[PCPER]Frame-rating, what it is, what is does, and why it's important.

[GC_PaNzerFIN]

Yes indeed this change is for the better for every gamer regardless the brand of their hardware. Almost put tears in my eyes when NVIDIA admitted the existence of microstuttering problems in the past (GTX 295) and now this is even better, the problem child also came forth.

In all honesty I think both need shared memory between the GPUs, better frame sync (also sub-frame piece syncing, tearing!) functionality from DX side and changing the rendering method to sub-frame pieces (this has been tried before, problems with tearing and lower performance makes it less sexy to use) to completely get away with this problem. Until then it is matter of trying to smooth things on the driver side which is very tricky indeed and exactly why even AMD is having so much trouble.

I have had too many crossfire and SLi setups in the past and at some point I just went to fastest possible single GPU and haven't had a problem since. This ment decreasing image quality settings but so far I rather take that than all the multi-gpu problems. 120Hz screen(s) would love to see some multi-GPU action though as performance is not good enough even with single Titan in many cases!

Now that we finally have this reasonable conversation about the rendering issues I will gladly do my best to help in any possible way to fix the problem. After all, this is what I always wanted to come out. The truth, not some marketing BS or fanboy chatter.

I can't belive how long ago I started, back in the days when we had no real term for frametime issues like microstuttering. Now often some people get confused about terms and completely mis-use the term microstuttering which is why I don't like it anymore. It is like you say there is criminal out on the street and people start calling tips that my neighbour looks suspicious and evil although he is doing gardening like every day.

Not all problems are related to frametimes, microstuttering term suffers from popularity?
Ball is now on AMD hands. Come out clean and explain what you are doing.

 

[cadaveca]

Did you not read AMD statements.

That's from a few days ago. When speaking of Anandtech, I was speaking in terms of what they reported three months earlier, which the article you linked is a follow-up to.
from your link:

AMD has been clear with us from the start that the primary reason they even ended up in this situation is because they weren’t doing sufficient competitive analysis, and that they have revised their driver development process so that they now do this analysis to prevent future problems.

and

They’re already working on changing how they do frame pacing for multi-GPU setups, and come July we’re going to have the chance to see the results of AMD’s latest efforts there.

So...what was your point?

That's not a public statement. That's Anandtech reporting what AMD told THEM, not AMD making a public statement.

I mean, really, that kinda settles all your questions in this thread right there, problems with testing or not. IF you want to call that AMD's public statement, then why are you trying to raise issues with the information provided?

AMD has been clear with us from the start that the primary reason they even ended up in this situation is because they weren’t doing sufficient competitive analysis

Because really.. it's just that.. it's info. In January AMD said they'd have a fix in March. They don't, and now it's April. In the meantime, here's some more info so you understand why it's taking so long, and why the issue is complex.


There's nothing new here, no conspiracy.

 

[Xzibit]

Here you go for the visualy impared.

I drew Xs to help your eyes out.

Additional lights that turn on

 

[cadaveca]

Here you go for the visualy impared.

I drew Xs to help your eyes out.

Nobody cares.

AMD has been clear with us from the start that the primary reason they even ended up in this situation is because they weren’t doing sufficient competitive analysis

Who cares about Ryan's videos? They are merely examples, not perfection...that's up to AMD's driver team to deal with. You're making a big deal out of nothing.

 

[Xzibit]

Nobody cares.


Who cares about Ryan's videos?

The whole purpose of benching is limiting influences if your not gonna bother doing that dont bench at all

The workload is different so offload is different to GPU sheesh

Really your asking that question.

 

[cadaveca]

Really your asking that question.

Sure. All Ryan is illustrating is what AMD is doing to fix things, and the types of testing they now use. In that, the videos are fine.

For somebody claiming to want to ensure the data is interpreted properly, you sure seem to have a completely different agenda. Really, you do seem to be claiming this teting is flawed, there are other issues, people are not looking at the right things, blah blah...


It's not meant to present fine details. I also didn't link his videos.. I linked his explanation of the testing hardware and what it shows. The rest is stuff the end user shouldn't have to worry about, and AMD has stated that how this ends up is going to be different for each user, so they are going to present different options, even.


AMD's already said the testing is right, and you're here months later trying to point out problems with it. Just saying, man.

 

[Xzibit]

For somebody claiming to want to ensure the data is interpreted properly, you sure seem to have a completely different agenda.

Which is what ?

I wants to know if the Data deing spit out by the game engine is equal in-both cases. If the 2 GPUs in question are processing same amount of data per frame. You can even throw in APUs.

Since FRAPS and FCAT take and insert frame info at the same spot before it reaches Direct X.

Then what ? Should we ignore that too ? Hope you dont conduct your own test like that.

Asking questions isnt a bad thing. Ignoring Answers is.

 

[cadaveca]

Ignoring Answers is.

Exactly. And you've ignored that this isn't an answer.


It's merely a way to effectively convey to the end user what's happening, and what all this "stutter" that people like me complain about. For years people have said they don't see it, it doesn't exist, blah blah blah...so here's a way it can be presented so that any user can see it. the flipping colors make it pretty obvious.

You are looking at image quality, when that's not what this is about. This is about the smoothness of the animation presented to the end user, and in those videos...the only important thing there is the motion of the character's arms.

It is clearly smoother in NVidia's implementation.


How or why it's smoother...is the question. The quality of the image...may have something to do with that...sure, but that is but one of many things can lead to problems.


And that's why none of this will change how VGAs are tested and reviewed. The second card in Crossfire IS working...it's simply not displaying it's image properly, so it appears that it's useless. It's not useless, though, it's merely out of sync, and in such a way that V-sync doesn't fix the issue.


AMD has stated, overall, that this is due to a memory management problem in the driver. A fix was expected in March, and we were given the first 13.1 beta.


Now it's march, there's no driver, so here's why. The problem is complex, has many facets, and takes time to manage. AMD has asked for four more months. I stated, before any of this "new" stuff went public, that I was expecting four more months. Go figure.

 

[DRDNA]

Xzibit, I also wonder about the visuals that are rendered to the end users monitor, I have wondered if both camps with equal settings do they show different things and if one camp may be suspect of not letting their card display everything like cutting out lights to get better render times. I have my opinion about that with the camps but is indeed a different topic that has some relations.

 

[Xzibit]

Xzibit, I also wonder about the visuals that are rendered to the end users monitor, I have wondered if both camps with equal settings do they show different things and if one camp may be suspect of not letting their card display everything like cutting out lights to get better render times. I have my opinion about that with the camps but is indeed a different topic that has some relations.

Thats what peaked my interest.

Like I said I dont have a SLI or Crossfire setup in either my Nvidia or AMD systems. Since those test arent limited to Multi-Card configurations. Hes also testing FCAT with single cards thats what i'm looking at myself.

People have noted discrepancies in the videos hes put out with the LOD differences.

It seams he did post the write-up

I'm interested to know if either 7000/600 series is doing more/less work for the benched results.

 

[the54thvoid]

Whether or not what card is doing more work is still not the point. The point is whether the delivery of the visual image is smoother or not.

Something people ought to be aware of as well is that AMD can deal with quite complex lighting (Dirt Showdown with global lighting destroys Nvidia cards because the lighting coding was developed from an AMD showcase of lighting for that game). Sleeping Dogs may use similar lighting effects (AMD title) which would help explain the differences you are highlighting.

The GCN architecture is fantastic. And in many respects is better (and sometimes worse) than Nvidia's Kepler. However, what Dave is so patiently putting across is not the difference in appearance of quality but the smoothness of the image.

And like he's said, AMD acknowledged this following the initial latency storm and then again in a discussion with Anandtech.

There isn't an argument about that to be had here. Your point is for another thread - who has better IQ. Certainly in your pics - it's AMD.

 

[GC_PaNzerFIN]

http://www.pcper.com/reviews/Graphi...ils-Capture-based-Graphics-Performance-Test-4

Take a look at the amount of only partially rendered frames with CF. Only thing they do is increase FPS although barely anything makes it to the screen.

 

[BiggieShady]

This image from guru3d's article illustrates the problem with muilti gpu setups. AMD is simply delivering frames as fast as it can and as soon as it can. nVidia is syncing and delaying frames.

 

[radrok]

Sadly you can't do much magic with AFR, you can just improve how smooth it presents.

I think that improving bandwidth between GPUs and making them render parts of the screen instead of 1x1x1x1 would be better.

 

[BiggieShady]

Sadly you can't do much magic with AFR, you can just improve how smooth it presents.

I think that improving bandwidth between GPUs and making them render parts of the screen instead of 1x1x1x1 would be better.

Problem with that approach is there would be much unnecessary duplicated work in a single frame on both GPUs. With alternate frames each GPU does separate work for its own frame only VRAM is duplicated.

Processing parts of the screen for different lighting passes in a deferred engine did prove to be a good idea, if I'm not mistaken Dice did it in Frostbite 2 engine using direct compute.

 

[Xzibit]

This image from guru3d's article illustrates the problem with muilti gpu setups. AMD is simply delivering frames as fast as it can and as soon as it can. nVidia is syncing and delaying frames.

http://www.guru3d.com/index.php?ct=articles&action=file&id=3108

I'm curious to know if this is forward thinking on Nvidias part for Cloud, Tegra, Shield Project (Remote Gaming)
Spacing out frames would mean less work for the GPU with less variance less noticable until you start pumping more data through and we arent there just yet.

Would it add overal game latency ?
Wouldnt more demanding games see higher variance and at higher resolutions ?

 

[cadaveca]

Would it add overal game latency ?
Wouldnt more demanding games see higher variance and at higher resolutions ?

Possibly, but I'm pretty sure that that was why G80 has separate I/O chip specifically to handle that load and minimize any latency that might be introduced. Eventually that silicon was integrated directly into the GPU itself.


AMD simply killed the AIW line, and that's truly when these problems started for me, personally. ATI used to have awesome capture and display hardware AND software.

And, that visual representation was covered by other articles already. We already know(or at least I know) that the issue is a lack of sync of the frames from the secondary card. AMD is claiming that the "runts" are these frames, and that's why the second card seems useless.

 

[brandonwh64]

I figured this would turn into an arguement somewere along the line. Dave is right with this article he linked us cause I am just now noticing things that my Xfire was doing that vanished when I went single 7970.

I know dave and he would not have posted this if he has not seen the effects of it himself.

 

[cadaveca]

I figured this would turn into an arguement somewere along the line. Dave is right with this article he linked us cause I am just now noticing things that my Xfire was doing that vanished when I went single 7970.

I know dave and he would not have posted this if he has not seen the effects of it himself.

Making the move from dual 6950/6970, to single 7970, where the performance is nearly identical, really does present things very differently than any other "upgrade" before.

Considering that the number of "shaders" has been reduced as well, I can't help but feel that it is hard to get an accurate idea of the full picture without actually having the hardware in hand, and swapping between the two configurations. It can be difficult to separate what's changed due to moving to a single GPU, and what's due to the more efficient GPU design.

But, at the same time, I hardly see this is as a movement that will change how reviews are done.


In the end, sure, I noticed this a long time ago, and yes, I have talked about it to countless people. Maybe two or three knew right away what I was talking about, while many more had similar configs, so it's always been clear to me that this is an issue that not everyone is really going to be affected by, number one because many simply don't do multi-GPU, and secondly, not everyone is going to be as sensitive to this as I am. The one thing that you have, Brandon, that others don't, is out countless conversations in TeamSpeak about it. You can now check and see if you see any of the things I did, now that you're going down that exact road I did.

Enjoy the trip.

 

[Xzibit]

Possibly, but I'm pretty sure that that was why G80 has separate I/O chip specifically to handle that load and minimize any latency that might be introduced. Eventually that silicon was integrated directly into the GPU itself.


AMD simply killed the AIW line, and that's truly when these problems started for me, personally. ATI used to have awesome capture and display hardware AND software.

And, that visual representation was covered by other articles already. We already know(or at least I know) that the issue is a lack of sync of the frames from the secondary card. AMD is claiming that the "runts" are these frames, and that's why the second card seems useless.

Yes.

I was more looking at the Single solution comparison and the difference on the Nvidia side with Single and SLI.

The Mid-end (Lower-end still to come) GPUs results seams more interesting

You'd expect when you add another card into the mix to lower the variance or maintain that of a single but it still increases. Once you go above 1440p resolution its seams its a question of how much you want to tolerate.

These results just deter multi-gpu solutions.

 

[cadaveca]

These results just deter multi-gpu solutions.

Perhaps. For sure, right now, on the AMD side of things. Perhaps some of the tech and know-how from buying 3dfx years ago has proved valuable here for Nvidia when dealing with this stuff.


I know, for sure, that ADM is trying to get this fixed. For me, it's very obvious, since some drivers "stutter" more than others, those that don't seem to be missing all benefits from the second card, and those in between tend to give me headaches or make me feel ill.


None of that every happens in single-GPU.


When moving up resolutions, you have to expect that the higher workload is going to affect the timings between frames. That's a given. You can really only accurately compare different resolutions, for this "investigation", if the performance offered by each is similar. You can't ignore that the workload is different.

If a GUP rendered 1080p and 1600p with the same level of performance, that's fine, but it should be quite clear to most users, by now, that there are different cards from each brand for a reason...not because they offer different performance...they are offerings for users running different resolutions.

You don't need a 7970 to run 720p, and trying to game @ 1600p on a 7770 is clearly a mistake. As far as I am concerned, the industry's lack of assigning GPUs to a specific resolution really hampers the ability of PC makers to provide well built and well-balanced configurations. There should be set "rules" for system configs to meet certain performance levels in certain resolutions.

 

[Xzibit]

Even the 7970 vs 680 comparison

@ 1080p
Going SLI seams only worth it to lower FT
At the price of adding another card at full price ?

@ 1440p
Going SLI looses its value by 50% FT slightly lowers and V rises.
Maybe if there was a SLI with a secondary card at 50% of the price with no outputs just SLI connectors.
I recall 3Dfx Voodoo series had a setup like that. Where you bought secondary card slightly cheaper.

@ 5760x1080
The only game that ran "remotely" decent was Dirt 3

Seams like just marketing to get you to buy an additional card for the full price.

 

[EarthDog]

Yeah, all of you that spent money on a second card for Crossfire, did it for nothing...really nothing...for now.

This is funny. I dont even notice with my 690. .. oh wait.......

 

[Vario]

I came close to pulling the trigger on a second 7850... glad I didn't. I don't think its really optimal to run either crossfire or sli, it seems like theres too many inherent problems, especially for AMD. Hopefully AMD will get a fix on this.

 

[qubit]

This is funny. I dont even notice with my 690. .. oh wait.......

690? Good choice there, dude.

Ya know what I've got? A pair of Asus HD 3850X2 cards I got cheap a few years ago. At the time, I ran them together and singly and in neither case did the performance even come close to my GTX 285. That's four AMD GPUs getting spanked by one NVIDIA GPU, lol.

Compare them to my GTX 590 and it's total humiliation.

What chance do you think that this runt frame problem was present even then? Shame I can't run those advanced tests and find out.