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

[Kaapstad]

I would love to keep the 970's but if they're killing performance at 4k because of this memory thing then I don't want them.

If they were choking due to lack of horse power then fair enough, no complaints from me, but I didn't drop over £600 on a pair of 970's with water blocks just to be robbed of performance at 4k because of a hardware limitation that Nvidia neglected to tell anyone about.

You realise that the 980s also suffer from a serious dip in performance @2160p.

 

[bru]

Surely if testing AMD cards it would have to the 290p, 280(non X) etc. the cut down cards. Actually anyone know it the 750Ti and 750 are showing any of the issues in anyway?

 

[almighty15]

I will say the same to you as I said to whoever it was earlier, in fact quoting it might be easier.

The earlier furmark reference is that when running furmark cards can throttle below their stated base clock speed, so by your reckoning they are all faulty.[/QUOTE]

The memory clock isn't as dynamic as the core clock and base bandwidth rates while not 100% achievable in the real world should be the same speed regardless of what partition is being read and written too.

So your post is wrong.

 

[humbug]

IF the buffer splitting is causing the slowdown / Stuttering, what can Nvidia do to fix it?

 

[Kipz]

I will say the same to you as I said to whoever it was earlier, in fact quoting it might be easier.



The earlier furmark reference is that when running furmark cards can throttle below their stated base clock speed, so by your reckoning they are all faulty.

This is the deciding factor but even with the up to 224Gb/s that point is still moot because it cannot perform this at the full 4GB. It's similar to saying that the core clock is 1200Mh/z but it can only achieve that speed if the core usage is under 80% and once it goes over that it clocks down to 700Mh/z. I'm sure that people would be just as upset if not more.

 

[almighty15]

You realise that the 980s also suffer from a serious dip in performance @2160p.

Dips? Yes, Stutters? No..

Anyway, we'll see what new information when it gets put up later.

 

[ALXAndy]

I use 980s and 290Xs

I can not say about the 970 but the 980s do have a serious performance dip @4K.

@1080p the 980s are in front of the 290Xs but @2160p in multi GPU setups the 290Xs are slightly faster.

Let's face it, everything currently on the market has a serious performance dip @ 4k. It's what's putting me off.

I read a review once comparing GPU set ups in SLI and Crossfire for 4k, and in pretty much every single test the 290s won at 4k. IIRC it was only F1 2013 that the 970s won. I just didn't want either hot stock cards or after market because they're poo in Crossfire/SLI. My 670 Jetstreams used to suffer really badly when stressed. Quite often saw the top card reach over 90c and for Kepler that wasn't good. Ended up having to seriously mod my case and stick two 120mm fans aiming right at them to get them back into the 80s and I still wasn't happy about that.

Sadly there were no stock blower versions of the 6gb 780 or I'd have gone with a pair of those.

 

[stooeh]

What games can I test these stutters apart from Mordor?

 

[humbug]

Surely if testing AMD cards it would have to the 290p, 280(non X) etc. the cut down cards. Actually anyone know it the 750Ti and 750 are showing any of the issues in anyway?

I already said my 290 (None X) does not slow down when (And will) use all of its 4GB buffer.

 

[ALXAndy]

IF the buffer splitting is causing the slowdown / Stuttering, what can Nvidia do to fix it?

If it is how the card is designed to work, like, mechanically? there's not a lot they can do.

Fact is that memory speed and bandwidth on that second chunk is just pure ass.

But hey, you never know, they might be able to fix it with a bios update. I strongly doubt that though, as they've basically come out and said yeah, the 970 uses its memory differently to the 980 and have posted a load of benchmarks showing games running up to 3.5gb but not more to try and worm out of it.

What games can I test these stutters apart from Mordor?

Any game that you can crank on and get to use over 3.5gb. Only issue? they're mostly all UBI crap that run horribly already.

 

[Ian Evey]

I already said my 290 (None X) does not slow down when (And will) use all of its 4GB buffer.

Have you tried it in Shadow of Mordor? Or are you just saying your 290 does not slow down when it uses all of its 4gb so your 290 is the same as a 970 in that case, unless you ran shadow of mordor.

 

[Tonester0011]

> that sign is more than. Their tests show margin of error difference in performance, also known as nothing to moan/cry/ whinge about.

Unless you're looking for free cake.

 

[almighty15]

What games can I test these stutters apart from Mordor?

Ryse can be beastly with the image quality cranked up.

Crysis 3 with IQ turned up

Crysis 2 with texture packs

Half Life 2 with cinematic MOD

STALKER with the complete mods

There's loads of games.

 

[humbug]

Have you tried it in Shadow of Mordor? Or are you just saying your 290 does not slow down when it uses all of its 4gb in other games like the 970 does?

I don't have Shadow of Mordor, i do have CoD AW.

 

[Tonester0011]

Wouldn't expect bug's cards to be anything other than perfect truth be told.

 

[ALXAndy]

You realise that the 980s also suffer from a serious dip in performance @2160p.

Everything does, dude. I ran the MSI VRAM test on a Titan and went from 260+ FPS at 1gb VRAM to 90 FPS when using it all.

Memory bandwidth is a serious issue that people have taken for granted because they don't think they need it. As soon as the new consoles came out and we garnered the layout (8gb, 2gb for the system the rest for graphics) it was pretty much written on the wall that you were going to need more VRAM.

If a console has 6gb on tap then the devs are going to work their hardest to make sure they utilise it, and the "prettiest game" war is well and truly on.

If there's ever been any gaming device in history that gets serious work and shed loads of optimisation it's the consoles. Mostly because they're a set architecture and don't change, so it gives devs time to work out how hard they can push it.

 

[Ian Evey]

I don't have Shadow of Mordor, i do have CoD AW.

Oh, Has someone said that they have cod aw and it slows down using a 970?

 

[Kaapstad]

Dips? Yes, Stutters? No..

Try running FC4 on some 980s, it is not a nice experience as the guy in the other thread has pointed out.

I actually bought FC4 not to play it but to see how awful it was.

I think people are mixing up an awful lot of things and blaming the 970 for all of them. Some of these would be -

Poor performance on the 9 series cards at high resolution (both the 970 and 980).

Rotten games like FC4 that are very badly written.

Dodgy benching software that can not detect what is really there.

Way too much negative speculating without solid evidence to back it up.

 

[almighty15]

Try running FC4 on some 980s, it is not a nice experience as the guy in the other thread has pointed out.

I actually bought FC4 not to play it but to see how awful it was.

I think people are mixing up an awful lot of things and blaming the 970 for all of them. Some of these would be -

Poor performance on the 9 series cards at high resolution (both the 970 and 980).

Rotten games like FC4 that are very badly written.

Dodgy benching software that can not detect what is really there.

Way too much negative speculating without solid evidence to back it up.

I know what you're saying but I can get stuttering in games that aren't really that demanding ALU wise, they just use big textures through mods so when I do get stuttering the GPU use is quite low, as is the CPU and system RAM use.

 

[Mei]

it might be just the kick up the butt nvidia needs to release some better stuff, instead of cost cutting their way to glory
so really any nvidia fans thinking its an atk on them its very silly when they could end up seeing better stuff as a result

be it hardware or drivers

just saying, dont kill me!