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

[old gamer]

Someone reported issues with Wolfenstein.

just gave it a go and apart from sli not working the game ran fine maxed out at 2560x1440

 

[Kaapstad]

No at all... their last reported quarterly profits were $1.1 billion... extrapolate that out for the year and let's say an even $4b (it's likely more)... that would equal the return of nearly 6 million 970's... and they can only dream they sold half that many. I'm being very crude in my example of course, but yes they could afford it, if they wanted to, but this will NEVER happen. The best we'll see is some kind of firmware update, but the way things are going it would seem even THAT won't happen.


This is the key point. Clearly needs to be more definitive testing and analysis to determine EXACTLY what's going on with that extra 0.5GB. We can't just take Nvidia's own figures and explanation as gospel on this, as I don't think that's detailed enough to answer the issues others have been reporting in legit benchmarks and games... not definitively anyway. We need unbiased impartial tests to put this to bed one way or the other. At the end of the day, if it does turn out to be a significant performance hit WITHIN that 0.5GB under 4GB, this is where Nvidia have a problem... and we ALL will eventually for that matter. Bottom line, the 970 should function perfectly well utilising UP TO 4GB VRAM... if it doesn't, to any significant degree, either with a game today or in the future, therein lies the fault. Now, I haven't seen definitive evidence either way on this... so that is what I am waiting for.

You realise people in the know would take one of the better 3gb 780ti's over a 4gb 970 everytime for heavy duty work.

 

[Mei]

^^^
This.......if it's the 1 to 3% in general,that nvidia are reporting then biiiiiiiig deal.if it's around the 20% and above region comparative to a 980 then it will be a big deal .obviously frame times would be nice to see on the 1 to 3% results to see if stutter is present.im sure the tech sites are on it now

nvidia were pretty smart replying with a bunch of inhouse benchmark results, to people who skim news it looks like its the review sites coming up with these results lol

...and really thats the majority of people
got to hand it to them!

 

[Legend]

You realise people in the know would take one of the better 3gb 780ti's over a 4gb 970 everytime for heavy duty work.

What do they know that us regular folk don't?

 

[Gregster]

Here is a tip for those having issues with stuttering and usage on a 970 showing 3.5GB or over - Turn down the AA a little till the stuttering stops and you will be golden. If a game uses more than the 4GB of VRAM 'which is usable', then that is where you will see the problems of stutter.

 

[Kaapstad]

What do they know that us regular folk don't?

That you run out of GPU grunt before you run out of VRAM.

The 780ti is the faster card and manages with 3gb yet people are carrying on like it is the end of the world if a slower 970 has a slight memory problem when using over 3.5gb.

 

[Legend]

Here is a tip for those having issues with stuttering and usage on a 970 showing 3.5GB or over - Turn down the AA a little till the stuttering stops and you will be golden. If a game uses more than the 4GB of VRAM 'which is usable', then that is where you will see the problems of stutter.

Well of course, if this is what's happening then it makes this whole debacle one big joke of a fiasco... but I have YET to see definitive proof that this is what's happening. Lots of egg all over the faces of those who've run benchmarks/games that show otherwise if all the time it's simply MORE than 4GB being used... but I'm not alone in wanting to see proof of this if that's the case.

 

[Bonjour]

just gave it a go and apart from sli not working the game ran fine maxed out at 2560x1440

Good stuff.

 

[Kaapstad]

Well of course, if this is what's happening then it makes this whole debacle one big joke of a fiasco... but I have YET to see definitive proof that this is what's happening. Lots of egg all over the faces of those who've run benchmarks/games that show otherwise if all the time it's simply MORE than 4GB being used... but as I say, I'm not alone in wanting to see proof of this if that's the case.

As String along with myself have already shown that is exactly what is happening with Mordor need I say more.

 

[Gregster]

Well of course, if this is what's happening then it makes this whole debacle one big joke of a fiasco... but I have YET to see definitive proof that this is what's happening. Lots of egg all over the faces of those who've run benchmarks/games that show otherwise if all the time it's simply MORE than 4GB being used... but I'm not alone in wanting to see proof of this if that's the case.

I don't like to state the obvious but it seems some need it said. That's not saying you or anyone in particular but having read through most of this thread, it does seem like some need some basic advice.

 

[Legend]

As String along with myself have already shown that is exactly what is happening with Mordor need I say more.

Then I look forward to seeing Nvidia publishing your results and heralding you and String as tech gods amongst us feeble minded men/women, and give you some sort of medal for saving their bacon.

 

[davida76]

Monitoring programs do report the 4GB in use. I think Nvidia just worded it poorly and it's just the effect that games generally use up spare VRAM that is generally not needed (as in not affecting performance). But with the 970 it will only fill spare VRAM to 3.5GB unlike the 980 which will fill out the full 4GB. This can be seen in Mordor pretty easily.

If the game does actually need more than 3.5GB it will use it and show in monitoring programs.

I'm confused as to where this monitoring programs can't read the usage over 3.5gb has come from too, did I miss something because there' plenty of examples in the various threads on the 970 subject with screenshots showing monitoring is just fine eg:




Don't get me wrong I have no allegiance to either brand red or green I buy what is most suitable for my needs at given time.

 

[Kaapstad]

Then I look forward to seeing Nvidia publishing your results and heralding you and String as tech gods amongst us feeble minded men/women, and give you some sort of medal for saving their bacon.

No need to be sarcastic as you are only making yourself look silly.

If you do need to be sarcastic though please carry on as I don't mind in the slightest.

I would like you to do something in your next post though, try and add something constructive as this is supposed to be a debate.

 

[Legend]

I'm confused as to where this monitoring programs can't read the usage over 3.5gb has come from too, did I miss something because there' plenty of examples in the various threads on the 970 subject with screenshots showing monitoring is just fine eg:

Don't get me wrong I have no allegiance to either brand red or green I buy what is most suitable for my needs at given time.

I didn't think it was usage, it can use it, it was the performance when it hit over the 3.5GB that was the reported issue. So I thought. But if it's actually using OVER 4GB, then obviously there will be a performance hit. Seems from what some people like Kaap are saying, we've been the victim of idiots who have misinterpreted benchmark/game data and/or not realised they're actually using more VRAM than the card actually has, yet reporting otherwise. And like I say, egg all over their faces if that's true! Or not. I don't know... time will tell.

No need to be sarcastic as you are only making yourself look silly.

If you do need to be sarcastic though please carry on as I don't mind in the slightest.

I would like you to do something in your next post though, try and add something constructive as this is supposed to be a debate.

I was being silly, as indicated by the . By all means I would like to see something definitive that TOTALLY puts this to bed. Otherwise this will be a 50-page thread by mid-week, with no sign of slowing down. If you can put an end to this madness, please do, I for one would welcome it.

 

[whyscotty]

Then I look forward to seeing Nvidia publishing your results and heralding you and String as tech gods amongst us feeble minded men/women, and give you some sort of medal for saving their bacon.

Are you actually having this issue in game with you're 970's?

 

[davida76]

I didn't think it was usage, it can use it, it was the performance when it hit over the 3.5GB that was the reported issue. So I thought. But if it's actually using OVER 4GB, then obviously there will be a performance hit. Seems from what some people like Kaap are saying, we've been the victim of idiots who have misinterpreted benchmark/game data and/or not realised they're actually using more VRAM than the card actually has, yet reporting otherwise. Or not. I don't know... time will tell.

Oh I'm aware of the initial complaint re the stuttering above 3.5gb due to bandwidth drop off. I'm questioning where the people who are saying the monitoring tools are giving false readings are getting this info from, theres a few replies saying this 2 pages back or so.

 

[Kaapstad]

I'm confused as to where this monitoring programs can't read the usage over 3.5gb has come from too, did I miss something because there' plenty of examples in the various threads on the 970 subject with screenshots showing monitoring is just fine eg:



Don't get me wrong I have no allegiance to either brand red or green I buy what is most suitable for my needs at given time.

If you run both a GTX 970 and 980 on the Shadow of Mordor bench using max settings @1080p the average and min fps scale correctly showing the difference between the cards (SMX units). If you look at the VRAM usage in GPUZ it will report the 980 using the full 4gb and the 970 only using 3.4gb, this can not be right for the 970.

 

[james.miller]

I didn't think it was usage, it can use it, it was the performance when it hit over the 3.5GB that was the reported issue. So I thought. But if it's actually using OVER 4GB, then obviously there will be a performance hit. Seems from what some people like Kaap are saying, we've been the victim of idiots who have misinterpreted benchmark/game data and/or not realised they're actually using more VRAM than the card actually has, yet reporting otherwise. Or not. I don't know... time will tell.

You were all over the beginning of this thread talking about lawsuits and other nonsense 'if the ram couldnt be utilised' when it clearly could from the screenshoots of the original benchmark in the OP! Which, despite me saying it over and over, you ignored. So what does that make you, one of those people misinterpreting benchmarks?

Whilst others and I have been trying to recreate the problem with actual games, you've done nothing but scaremonger. I didnt realise you actually owned a 970 since you contributed so little to the thread. You just questioned how the thread has got so long for a second time - you do realise you've made more posts in it than anybody else, right?

 

[opethdisciple]

Ok so what we know so far, is that the way the 970 has been cut down from a 980 means that the card sees it's vram in two sections, 3.5Gb and 0.5Gb.

Also maybe monitoring programs also cannot see the 0.5Gb of vram?

Do we yet have any evidence of reduced performance when the other 0.5Gb of vram is utilized?

 

[Mei]

dont make me feel bad for being 2nd james!