new AMD Catalyst driver from Alpha Micro stuttering

[ICDP]

According to THG's review, which has gamers test their systems, without knowing which was which, people still can tell the difference. Though they were not using the prototype driver. Later, when they tested the prototype driver, they noted an improvement.

So we wait for the prototype to be released and compare again. They should improve it more by then. In the THG and Pcper articles, it still had a lot more variance than Nvidia's frame metering, but definitely better than before.

I don't know who THG are sorry, can you enlighten me? Also if they didn't carry out the blind tests using vsync or FPS caps then of course it would be possible to tell the difference. CF without these fixes has serious and unmissable micro stutter.

 

[Black Octagon]

My apologies, of course that is true. My point is that at some point MS is indeed to all intents and purposes, gone and I posted graphs to back that claim up. Now many may counter claim that maybe I have more tolerance and I only think it is fixed. That is why I called it ironic to ask for tangible data only to dismiss that tangible data with a flipant, "Maybe different people just have different levels of sensitivity to microstutter".

ICDP, I was not being flippant when I said that. Your frame time graphs are appreciated, but you took them with FRAPS, meaning that the results only show the beginning of the rendering pipeline...and this is something that AT and other sites have clearly demonstrated to be problematic for capturing what the end user is likely to experience.

If you can show me the same smoothing of the frames (thanks to vsync) using FCAT or something equivalent then I will reconsider the possibility that I somehow wasn't doing the 'fix' properly. Until then, my working theory is that your Crossfire setup is microstuttering, but that you are less sensitive to it than I (and others).

 

[loccothan]

http://www.guru3d.com/articles_pages/radeon_hd_7990_review,23.html

 

[ICDP]

ICDP, I was not being flippant when I said that. Your frame time graphs are appreciated, but you took them with FRAPS, meaning that the results only show the beginning of the rendering pipeline...and this is something that AT and other sites have clearly demonstrated to be problematic for capturing what the end user is likely to experience.

They may have shown FRAP graphs to be problematic but they haven't shown it to be worthless. If you get a perfectly even frame time at the start of the rendering pipeline and it looks smooth on the screen (the end of the rendering pipeline), then it doesn't need further testing to conclude that it is in fact smooth.

If you can show me the same smoothing of the frames (thanks to vsync) using FCAT or something equivalent then I will reconsider the possibility that I somehow wasn't doing the 'fix' properly. Until then, my working theory is that your Crossfire setup is microstuttering, but that you are less sensitive to it than I (and others).

It isn't a working theory, it's an assumption based on arrogance and it doesn't even deserve to be called a theory. The old "your getting stutter, you just don't see it but I bet I would" reply. Once again, how do you know I don't have a worse tolerance for micro stutter than you? How about I reply with an arrogant and flippant, "you just weren't doing it right"?

 

[bystander36]

I don't know who THG are sorry, can you enlighten me? Also if they didn't carry out the blind tests using vsync or FPS caps then of course it would be possible to tell the difference. CF without these fixes has serious and unmissable micro stutter.

THG is short for Tom's Hardware Group. Here is the test they did within the 7990 review, that includes runt frame analysis and frame variance. AMD's prototype drivers helps a lot with the runt frames, though a runt frame is only 20 lines or less, it could be made larger, but the frame variance still has some work to do.

http://www.tomshardware.com/reviews/radeon-hd-7990-review-benchmark,3486-13.html
(That shows the gamer testers who played two machines without being told which was which. They all preferred the 690 btw.)

 

[Black Octagon]

How about I reply with an arrogant and flippant, "you just weren't doing it right"?

You could do that...and judging from how I seem to have pissed you off I daresay that's what you're trying to say without actually saying it.

But I was 'doing' it. RadeonPro, preview version. Vsync (normal and dynamic). Frame rate limiter. Triple buffering. Flip queue sizes. In pretty much every combination I could think of. Plus a multitude of different driver versions. Not to mention various other 'fixes' being touted on internet forums. I'm not a technical expert, but I do know what I'm doing, and I am thorough. I also was on something of a mission to defend AMD as I was sick of the Kepler hit squad's nonsense through 2012. No...dice.

I'm not exactly the first person to complain about it, and as already stated the review sites seem convinced that microstutter is still an insurmountable issue affecting Crossfire.

I'm delighted that your eyes consider the issue fixed. I truly am. But I know what I experienced and others experience it too. Not much else to say about it really.

 

[Silverforce11]

Some people complaint about microstutter and or input lag with SLI as well, hence, they tend to stay way from multi-gpus.

Its no stretch of the imagination to assume there are those who perceive the world minutely different to oneself.

 

[ICDP]

THG is short for Tom's Hardware Group. Here is the test they did within the 7990 review, that includes runt frame analysis and frame variance. AMD's prototype drivers helps a lot with the runt frames, though a runt frame is only 20 lines or less, it could be made larger, but the frame variance still has some work to do.

http://www.tomshardware.com/reviews/radeon-hd-7990-review-benchmark,3486-13.html
(That shows the gamer testers who played two machines without being told which was which. They all preferred the 690 btw.)

Ah thanks for the clarification. I'm not surprised they picked the GTX690, AMD still has a lot of catching up to do.

 

[railven]

If you can show me the same smoothing of the frames (thanks to vsync) using FCAT or something equivalent then I will reconsider the possibility that I somehow wasn't doing the 'fix' properly. Until then, my working theory is that your Crossfire setup is microstuttering, but that you are less sensitive to it than I (and others).

V-sync helped, a lot, even to the point that the guys using FCAT the longest and beating the drum the loudest even acknowledge it (of course with a huge asterisk about input latency.)

http://www.pcper.com/reviews/Graphi...ils-Capture-based-Graphics-Performance-Tes-11

For both NVIDIA and AMD multi-GPU solutions with standard Vsync, enabling it definitely changes the story. NVIDIA’s cards pretty much perform as we expected but for CrossFire we didn’t really know what expect with the various visual concerns. It does appear that the runts problem was at least mostly solved with the enabling of Vsync though to be clear we are only testing a couple of game at this point – much more needs to be done.

Of course, the new driver is even better, but this is just reviewer proof of what some people have been saying - Vsync seems to help, a lot.

I've yet to encounter any stuttering in the games I play beside WoW, but that stuttered with 1 card and it stutters on all my systems (GTX 680/GTX 660 Ti) it shows slight stutter in the HD 5770 system I was just tinkering with it.

I'm starting to wonder the WoW issue isn't driver/hardware related and game engine related. That's 4 cards I've experienced the stuttering in. When I build up the GTX 460 system I plan on giving to someone, I will test WoW on it.

I wonder if it's an add-on...never tested with all add-ons disabled.

 

[ICDP]

You could do that...and judging from how I seem to have pissed you off I daresay that's what you're trying to say without actually saying it.

I was being crass deliberately to demonstrate how your, "Maybe different people just have different levels of sensitivity to microstutter" quote comes across in exactly the same arrogant way. I inferred that your statement was a thinly veiled and arrogant assumption that I'm seeing stutter but I just don't know it. Thanks for playing along

But I was 'doing' it. RadeonPro, preview version. Vsync (normal and dynamic). Frame rate limiter. Triple buffering. Flip queue sizes. In pretty much every combination I could think of. Plus a multitude of different driver versions. Not to mention various other 'fixes' being touted on internet forums. I'm not a technical expert, but I do know what I'm doing, and I am thorough. I also was on something of a mission to defend AMD as I was sick of the Kepler hit squad's nonsense through 2012. No...dice.

I don't doubt for one second that you tried all those things, what I was pointing with my graphs was that it is fixable and at some point MS becomes totally fixed. If you think an almost totally flat line in FRAFS does not equal totally smooth on the screen then no amount of tangible data will change your mind, FCAT or otherwise.

I'm not exactly the first person to complain about it, and as already stated the review sites seem convinced that microstutter is still an insurmountable issue affecting Crossfire.

It is a major problem but it is not insurmountable, I have shown graphs that prove it. You can't have it both ways by saying show me tangible data and then discount any tangible data when presented. Do you have FRAPS data from your own testing so we can compare?

I'm delighted that your eyes consider the issue fixed. I truly am. But I know what I experienced and others experience it too. Not much else to say about it really.

Again we are back to the question, how do you know I don't have worse tolerance for micro stutter but just am better at fixing it than you? It is arrogant to assume either position without testing, so simply claiming "your eyes are worse than mine" is both flippant and arrogant in the extreme.

Answer me this, if my eyes could not see the MS you claim is still present when I say it isn't. How come CF with the Tomb Raider settings I posted is far, far smoother on CF than single GPU for me?

 

[Black Octagon]

^ Dude, I'm not here to fight. If it makes you happy, I apologise and, again, was not trying to be flippant, arrogant or anything else. If PCPer' FCAT testing backs up what you say then I hereby officially change my mind and an open to the possibility that I simply had a case of bad luck with the (pre-2013) games that I tried. Sucks for me. Great for you. I truly hope that this is a viable fix for everyone and that this will soon lead to out of the box smoothness for future Crossfire users

 

[bystander36]

The one thing that always lingers in my mind when I hear people say, "if you don't notice microstutter, it isn't a problem", is that they probably would still see a difference in smoothness when compared to a system without the microstutter, or they have far more power than they need.

It seems to me that smoother is smoother, whether we see the difference or not, but if you don't, you must have passed the threshold for smooth game play with both systems.

The part that still needs to be explored, is at what point does FPS or consistent frame deliver, matter more.

Is it just a matter of the worst case frames? If so, how frequently do those worst case scenarios have to take place before it is worse? Is constant variation a bigger problem? Is there a magic max frame time that we just have to avoid?

I'd love to see these reviewers pin point some formula that results in smooth game play. It might be easier to test on weaker cards, but my guess is we'll always have some uncertainty.

 

[UaVaj]

if you are currently happy with the level of microstutter from crossfire. more power to you. as long as you DO NOT try sli. you will never know what you are missing.

ignorance is bliss!

-----

per tom's review. with the protype driver - looks like the level of microstutter is just a hair below acceptable level with a few spikes.

 

[ICDP]

No apologies required, I know CF has serious problems and that out of the box it is horrible. The fact that you have a 120Hz monitor may be the issue, so I don't dispute that you had problems. I dispute your assertion that it is unfixable by the simple reason that you (or I ) can't possibly assume to talk for everyone.

 

[railven]

The one thing that always lingers in my mind when I hear people say, "if you don't notice microstutter, it isn't a problem", is that they probably would still see a difference in smoothness when compared to a system without the microstutter, or they have far more power than they need.

It seems to me that smoother is smoother, whether we see the difference or not, but if you don't, you must have passed the threshold for smooth game play with both systems.

The part that still needs to be explored, is at what point does FPS or consistent frame deliver, matter more.

Is it just a matter of the worst case frames? If so, how frequently do those worst case scenarios have to take place before it is worse? Is constant variation a bigger problem? Is there a magic max frame time that we just have to avoid?

I'd love to see these reviewers pin point some formula that results in smooth game play. It might be easier to test on weaker cards, but my guess is we'll always have some uncertainty.

This post is very interesting because now I wonder where the conversation will go.

Someone else posted something very similar to this, and I agree with it then like I agree with it now. The issue was, back then the poster was a known Radeon prefer so of course his post was turned into a moving the goal-post/deflect slant that a conversation never really followed.

I've very sensitive to tearing, and stutter (that I contribute to frame dips) but I'm not seeing the stutter that is discussed in forum posts and reviews. I, however, use v-sync and refuse to not use it. Turning it off to bench for the sake of "highest FPS" I get so much tearing I can't really say the game is smooth, feels smooth, or is a pleasant experience. So I deal with the input lag (which honestly doesn't seem to bother me, but I'm not a professional gamer.)

Eitherway, AMD has issues and they seem to be addressing them. I'm just glad to see a nvidia prefered poster to a similar thought process that was presented a long time ago. Where do we draw the line in the sand on what is essentially a subjective experience?

We can look at FCAT charts all day, but if 80% of those watching don't notice it or let it affect them is it an issue? I dunno. For the pom-pom waivers, it definitely is because it's something they can hold over their adversary. What about those of us who don't care about anyone's bottom line or market share? /shrug.

 

[ICDP]

if you are currently happy with the level of microstutter from crossfire. more power to you. as long as you SO NOT try sli. you will never know what you are missing.

ignorance is bliss!

-----

per tom's review. with the protype driver - looks like the level of microstutter is just a hair below acceptable level with a few spikes.

I used 2x GTX680 in SLI and they were much, much better than CF, but they absolutely were not MS free. I was still having to apply FPS caps and vsync to eliminate MS in quite a number of games.

 

[railven]

I used 2x GTX680 in SLI and they were much, much better than CF, but they absolutely were not MS free. I was still having to apply FPS caps and vsync to eliminate MS in quite a number of games.

I'm almost tempted to get a second GTX 680 just to test it and see for myself if the grass really is greener on the other side (no pun intended.)

I'm actually very satisfied with my short CFX exposure. All the games I play work, I get scaling in games multiple reviews from my own reading say have negative/zero scaling, and my cards run cooler than I was expecting since I can run them at lower volts than stock.

 

[bystander36]

This post is very interesting because now I wonder where the conversation will go.

Someone else posted something very similar to this, and I agree with it then like I agree with it now. The issue was, back then the poster was a known Radeon prefer so of course his post was turned into a moving the goal-post/deflect slant that a conversation never really followed.

I've very sensitive to tearing, and stutter (that I contribute to frame dips) but I'm not seeing the stutter that is discussed in forum posts and reviews. I, however, use v-sync and refuse to not use it. Turning it off to bench for the sake of "highest FPS" I get so much tearing I can't really say the game is smooth, feels smooth, or is a pleasant experience. So I deal with the input lag (which honestly doesn't seem to bother me, but I'm not a professional gamer.)

Eitherway, AMD has issues and they seem to be addressing them. I'm just glad to see a nvidia prefered poster to a similar thought process that was presented a long time ago. Where do we draw the line in the sand on what is essentially a subjective experience?

We can look at FCAT charts all day, but if 80% of those watching don't notice it or let it affect them is it an issue? I dunno. For the pom-pom waivers, it definitely is because it's something they can hold over their adversary. What about those of us who don't care about anyone's bottom line or market share? /shrug.

Well, if it makes you feel better, while I have 680's now, my previous setup was 6950's, and 5870's before. I had a 9800pro as well, and some Nvidia cards in between. We can only make choices on what we see on the review sites, and I've always tried to be unbiased when I buy my cards, but I have to admit, I've always noticed a general "it just works" experience with Nvidia, where I always had to fiddle a lot with my AMD cards.

Anyways, I'm currently stuck with Nvidia, as 3D Vision caught my attention. I tried HD3D first (on same monitor that supported both, though limited to 720p with HD3D), but found 3D Vision was better with a great mod community to fix many games that normally doesn't work.

It would be great if they fix the issue. I have been impressed at their new driver support lately. They've been much better at fixing issues by the reviews you see. I will likely give them another go in the future. I always do.

 

[ICDP]

The one thing that always lingers in my mind when I hear people say, "if you don't notice microstutter, it isn't a problem", is that they probably would still see a difference in smoothness when compared to a system without the microstutter, or they have far more power than they need.

I have no doubt that is true, I frequently see people who claim CF is perfect out of the box and my 1st thought is "how can they say that with a straight face". I also feel the same when someone claims SLI is always perfect and that it is MS free. Yes SLI is better out of the box than CF but somehow the consensus on most forums is CF = always broken and SLI = always perfect.

It seems to me that smoother is smoother, whether we see the difference or not, but if you don't, you must have passed the threshold for smooth game play with both systems.

The part that still needs to be explored, is at what point does FPS or consistent frame deliver, matter more.

Is it just a matter of the worst case frames? If so, how frequently do those worst case scenarios have to take place before it is worse? Is constant variation a bigger problem? Is there a magic max frame time that we just have to avoid?

I'd love to see these reviewers pin point some formula that results in smooth game play. It might be easier to test on weaker cards, but my guess is we'll always have some uncertainty.

There in lies the problem, everyone has different thresholds of what is acceptable so it is always going to be subjective.

 

[bystander36]

There in lies the problem, everyone has different thresholds of what is acceptable so it is always going to be subjective.

That is where we are at, but as you are seeing now, we have found that we were working with incomplete data. It may be possible to get a more narrowed view on what is better or not.

Before, it was strictly about FPS, maybe now, it'll be some computation about 90thPercentileFrameTimeAve with variance under X.

I expect some sites will find some secret sauce to at least give us a better clue on what is better for them at least.

Perhaps they can have a formula that determines what is smoother with X frame time and Y variance, and people can learn just how smooth they need to be happy, but you might be right in that what is smooth could vary from person to person. Variance might be worse to one person than another, while FPS is bigger to one person or another.

 

[bystander36]

On the above topic, TechSpot has an interesting way to do what I was asking about.
http://www.techspot.com/review/663-amd-radeon-hd-7990/page4.html
Look at the charts at the bottom of each review with the 99th percentile frame time. I'm not exactly sure how they came up with those numbers yet, but they clearly are some form of formula to determine what is smoothest. This is the type of thing I'm be very interested in seeing (edit: to add one more favorable to AMD, as the other may have been the worst case).

 

[Black Octagon]

No apologies required, I know CF has serious problems and that out of the box it is horrible. The fact that you have a 120Hz monitor may be the issue, so I don't dispute that you had problems. I dispute your assertion that it is unfixable by the simple reason that you (or I ) can't possibly assume to talk for everyone.

Fair enough, and indeed we cannot speak for others. 120Hz may indeed be the issue because, since moving to it, I find myself quite unsatisfied gaming at 60Hz/fps on other people's screens/systems. Some people claim not to notice a difference between 60 and 120Hz/fps gaming, but I am not one of them. Perhaps this change has made me particularly intolerant of imperfect smoothness...I'm not sure.

However, before wrapping this up I would like to ask this: you say that we can look at a near-smooth frame metering graph produced via FRAPS and can conclude that the same smoothness should be displayed on an FCAT graph based on the same sample of gameplay. What's your thinking here? I'm not saying that you're wrong. It's just that Anandtech's coverage from their interview with AMD seemed pretty clear that:

FRAPS is at the very start of the rendering pipeline; it’s before the GPU, it’s before the drivers, it’s even before Direct3D and the context queue. As such FRAPS can tell you all about what goes into the rendering pipeline, but FRAPS cannot tell you what comes out of the rendering pipeline.

So to use FRAPS in this method as a way of measuring frame intervals is problematic. Considering in particular that the application can only pass off a new frame when the context queue is ready for it, what FRAPS is actually measuring is the very start of the rendering pipeline, which not unlike a true pipe is limited by what comes after it. If the pipeline is backed up for whatever reason (context queue, drivers, etc), then FRAPS is essentially reporting on what the pipeline is doing, and not the frame interval on the final displayed frames. Simply put, FRAPS cannot tell you the frame interval at the end of the pipeline, it can only infer it from what it’s seeing.

That whole quotation is interesting in itself, but particularly the part in bold. FRAPS measures the part of the pipeline that comes before the GPU and the drivers, and yet it is well-accepted that microstutter from Crossfire/SLI stems directly from AFR. AFR is, as I understand it, something handled near-entirely by the GPU. So how can we conclude that FRAPS graphs from a Crossfire setup - even smooth graphs - tell us with confidence that the graph would also be 'smooth' if done using FCAT? This is one major reason why I remain(ed) skeptical of your graphs as evidence that vsync fixed microstutter.

Or am I missing something?

 

[BrightCandle]

If fraps is showing a stuttering trace then there is a still a problem, but a lot of people misunderstand what it is measuring. The key part people are missing is that fraps measures the time when the present method is called in the DirectX API, which signifies to the card its time to show the frame. What the graphics card driver does in response is part all the instructions for the frame into the context queue. However if there isn't space in the queue then the thread calling onto present is blocked, it just sits idle until there is space.

Now the key point is that almost all games are GPU limited, at least to some extent. Which means that there is always a thread waiting on the present call and it wont get released until the GPU driver has space for it to do so. This is how games regulate their frame rate. They dont try to create the next frame until the previous one has been presented. Some people are calling this backpressure.

So if a fraps trace shows uneven frames then the game simulation itself is being told to produce frames at varying pace, which means the animation in the game will be at odd times. So you can smooth the output onto the screen but the contents of the frames themselves are stuttering. You can stutter the pixels of the frames (like the AMD AFR does with crossfire resulting in the frame not being displayed) but if the card is uneven the contents of the frame is also not produced on an even schedule and it makes things even worse.

AMDs solution is causing both things to happen, which makes it doubly bad. FCAT measures the output to the screen, the end of the pipeline but fraps measures the input, and they both need to be smooth as I have explained.

Pcper have some lovely videos of the impact of vsync on games as well as FCAT graphing. While it solves the runt frame issue it does not solve the animation issue and it introduces its own type of stuttering effect which makes the fraps results even worse than otherwise. In essence it might fix one problem but it introduces another, its far from a reasonable fix.

 

[UaVaj]

I have no doubt that is true, I frequently see people who claim CF is perfect out of the box and my 1st thought is "how can they say that with a straight face". I also feel the same when someone claims SLI is always perfect and that it is MS free. Yes SLI is better out of the box than CF but somehow the consensus on most forums is CF = always broken and SLI = always perfect.

There in lies the problem, everyone has different thresholds of what is acceptable so it is always going to be subjective.

a. level of microstutter is objective n measureable. amd simply sutter more. a lot more.
b. perception of microstutter is where it gets subjective. some folks are more perceptable than others.

both side stutter. nvidia just a lot lot less.
the biggest question is. at what level of microstutter is consider acceptable to: (a) the occasional gamer, (b) the regular gamer, and (c) the hard core gamer?

 

[Silverforce11]

I'm told hardcore gamers dont play with vsync cos the input lag is too much "OMG WTFBBQ" horrible and it would affect their ability to perform uber flick headshots....