

 

[trick0502]

You seem to be missing that the article was updated, is that on purpose?
Do you have anby gaming benchmarks (aka real world usage) that shows this behaviour?


I am now the owner (via my daughter) of a GTX970 unlike, I suspect, most of the people that are "up in arms over this"...are you?

I don't think you understand what I'd going on here. Nvidia strait up lied to you about the specs of this card. Sold millions of units (totally guessing), and never came clean about the specs until they were called out on it. This is about being lied to as a consumer, not about "real work benchmarks". The "real world benchmark" is their release information on the gpu, with the wrong specs. I don't buy for a second that marketing messed this up either. If this was really on marketing, some one would have lost their job by now, and that hasn't happened.

 

[Creig]

It's amazing the lengths some people will go to defend Nvidia even after Nvidia themselves finally admitted that the specs they gave reviewers were wrong. And the fact that Nvidia did not own up to the error until forced to do so after the gaming community discovered it on their own through benchmarking seems to make little difference to them whatsoever. It's as if Nvidia can do no wrong.

EVER.

 

[spaceace420]

Okay so I did manage to get back into the last 512 with shadows of mordor but it took some insane 4k at ultra and getting into a huge fight with 3x level 12 captains at once. Pretty much unplayable but I did win the fight, performance sucked as one would expect at those settings and my pagefile hit 13 GB. Now what is strange is having to use 4k when before 1080p at ultra would get me up in that range easy. If the full 4 GB is truly usable at the same time why does it fight so hard to keep us out of it? GPU usage did go down again while up in the last segment ( in the 80 % instead of pegged at 99% where the game normally is). Again I love the card and performance just cannot stand corporations being defended when they obviously have done something very wrong. I'll fire the game up later at 1080p ultra and try to recreate the fight where I hit above 3600 but that 4k should have got me to the max not just the ~3800's I was in.
Maybe these same people defending Nvidia can be witnesses in the court case that is sure to be coming They might not have to pay off the system as much then.

 

[Starrbuck]

performance sucked as one would expect at those settings and my pagefile hit 13 GB. Now what is strange is having to use 4k when before 1080p at ultra would get me up in that range easy. If the full 4 GB is truly usable at the same time why does it fight so hard to keep us out of it?

Pretty certain they made a stealth change to the drivers to try to hide the issue. No way the 4GB is actually "useable" at full-speeds.

 

[FlipCody]

Okay so I did manage to get back into the last 512 with shadows of mordor but it took some insane 4k at ultra and getting into a huge fight with 3x level 12 captains at once. Pretty much unplayable but I did win the fight, performance sucked as one would expect at those settings and my pagefile hit 13 GB. Now what is strange is having to use 4k when before 1080p at ultra would get me up in that range easy. If the full 4 GB is truly usable at the same time why does it fight so hard to keep us out of it? GPU usage did go down again while up in the last segment ( in the 80 % instead of pegged at 99% where the game normally is). Again I love the card and performance just cannot stand corporations being defended when they obviously have done something very wrong. I'll fire the game up later at 1080p ultra and try to recreate the fight where I hit above 3600 but that 4k should have got me to the max not just the ~3800's I was in.
Maybe these same people defending Nvidia can be witnesses in the court case that is sure to be coming They might not have to pay off the system as much then.

Interesting. You have a similar setup to me and I have zero problems at 1440p. I have been playing this game a lot at 1440p with ultra HD texture pack and getting over 3,584mb (3.5gb) very frequently with no issues. All my settings are maxed out. I have had times when I'm fighting over 30 orcs, etc. and it is silky smooth. I have been monitoring memory use with MSI Afterburner.

How are you monitoring memory use? Are you running with DSR and at what percent? What speed are you running your computer ram at?

 

[spaceace420]

cpu down to 4.5 right now for testing normally 4.7-5.0 and no overclock on the card beyond the factory on this run. If you are at 1440 you should be above the 3.5 I get at 1080p shouldn't you? Turn up the in game rez just to see if you can get full utilization if you would for me. I was getting well over 3584 with 1080p ultra and hd textures before the latest driver. Even remember some graphs getting up to 3960 and with Afterburner monitoring since it is known to work. Last nights testing was at 200% = 4k just to force it above.

 

[spaceace420]

Pretty certain they made a stealth change to the drivers to try to hide the issue. No way the 4GB is actually "useable" at full-speeds.

That is my opinion as well, Hoping to get someone to help me confirm it though. There is no way I shouldn't have been maxing out the 4Gigs last night but I forgot to have afterburner monitor frame times but I know they were all over the place just from the lag I was hitting on button presses. In no way am a saying the card wasn't pushed beyond what I would actually play at but I kept at it to make sure it was in the last bit, which it still didn't quite hit the end of the ram and 13 gigs of page file seems insane to me at least.

 

[Starrbuck]

There is no way I shouldn't have been maxing out the 4Gigs last night but I forgot to have afterburner monitor frame times but I know they were all over the place just from the lag I was hitting on button presses.

I believe data gets written to this "page file" in Afterburner when it won't fit in the graphics memory.

Here is a recent reddit discussion:

http://www.reddit.com/r/EvolveGame/comments/2l1z7p/warning_to_ssd_users_check_your_pagefile_usage/

I have 16GB of DDR4. I'm thinking of disabling the page file to see what happens. I don't like the idea of all this data being written to SSD.

In no way am a saying the card wasn't pushed beyond what I would actually play at but I kept at it to make sure it was in the last bit, which it still didn't quite hit the end of the ram and 13 gigs of page file seems insane to me at least.

Problem is we would be pushing the cards harder and harder over the next few months and what seems past the point you want to push it today may seem mainstream soon!!!

 

[FlipCody]

cpu down to 4.5 right now for testing normally 4.7-5.0 and no overclock on the card beyond the factory on this run. If you are at 1440 you should be above the 3.5 I get at 1080p shouldn't you? Turn up the in game rez just to see if you can get full utilization if you would for me. I was getting well over 3584 with 1080p ultra and hd textures before the latest driver. Even remember some graphs getting up to 3960 and with Afterburner monitoring since it is known to work. Last nights testing was at 200% = 4k just to force it above.

Did you only have issues running with DSR at 200%? If that is the case the performance issue is not related to memory. If 200% is equivalent to 4K I do not believe a single GTX 970 would be able to run this game playable with ultra settings at 4k. I have seen benchmarks at 4k with ultra settings with a single card on this game and they are around 30 fps. If you get in a fight with lots of orcs I suspect it would drop well below that. The GTX 970 is not a single 4K card. Arguably even a single GTX 980 is not a 4K video card. I believe for a game like this with ultra settings, smooth game play, and 4K you would have to run SLI.

edit: I believe 200% is 2160p and not 4K.

As I stated I'm running 1440p and getting extended runs above 3.5gb ever time I play the game and I have been playing it a lot. Silky smooth.

 

[spaceace420]

Did you only have issues running with DSR at 200%? If that is the case the performance issue is not related to memory. 200% is equivalent to 4K and I do not believe a single GTX 970 would be able to run this game playable with ultra settings at 4k. I have seen benchmarks at 4k with ultra settings with a single card on this game and they are around 30 fps. If you get in a fight with lots of orcs I suspect it would drop well below that. The GTX 970 is not a single 4K card. Arguably even a single GTX 980 is not a 4K video card. I believe for a game like this with ultra settings, smooth game play, and 4K you would have to run SLI.

As I stated I'm running 1440p and getting extended runs above 3.5gb ever time I play the game and I have been playing it a lot. Silky smooth.

Everything you said I know, this is not about performance this is about the driver using 13gb of pagefile before it allocates the whole 4 gigs of vram. I absolutely know it is not capable of 4k 60fps and has nothing to do with what I am testing. Not sure if you misunderstood what I am trying to accomplish here but the performance at 1080p is fine for me. Starbuck seemed to have a firm grasp on what I am talking about and when I was testing other drivers with a smaller page file I ended up crashing from using all 8 of ram and the pagefile. I was hitting closer to the 4 gigs of vram before the latest driver and now I am not and the pagefile is even larger. som at 4k should be pegging out the vram but it is not is the issue not the performance at that rez. Games going forward are going to want more ram and I moved my pagefile off of my ssd just because of how much it was using to write too. To summarize I am happy with my card and not returning it but Nvidia pulled some major shit with this card and IMO still are with the drivers.
Chris

 

[FlipCody]

Everything you said I know, this is not about performance this is about the driver using 13gb of pagefile before it allocates the whole 4 gigs of vram. I absolutely know it is not capable of 4k 60fps and has nothing to do with what I am testing. Not sure if you misunderstood what I am trying to accomplish here but the performance at 1080p is fine for me. Starbuck seemed to have a firm grasp on what I am talking about and when I was testing other drivers with a smaller page file I ended up crashing from using all 8 of ram and the pagefile. I was hitting closer to the 4 gigs of vram before the latest driver and now I am not and the pagefile is even larger. som at 4k should be pegging out the vram but it is not is the issue not the performance at that rez. Games going forward are going to want more ram and I moved my pagefile off of my ssd just because of how much it was using to write too. To summarize I am happy with my card and not returning it but Nvidia pulled some major shit with this card and IMO still are with the drivers.
Chris

I edited my other post. Not really really sure how the percents work with DSR. Is 200% (at 1080p) 2160p and not 4k?

Edit: after re-reading your original post I get what you were getting at with the pagefile.

 

[spaceace420]

No 200% at 1080p is 4k, It's an in game option for SoM kind of like BF4's overscaling. This game is my addiction right now with a little metro 2033 redux tossed in as well. Kick ass game with great animation and lots of orcs to kill. Thats why I still finished the battles at what I would call almost unplayable settings, some of the time in battle it was down to 17fps and close to a blurry mess but I wanted to see if it would get the vram maxed out and not drop the gpu use. Didn't max out the vram and gpu usage dropped into the low 80's in some parts.

 

[rinaldo00]

I edited my other post. Not really really sure how the percents work with DSR. Is 200% (at 1080p) 2160p and not 4k?

200% is double and it doubles in both directions, so it is 4 times 1080p which equals 4K

EDIT is it a matter of terminolgy? 2160 IS 4 times 1080 and is also called 4K UHD. What are the dimensions that spaceace420 is calling 4K? The other 4K res, 4098 x 2160, is mostly in movies not monitors or TVs

 

[AFD]

It's okay guys. Jen-Hsun just apologized and stated that this issue is a good design, and that otherwise, it would've been impossible for them to have given us anything larger than 3GB for the 970..

Instead of being excited that we invented a way to increase memory of the GTX 970 from 3GB to 4GB, some were disappointed that we didn't better describe the segmented nature of the architecture for that last 1GB of memory.

http://www.techpowerup.com/210149/it-wont-happen-again-nvidia-ceo-breaks-silence-on-gtx-970-controversy.html​

It's all better, I'm excited for having gimped 4GB

 

[Revdarian]

Darn you guys, how could you haven't been happy about the "feature" of the last 0.5GB of on board ram?

http://blogs.nvidia.com/blog/2015/02/24/gtx-970/

^^ Jen-Hsun's take on the issue...


AFD beat me to it

 

[Brackle]

He's calling it a feature. So you are telling me Nvidia marketing department, who spins Physx to people and brainwashes them, totally forgot to mention a Feature of a video card to people?

Wow......call my mind blown, talk about pure ignorance LOL

Sad really.

 

[rinaldo00]

He's calling it a feature. So you are telling me Nvidia marketing department, who spins Physx to people and brainwashes them, totally forgot to mention a Feature of a video card to people?

Wow......call my mind blown, talk about pure ignorance LOL

Sad really.

More like arrogance

 

[{NG}Fidel]

Well at least he finally opened his mouth, but yeah thats quite the snake oil salesman explanation on the issue.

 

[spaceace420]

3840 x 2160 4k is what I was referring to as 4k
His spin is why he is the ceo, glad they finally spoke up but that is a horrible apology. feature my ass

 

[rinaldo00]

3840 x 2160 4k is what I was referring to as 4k

OK, So you both agree. 3840 x 2160 is called 2160p or 4K UHD

 

[spaceace420]

yep 4x the pixels-I can't believe how arrogant the ceo comes off in that statement, just blown away by them now. My next card will not be green team out of principle

 

[TekRok]

"This new feature"

HOLY SHIT...wow

 

[Deleted member 83233]

yep 4x the pixels-I can't believe how arrogant the ceo comes off in that statement, just blown away by them now. My next card will not be green team out of principle

If only it were that simple a choice for me. Unfortunately there really isn't parity between AMD and Nvidia. Performance-wise they teeter back and forth, so that's not the concern. Unfortunately every other difference between the two very much sway me toward NV. I guess if I'm going to take issue with them, it's back to Matrox for me.

 

[filip]

Instead of being excited that Nvidia invented a way to increase memory of the GTX 970 from 3GB to 4GB you guys are here getting mad. This is a revolution, Nvidia has invented a way to increase 3GB GPUs to 4GB GPUs!!!!!

 

[Sp33dFr33k]

The explanation just didn't make any sense to me. They had to sort of cripple the GTX 970 so lesser models would better benefit from the architecture? Granted I'm no engineer but seems like they should have found a better way to enable 4GB. I'd still buy one I'm just waiting for the price to drop.

 

[rinaldo00]

The explanation just didn't make any sense to me. They had to sort of cripple the GTX 970 so lesser models would better benefit from the architecture? Granted I'm no engineer but seems like they should have found a better way to enable 4GB. I'd still buy one I'm just waiting for the price to drop.

They found a way to turn parts not good enough for the full 4 GB 980 into the 3.5 GB + .5 GB 970, The spin is you get the extra .5 GB 'for free'

 

[Nenu]

They found a way to turn parts not good enough for the full 4 GB 980 into the 3.5 GB + .5 GB 970, The spin is you get the extra .5 GB 'for free'

Its brilliant, as long as you dont use it.

They could stop any further problems by making the card report it only has 3.5GB.
But then they would have to concede a lot, take a lot of flak and lose heavily in court.
I dont think they have enough humble pie left for this one.

 

[jwcalla]

I predict nvidia's going to have another blow-out quarter.

 

[Rock_Lobstah]

At what point would then a straight 4GB instead of 3.5GB + 0.5GB make any sort of difference since you're still short over 1GB on a 980...this outcry is fucking asinine. Your card still has 4GB but just doesn't pull from it until needed - with a measured performance hit of a whopping 2% or less unless playing at 4K. And if you're playing at 4K on a single 970 you're just a fucking moron to begin with.



Except 192HP is probably what you see from the drive train as the 224HP is only to the rear differential/wheelbase. And even one better than that, the 224HP is probably about 1/4 the actual power being produced by burning the fuel due to efficiency losses. Congrats, everybody is a liar, sue the whole bloody world!

You are truly a useful consumer. I bet companies all over the world wish everyone had your mindset, so they could continuously rip people off with false advertising. It's the principle of the matter. Where do you draw the line?

In business and in life you say what you mean and mean what you say. You don't take someone's money advertising one thing, and give them something different. It's as simple as that. The fact that you are attempting to rationalize and defend this error by Nvidia speaks volumes.

 

[Dayaks]

Darn you guys, how could you haven't been happy about the "feature" of the last 0.5GB of on board ram?

http://blogs.nvidia.com/blog/2015/02/24/gtx-970/

^^ Jen-Hsun's take on the issue...


AFD beat me to it

Don't care for the "feature" nonsense. I did like consumers made enough noise they might make a point not to do it again. Really contemplating a 390x this next go around.

For car analogies, this is still my favorite from page 24. I still giggle like a little girl:

This is probably the worst analogy since this discussion started. It's more like you are doing 140 and your car wants to downshift to first gear.

 

[lee0539]

At what point would then a straight 4GB instead of 3.5GB + 0.5GB make any sort of difference since you're still short over 1GB on a 980...this outcry is fucking asinine. Your card still has 4GB but just doesn't pull from it until needed - with a measured performance hit of a whopping 2% or less unless playing at 4K. And if you're playing at 4K on a single 970 you're just a fucking moron to begin with.



Except 192HP is probably what you see from the drive train as the 224HP is only to the rear differential/wheelbase. And even one better than that, the 224HP is probably about 1/4 the actual power being produced by burning the fuel due to efficiency losses. Congrats, everybody is a liar, sue the whole bloody world!

Car companies have been sued many times for false specifications and people who bought the car received some level of compensation. Everyone who bought 970 I believe should get around 20% back and if possible a free refund (no shipping cost back). Nvidia should take this loss for lying to people about the specifications of the card.

I would have rather paid 70 dollars more for a true 4 gb card. Gtx 670 was the same memory size as the bigger 80's brother and 770 was 1 gb less. These were true numbers so no one complained You knew when you bought the card it was X memory capacity. To say we should be happy they tricked us is retarded. Anyone buying the gtx 970 would have thought it was the same as previous generation.. a less powerful clocked card with the same memory capacity as advertised. So a brand new architecture now kills 0.5 gb of the capacity so we should be thankful they didn't take out the whole 1 gb?

The thing that annoys me most is that new games that run at high vram are going to stutter like shadow of modor until nvidia brings a driver to lock it at 3.5 gb.
If I'm not mistaken people on gtx 780 at 3 gb don't have those issues with shadow of modor because it is a fixed 3 g

 

[n=1]

Well to be fair they should only get a 12.5% refund.

 

[Rock_Lobstah]

By the way, just to remind folks, and from what I've gathered, it's not just the VRAM capacity that's falsely advertised. It also doesn't have the 64 ROPs or the 2048kb of L2 cache that was claimed and advertised.

As someone who bought a MSI 970 because they intend to upgrade to a 1440p monitor one day, the 4gb and 64 ROPs was one of the main selling points for me. What irks me even more about this is Jen-Hsun's lousy attempt at damage control by claiming this technical and advertising fiasco was all by design and an intended "feature". From the article: http://blogs.nvidia.com/blog/2015/02/24/gtx-970/

GTX 970 is a 4GB card. However, the upper 512MB of the additional 1GB is segmented and has reduced bandwidth. This is a good design because we were able to add an additional 1GB for GTX 970 and our software engineers can keep less frequently used data in the 512MB segment.

Really?

A) if it's a good design then how come the 980 doesn't have it? (Would be interesting to check if the GTX 960 and/or 960 Ti have this "feature".)

B) How come your primary competitor, AMD, has no problem making GPUs with bandwidth to support larger capacities like 4gb? Am I missing something here? Why does a card suddenly need this "good design" and "feature" in order to achieve 4gb?

I think at the end of the day, Nvidia realized the 970, at its original and thought-to-be specs, was going to cannibalize the over-priced 980. So, either lower the price of the 980 or gimp the 970. Pretty obvious which route they took.

 

[jwcalla]

By the way, just to remind folks, and from what I've gathered, it's not just the VRAM capacity that's falsely advertised. It also doesn't have the 64 ROPs or the 2048kb of L2 cache that was claimed and advertised.

The ROPs and L2 cache weren't advertised. And the VRAM capacity is as advertised. Just smh that people keep going with this. You could not have packed more error into your two short sentences.

As someone who bought a MSI 970 because they intend to upgrade to a 1440p monitor one day, the 4gb and 64 ROPs was one of the main selling points for me.

I refuse to believe that you bought this card because of the ROP count. You don't even know what a ROP is, just admit it.

Either way, if this card did have 64 ROPs it would make zero difference, because its performance is not ROP-limited.

A) if it's a good design then how come the 980 doesn't have it? (Would be interesting to check if the GTX 960 and/or 960 Ti have this "feature".)

The 970 is the same exact GPU as the 980 except with the defective parts turned off.

I think at the end of the day, Nvidia realized the 970, at its original and thought-to-be specs, was going to cannibalize the over-priced 980. So, either lower the price of the 980 or gimp the 970. Pretty obvious which route they took.

You mean NVIDIA intentionally made the 970 perform less than the 980? When did they start doing such things?!