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

[H3x0rc!st]

Just been reading GeForce forums (970 thread), full of childish uneducated morons. It's cringeworthy reading.

I'm happy with my 970 for now but I may think twice before buying my next graphics card, and hopefully future reviewers will look more into 100% memory utilisation without slowdowns.

 

[mattyc1977]

I'm wondering if this is what is causing all the stuttering problems with dragons age inquisition. Would be interested to know which partition physx uses.

 

[ChrisD.]

If I wasn't having problems with my gtx970, I wouldn't be reading these forums in the first place. I'm having serious problems with dragons age inquisition and shadows of modor. I'm more worried about this time next year when aaa games need over 4gb of vram. It's not so much of a problem now but fear it will be in the future.

As has already been stated in this thread, Shadows of Mordor has a 6GB VRAM requirement with everything turned up.

 

[r7slayer]

Spoiler

Now I can understand the difference.

http://www.guru3d.com/articles_pages/call_of_duty_advanced_warfare_vga_graphics_performance_benchmark_review,7.html

Well this explains it and pretty much mirrors what someone said earlier in the thread. This mystical second memory bank of 0.5gb just isn't being read by monitoring programs and any bandwidth throttling would appear to be from the user unknowingly hitting 4gb and system ram helping out. Seems plausible and likely true.

Now surely all those with tin foil hats should be directing this anger towards the creators of afterburner/gpus/anythingelsethatmonitorsvram for not being set up to read the second memory bank?

This thread was absolutely painful to read btw, half of you lot are acting like you've just caught Jen smashing your mums back doors in.

Really they claimed a drop of 3% difference in Call of Duty AW but if you look at the graph dropping from 57FPS (980) to the 45FPS (970) is certainly more than a 3% drop if my maths are correct? Looks like something is wrong to me especially when a 290 is smashing a 970 by 5 FPS when a 970 is supposedly meant to be faster.

edit - never mind it looks like ive woken up to early this morning and not read it right lol.

Still its not right though its still not 3% the 980 got 57FPS not 55

 

[mattyc1977]

As has already been stated in this thread, Shadows of Mordor has a 6GB VRAM requirement with everything turned up.

Surely not at 1080p?

 

[chef-borjan]

If that's the statement from nvidia then I'm happy. As others have said, far too many people jumped on assumptions and got riled up.

 

[stooeh]

Well I was messing about with Mordor last night. At 2160p max settings massive stutter - 4GB VRAM use showing in Afterburner. Turn down Texture Quality a notch and 3750MB max VRAM showing and no stutter.

A bigger issue is the dodgy SLI support in recent games...

 

[pgi947]

I think the question is how much slower is that 0.5GB? And does it effect performance over 3.5GB but below 4GB?

It already says in that big ol' chart above, 1-3% slower then a gtx980 with a full single 4gb bank, so margin of error stuff.

 

[Davanius]

Seems to me that Nvidia hasn't done much wrong except not disclosing how this memory allocation works from the get go. This should have been reported by the reviewers before launch so that the consumer knew exactly what they were buying.

I reckon most buyers would have bought it anyway and maybe this is how Nvidia will proceed for the next generation of cards. This will be all forgotten by then.

It will be very difficult imo to get a refund as it's a design feature unless the seller is very understanding. As I see it, the card has 4GB of ram and that the ram itself isn't slower per say, it's just separated and some software will show a speed reduction when crossing that threshold. But if software it written for it properly, than there would be no noticeable difference. I could be wrong though so can someone please explain it to me if I am.

I would have bought one if I didn't already have a ghz edition 780. Saying that, I can understand the frustration from an unhappy 970 owner.

 

[Mei]

it would be nice if some of these review sites did their own testing instead of just going by what nvidia's internal tests say, part of their job is meant to be protecting the consumers, i can hardly say that without laughing but its true!

i think atleast anandtech was a little less lame and said they cant comment until they look further? something like that

 

[Marine-RX179]

So I guess the conclusion is that for people that game on 4K (or 2560 res), avoid the 970 and get either the 290/290x or 980, but for 1920 res shouldn't be an issue for "most" games?

Still with that kind of performance hit, the 970 should really had been better off just marketed as 3.5GB...

 

[Awahwah]

As I had an existing RMA in for my Infinity Card, I've just updated it asking if OcUK have spoken to nVidia or their board partners regardining this. It's possible that they know more about this 'problem' and can give us all more info.

 

[old gamer]

So no issues with games -

none that ive noticed

 

[eyetrip]

Well, Inquisition maxed reads 3672MB in emerald graves area.

The fps was pretty constantly around 28 to 30,, but it was an absolute stutter fest.

When I brought up the perfoverlay graph though, the gpu usage was pretty flat, but the cpu usage was mental.

Not drawing any conclusions from this - I really just would like to compare the reported vram usage against someone with a 980.

 

[r7slayer]

As I had an existing RMA in for my Infinity Card, I've just updated it asking if OcUK have spoken to nVidia or their board partners regardining this. It's possible that they know more about this 'problem' and can give us all more info.

I'm sure gibbo and such are aware of this as it has a potential to hurt their sales and probably in close talk with nVidia but being who they are, nVidia are probably taking a long time to get back to OcUK which is why all OCuK guys are keeping out of it for now untill they have some official words.

 

[Twirlywoos]

Which games use texture packs above 4gb or mods? I'd like to see comparable tests not ones where Nvidia change the resolution or turn on super sampling.

 

[davida76]

Surely not at 1080p?

Correct, at 1080p SOM tops out around the magical 3.5gb number.

 

[whyscotty]

none that ive noticed

Cheers

What res do you normally game at please?

 

[r7slayer]

I'm going to call it right here.

Im going to assume seen as monitoring programs can't pick up the last 500mb of data no one can actually tell their usage when it hits over 3.5GB of data so those with stutter and such could actually be hitting the 4GB wall for all they know? And other 970 users are just getting slightly over 3.5GB and not hitting the wall, not noticing any problems in games what so ever! If there was a performance drop off from using the extra partitioned 500MB i'm going to assume its 3% difference in most cases, give or take a percentage. And whats 3% of less FPS in a game going to look like? Lol i can tell you won't tell the difference!

So there called it!

 

[Bonjour]

All's well that ends well..