[FALSE ALARM] GTX 980 Ti 5.5GB VRAMgate Reloaded?! WTF

[skullbringer]

I did the testing wrong and therefore my results were wrong. How to do it right:

 

Here you go. As already explained if it's not in headless mode, it means VRAM is always allocated and used, so that ram will always be slower than the rest when tested as it's already in use.

 

http://http.developer.nvidia.com/ParallelNsight/2.1/Documentation/UserGuide/HTML/Content/Setup_Local_Headless_Debugging.htm

 

This is my correct result:

 

So no VRAMgate with 980 Ti.

 

Sorry for causing any confusion!

 

 

 

 

 

 

Ok, so another forum member just messaged me, asking what my experiences with the GTX 980 Ti were: "What I want to know is, are you limited to utilizing only 5.5 GB or can you use a full 6.1 GB. What's your FPS in the Witcher? Which card do you have?"

 

I kinda lol'd at the thought that there might be a VRAM restriction like with GTX 970, where one eighth of the VRAM is attached serially and therefore runs much slower than the rest.

 

So I just ran the test (You can download it here from Guru3d.) and this is what I got:

 

 

 

To be honest, I dont know what to think right now... GTX 980Ti was advertised to have 6GB of VRAM, but it does not really?!

 

With GTX 970 Nvidia admitted it to be an error in their marketing:

 

NVIDIA says this was an error in the reviewer's guide and a misunderstanding between the engineering team and the technical PR team on how the architecture itself functioned.

 

So did I get something wrong here or was there an internal "misunderstanding" at Nvidia again?! And why did nobody notice this earlier?!

 

 

 

Sources:

http://www.gamespot.com/articles/nvidia-admits-to-error-in-gtx-970-specs-and-memory/1100-6424915/

http://www.guru3d.com/news-story/does-the-geforce-gtx-970-have-a-memory-allocation-bug.html

[NeatSquidYT]

How do you do this test btw?

[Glenwing]

As noted in the quote you quoted, it's part of how the architecture functions. 980 Ti is still Maxwell architecture, so if you disable SMMs like on the 970, you'll get a VRAM configuration like on the 970. This is kind of expected TBH. The only thing I'm surprised about is that apparently nobody decided to check for this on the launch day reviews.

[skullbringer]

How do you do this test btw?

added link

[beach_boy98]

That looks pretty convincing. Is it to do with the Gm200 core in the 980ti not being fully enabled the GM204 in the 970?

[ZetZet]

As noted in the quote you quoted, it's part of how the architecture functions. 980 Ti is still Maxwell architecture, so if you disable SMMs like on the 970, you'll get a VRAM configuration like on the 970.

why doesn't 980 do this then?

[Glenwing]

why doesn't 980 do this then?

 

Why would it? The 980 doesn't have any SMMs disabled.

[NeatSquidYT]

added link

Thanks.

 

*GASP*

 

THE 970 HAS LOW PERFORMING VRAM

 

[Glenwing]

Also, this makes me question the Titan X's VRAM.  

 

The TITAN X doesn't have any SMMs disabled, so it doesn't have a memory configuration like this.

[ZetZet]

Why would it? The 980 doesn't have any SMMs disabled.

Oh it's related to that.

[Sunako]

Where is a source for this with the 980 ti you have only showing old articles on the 970

[itsmyjobtoknow]

If this is true, and hopefully its not...then we have a trend here.

 

Notice how the 980 launched and the one tier down card which is the 970 had the .5GB issue.

 

And now with the Titan X, the one tier down card is the 980 Ti....and it have the same issue as the 970 did....

 

Are they done on purpose or is it the fault on how NVidia's method of cutting down the Maxwell core of each respective flagship. 

[UnknownEngineer]

Also, this makes me question the Titan X's VRAM.  

But that one doesn't have any SMM's removed.

[skullbringer]

As noted in the quote you quoted, it's part of how the architecture functions. 980 Ti is still Maxwell architecture, so if you disable SMMs like on the 970, you'll get a VRAM configuration like on the 970.

But why would Nvidia do that?

With 970 and 980 the amount of RAM is the same so thats why there has to be a downgrade somewhere else.

 

980Ti already has half the RAM, so why downgrade it again by disabling SMMs??

 

And why did Nvidia not tell us?!?!?!?!

[asim1999]

Also, this makes me question the Titan X's VRAM.  

The Titan X has a full core. JayzTwoCents did the test, and all 12GB worked properly 

[Fetzie]

Where is a source for this with the 980 ti you have only showing old articles on the 970

 

His command line output picture in the OP clearly states it is running on a GTX 980ti

[Misanthrope]

Oh come on! Not this again. Look I sorta saw the point on 3.5 vs 4 but I seriously doubt you will ever get to use 5, let alone 6. There might be a case scenario for SLI that maybe gets to 6gb vram in the future but in the future DX12 will be able to pool all ram so you'll get 11 vs 12, still plenty.

[asim1999]

If this is true, and hopefully its not...then we have a trend here.

 

Notice how the 980 launched and the one tier down card which is the 970 had the .5GB issue.

 

And now with the Titan X, the one tier down card is the 980 Ti....and it have the same issue as the 970 did....

 

Are they done on purpose or is it the fault on how NVidia's method of cutting down the Maxwell core of each respective flagship. 

Nvidia's manufacturing issue.

Now i am thinking about it, the amount of ROPs (96) may not be correct, probably has 88

[Glenwing]

But why would Nvidia do that?

With 970 and 980 the amount of RAM is the same so thats why there has to be a downgrade somewhere else.

 

980Ti already has half the RAM, so why downgrade it again by disabling SMMs??

 

And why did Nvidia not tell us?!?!?!?!

 

Because it would make the TITAN X even more irrelevant than it is already. Realistically, no gamer is going to care about the difference between 6GB and 12GB of VRAM, so that's not a downside for these customers. NVIDIA has to do something that actually affects performance if they want to differentiate the product.

[Fetzie]

Oh come on! Not this again. Look I sorta saw the point on 3.5 vs 4 but I seriously doubt you will ever get to use 5, let alone 6. There might be a case scenario for SLI that maybe gets to 6gb vram in the future but in the future DX12 will be able to pool all ram so you'll get 11 vs 12, still plenty.

 

Some reviewers noted different performance values in certain games to other reviewers. Could be that they hit this restriction?

 

Although while I can see 4GB of VRAM being a bottleneck, especially when games load every texture they can into the VRAM even if it isn't going to be needed, 5.5 or 6 really isn't.

[Analog]

My main issue is how the benchmark just floods the card with data on the expectation no slowdown will occur. I'm pretty sure if you had the same kind of benchmark for an AMD card that you would notice similar slowdowns.

 

The slowdowns occur simply because that VRAM is slower. There is absolutely no other reason for it. 

[henrikfraannk]

Hmm.. but why does my GTX 980 do the same thing? Is my GTX 980 broke? http://tinypic.com/r/b5gor9/8

[itsmyjobtoknow]

Nvidia's manufacturing issue.

Now i am thinking about it, the amount of ROPs (96) may not be correct, probably has 88

 

But wouldn't they have fixed this issue knowing they messed up once already...?

[Misanthrope]

Some reviewers noted different performance values in certain games to other reviewers. Could be that they hit this restriction?

 

Although while I can see 4GB of VRAM being a bottleneck, especially when games load every texture they can into the VRAM even if it isn't going to be needed, 5.5 or 6 really isn't.

 

With any game out right now? I seriously doubt that unless you're doing something stupid like triple 4k monitors while still having all detail on ultra on GTA V but I'm pretty sure you'd get to GPU limit way before vram limiting performance so it would be a non issue.

[Sunako]

http://i.imgur.com/nnkcC4e.png

this is from  2 mins ago run on my own system i will be doing more in just a moment that should be card 1 of 2