970's having performance issues using 4GB Vram - Nvidia investigating

Gregster · 25 Jan 2015 at 19:01

Someone did some proper testing at Anand on a 970 and clearly you can see that the extra .5GB is being used.

dyonisis007 · 25 Jan 2015 at 19:03

Ran the SOM bench at ultra 4K a few weeks ago for reddit in relation to this issue. SLI 970s

View it here: http://youtu.be/C2-OHQGwMkw

humbug · 25 Jan 2015 at 19:07

Gregster said:
Sorry for coming across as blunt and apologies. This thread just seems to have a few people who are missing what is what or reading things completely differently to how it actually is and trying to find problems where there aren't any.

Then perhaps you should show all those who do have problems where they are going wrong, clearly you know something they don't.

Maybe start with this thread and then go to the huge thread's around the rest of the internet.

Dirk Diggler · 25 Jan 2015 at 19:10

humbug said:
Then perhaps you should show all those who do have problems where they are going wrong, clearly you know something they don't.

Maybe start with this thread and then go to the huge thread's around the rest of the internet.

What problems. What problems do these people have? Be specific.

Rroff · 25 Jan 2015 at 19:11

randal said:
So the general rule here is that:

If NVidia use a cut down version of their GPU resulting in fewer shader modules, and thusly fewer crossbar points into the memory system then they have to divide the memory into directly accessed (high speed) and indirectly (SLOW speed) chunks?

Meaning that unless NV have a major rearchitecture then this is going to be the case on anything that isn't a flagship GPU?

If so, that's a hardcore own goal.

Maxwell is likely to be relatively short lived (in terms of cards produced on it) compared to other generations anyhow, wouldn't be surprised if Pascal gets brought forward (and ramped up to full feature as and when tech makes it feasible).

humbug · 25 Jan 2015 at 19:11

String said:
What problems. What problems do these people have? Be specific.

There is more than one?

Mei · 25 Jan 2015 at 19:12

Gregster said:
Someone did some proper testing at Anand on a 970 and clearly you can see that the extra .5GB is being used.

not meaning to be picky but i dont think a 2min test proves much
ingame testing you would need to test at say 3.8gb then at below 3.5gb and then examine the frametimes, and it would need to be a long test

its maybe accessing that supposedly slow 500mb a very small amount of the time if at all?

that test could maybe prove it does atleast have access to 4gb but it doesnt test the performance of all of it

humbug · 25 Jan 2015 at 19:16

I didn't know Gregster had 970's, did you get yourself some 970's Greg?

Ian Evey · 25 Jan 2015 at 19:18

Mei said:
not meaning to be picky but i dont think a 2min test proves much
ingame testing you would need to test at say 3.8gb then at below 3.5gb and then examine the frametimes, and it would need to be a long test

its maybe accessing that supposedly slow 500mb a very small amount of the time if at all?

that test could maybe prove it does atleast have access to 4gb but it doesnt test the performance of all of it

It actually shows that there is no drop in performance when well over 3.5gb is used, Whether that be 10 seconds or 10 minutes the graph is still pretty consistent.

Stanners · 25 Jan 2015 at 19:18

Mei said:
not meaning to be picky but i dont think a 2min test proves much
ingame testing you would need to test at say 3.8gb then at below 3.5gb and then examine the frametimes, and it would need to be a long test

its maybe accessing that supposedly slow 500mb a very small amount of the time if at all?

that test could maybe prove it does atleast have access to 4gb but it doesnt test the performance of all of it

If you look at the graph when the memory usage is above 3500mb and the corresponding frame time you can see in this test it has little detrimental effect.

The time is not just a second or two either.

Gregster · 25 Jan 2015 at 19:22

Mei said:
not meaning to be picky but i dont think a 2min test proves much
ingame testing you would need to test at say 3.8gb then at below 3.5gb and then examine the frametimes, and it would need to be a long test

its maybe accessing that supposedly slow 500mb a very small amount of the time if at all?

that test could maybe prove it does atleast have access to 4gb but it doesnt test the performance of all of it

It shouldn't make a difference, as the question was "Is the last .5GB being used" and it clearly is and at no extra cost of performance.

Mei · 25 Jan 2015 at 19:24

Ian Evey said:
It actually shows that there is no drop in performance when well over 3.5gb is used, Whether that be 10 seconds or 10 minutes the graph is still pretty consistent.

ofcourse it matters because you have no reference
just because it goes over 3.5gb usage does not mean suddenly all textures are being loaded from the slow 500mb, textures are not drawn in sequentially like that

its not going to be that dramatic of a change, not while your testing all of the memory at once

atleast this is what logic tells me

Mei · 25 Jan 2015 at 19:26

Gregster said:
It shouldn't make a difference, as the question was "Is the last .5GB being used" and it clearly is and at no extra cost of performance.

you cant prove at no performance cost in a 2min test, atleast i dont believe that, if its being used yes!

Final8y · 25 Jan 2015 at 19:27

Mei said:
interesting i bet!, wish i understood half of it lol

In other words the GPU can't directly process the data in the other 500MB, data in there has to be swapped to the 3500MB part first and depending on the game it could cause performance issues.

Some games even those saying its using more than 3.5GB will only being using it for caching so it will make no difference.

If the game does really need to access all of the 4GB at once then there could be issues, also even when pushing things to the point that 4GB is used and really needed the bottleneck at the GPU can be greater than the bottleneck of the issue so again no issue is noticeable.

crinkleshoes · 25 Jan 2015 at 19:27

Buckster said:
Surely it should just be a yes/no answer

does the card have 4gb of 224gb/sec of memory

the bandwidth includes both the physical design of the card and the chip memory speeds itself

No... it does not... as made clear from comparison benchmarks.

Bonjour said:
It appears to only affect the 970 series...

http://i.imgur.com/HJHKob2.jpg

The RAM is clearly segregated at the bus.

Take a look at the bandwidth throughput. The total speed you quote sounds like the theorectical maximum... it's rare things which reach that max... best case scenario in comparison would be the GTX980 which gets 178GB/s. The GTX970 gets 150GB/s.

Card ratio:

GTX980 : GTX970
178GB/s : 150GB/s
1.187 : 1

16/16 : 13/16
1.23 : 1

256bit : 208bit
1.23 : 1

The benchmark throughput is remarkably close to the theoretical speed difference of having 3/16 of the bus width cut while keeping the same frequency.

Ian Evey · 25 Jan 2015 at 19:32

Mei said:
ofcourse it matters because you have no reference
just because it goes over 3.5gb usage does not mean suddenly all textures are being loaded from the slow 500mb, textures are not drawn in sequentially like that

its not going to be that dramatic of a change, not while your testing all of the memory at once

atleast this is what logic tells me

I am sorry but logic to me is the 970 anand used has no problem on shadow of mordor using 4gb, as shown above, its that simple.

All the people having issues with there own card should definitely contact Nvidia, all the other 970 owners that dont have any issues should carry on playing their games as maxed out as they wish.

Kaapstad · 25 Jan 2015 at 19:35

Gregster said:
What are you going on about?

Just been reminded why I don't have this on my AMD PC

The drivers are awful, hopefully there are only NVidia users in this thread.

Single 290X
SoM
1080p maxed
Stuttery mess

Edit do I start a law suit against AMD because the game runs worse on my fully functional 290X than it does on a broken 970.

Mei · 25 Jan 2015 at 19:36

Final8y said:
In other words the GPU can't directly process the data in the other 500MB, data in there has to be swapped to the 3500MB part first and depending on the game it could cause performance issues.

Some games even those saying its using more than 3.5GB will only being using it for caching so it will make no difference.

If the game does really need to access all of the 4GB at once then there could be issues, also even when pushing things to the point that 4GB is used and really indeed the bottleneck at the GPU can be greater than the bottleneck of the issue so again no issue is noticeable.

just another theory i guess?
not a bad one tho
this stuff seems very tricky to prove or disprove lol,

Mei · 25 Jan 2015 at 19:37

nvida need to be more open about it instead of "its only 3% slower trust us we wouldnt lie!"

Final8y · 25 Jan 2015 at 19:38

Mei said:
just another theory i guess?
not a bad one tho
this stuff seems very tricky to prove or disprove lol,

Indeed.