PCPer on Crossfire problems in the Titan review

[Final8ty]

I saw those videos quite a while ago and I had a similar reaction. Right, there IS a stuttering moment, but honestly I would already say that example is unplayable due to that crazy tearing they are seeing. I think it only makes sense to look at micro stuttering when there is not any glaringly obvious tearing, otherwise it is like "Hey we take this example which is completely unplayable and we found a second, less noticeable, problem with it". If vsync is the best solution, I do not know, but it is one solution. I would rather have tearing free experience without having to turn on vsync though.

Indeed!
Massive in your face unbearable tearing but lets not worry about that and go slowmo to find that little bit of stutter.

 

[BrightCandle]

I want these new techniques done at 120 hz because in theory at least the problem of stutters and especially microstuttering should be reduced. I would like to think it will halve the problem but I suspect its more likely to only reduce the waiting for frame part of the equation and not the actual frame render part of stuttering.

Its a shame tech report, pcper and others are still at 60hz only, we need a comparison to 120hz to know if it makes a difference and if the input lag is doable with sync on at 120hz.

 

[Rikard]

Indeed!
Massive in your face unbearable tearing but lets not worry about that and go slowmo to find that little bit of stutter.

I am sure they could have found tearing free examples with micro stuttering if they wanted to. But I guess that would be boring to see when a couple of frames are identical to their predecessor. Instead they show multiple frames simultaneous on screen. That is really a measure of tearing that they have, and they could do some great stuff with it if they just decide to interpret their results for what they are.

 

[SickBeast]

I will say that my crossfire experience was 2nd rate compared to my SLI experience. There was definitely microstutter in certain games with both setups, but it was worse with the AMD cards, plus I was unable to adjust my 2nd card's voltage in crossfire and overclocking was just a general PITA.

 

[Ferzerp]

This whole tearing fixation is a red herring and you folks should know better.

*every* game with every card exhibits tearing at a mostly random place on a frame unless vsync is turned on.

Why exactly are you trying to steer the conversation away from what they're using that tearing to measure to the tearing itself?

 

[VulgarDisplay]

This whole tearing fixation is a red herring and you folks should know better.

*every* game with every card exhibits tearing at a mostly random place on a frame unless vsync is turned on.

Why exactly are you trying to steer the conversation away from what they're using that tearing to measure to the tearing itself?

I'm sorry but isn't the whole point of these testing methods to measure smoothness? Isn't introducing massive amounts of tearing by outrunning your refresh rate the exact opposite of encouraging a smooth gaming experience?

This is a remnant of the Max FPS benchmark approach that the same reviewers were vilifying to justify their new testing approach. The two do not go hand in hand. They continue to shove Max FPS down our throat, while at the same time saying it's no longer good enough.

The only benefit of running faster than 60fps on a 60hz monitor is the increased responsiveness of the controls. The monitor is not displaying more than that so there is absolutely no point in testing for a smooth experience beyond 60hz.

They need to switch to 120hz monitors and then get back to us on this whole issue.

 

[BrightCandle]

Switching to 120hz won't remove the tearing, you don't magically get rid of the tears because the frame rate is below the frequency of the monitor. If you don't sync every single frame will have a tear in it somewhere except in those rare 1 in 1080 moments when it aligns perfectly.

What these runt frames are about is AMD not doing frame metering. They produce 2 frames very close to each other, on of which is swapped in less than a millisecond after the other and did nothing for the player as the frame was never really displayed. Greater than 60 fps on 60 hz screens just means more than 1 tear is increasingly likely.

Pcper are doing something completely reasonable when they remove the runt frames. A frame that only shows an few lines did nothing for the user at all, it was wasted effort entirely and should be removed from the fps chart to correct for the problem. The small tail of a picture that is the next frame should not be removed because it will form the top of the next frame, although it would be nice to know if its also cut short and its total display lines is less than the useful threshold.

What pcper has come across is a problem none of us knew about. AMD is getting artificially high fps in xfire with frames being swapped very close. Its effectively worthless to have xfire over a single card in the game shown. The same is not true on Nvidia.

The previous fraps traces gave us an idea of what was wrong but not the whole story, clearly these traces are a superior test and much more accurate. I have no problem saying that now I would prefer a captured trace over frame times, but I still prefer frame times over averaged fps as it still shows up problems that fps does not. But we also know that the jitter nvidia cards show is reduced by the time it gets on screen and on amd its increased massively. That is a bit of shock for many here but not for me, I maintain there is something deeply wrong on the 7970s I own and its not that the cards are faulty. Looking more and more like amd has a terrible solution, and I want more games tested and some way to review the results myself.

 

[VulgarDisplay]

Switching to 1120hz won't remove the tearing, you don't magically get rid of the tears because the frame rate is below the frequency of the monitor. If you don't sync every single frame will have a tear in it somewhere except in those rare 1 in 1080 moments when it aligns perfectly.

I know that but doesn't it stand to reason that there is a much greater chance of a 120hz monitor only displaying complete, or more complete frames at say 90fps than a 60hz monitors at 90fps?

I think this is why 120hz monitors seem so much smoother than a 60hz monitor. Before I put my fist through my old 60hz Samsung I noticed that the animations in BF3 while good seemed jittery. I thought this was the way that they were.

After I smashed that monitor and got my 120hz monitor the animations seemed fluid, and I had no idea that I was missing out on these animations. If you haven't seen 120hz before you are truly missing out.

 

[BrightCandle]

I know that but doesn't it stand to reason that there is a much greater chance of a 120hz monitor only displaying complete, or more complete frames at say 90fps than a 60hz monitors at 90fps?

I think this is why 120hz monitors seem so much smoother than a 60hz monitor. Before I put my fist through my old 60hz Samsung I noticed that the animations in BF3 while good seemed jittery. I thought this was the way that they were.

After I smashed that monitor and got my 120hz monitor the animations seemed fluid, and I had no idea that I was missing out on these animations. If you haven't seen 120hz before you are truly missing out.

The chance is basically 1/1080 (on a 1080p) screen. So at 60 hz you get half the number of perfect frames per second of 120hz, but its very unlikely you will get that frame within a second, its only really going to happen once every 1080/60 seconds or once every 1080/120 seconds. That is such a rare event that it will have no impact on smoothness at all.

At >60fps on a 60hz monitor (or >120fps on 120hz) you will see more than one tear becoming increasing common on a screen. So to some extent you will notice more tearing, and the chance of a perfect frame decreases further but perfect or near perfect is so rare that it doesn't matter.

 

[Ferzerp]

I know that but doesn't it stand to reason that there is a much greater chance of a 120hz monitor only displaying complete, or more complete frames at say 90fps than a 60hz monitors at 90fps?

I think this is why 120hz monitors seem so much smoother than a 60hz monitor. Before I put my fist through my old 60hz Samsung I noticed that the animations in BF3 while good seemed jittery. I thought this was the way that they were.

After I smashed that monitor and got my 120hz monitor the animations seemed fluid, and I had no idea that I was missing out on these animations. If you haven't seen 120hz before you are truly missing out.

Amazing how you've turned this thread about whether or not a specific line of cards is actually creating every frame that it reports it is in to something else entirely. I suggest that if you want to discuss this topic, you do it in a thread that you aren't mysteriously hijacking for some odd reason.

 

[BrightCandle]

Amazing how you've turned this thread about whether or not a specific line of cards is actually creating every frame that it reports it is in to something else entirely. I suggest that if you want to discuss this topic, you do it in a thread that you aren't mysteriously hijacking for some odd reason.

I think its relevant to address it since the pcper data shows up something i think surprises a lot of people, that almost every screen has a tear. The question of if tears affect the smoothness is perfectly relevant, and a tough topic to address technically so if someone has a better analysis than mine I am all eyes.

But I also agree its somewhat orthogonal to the overall point which is pcper new review showing a really big problem on crossfire and some amazing work done their on really delivering on the promise of their new review approach.

 

[Ferzerp]

I think its relevant to address it since the pcper data shows up something i think surprises a lot of people, that almost every screen has a tear. The question of if tears affect the smoothness is perfectly relevant, and a tough topic to address technically so if someone has a better analysis than mine I am all eyes.

But I also agree its somewhat orthogonal to the overall point which is pcper new review showing a really big problem on crossfire and some amazing work done their on really delivering on the promise of their new review approach.

It seems a common tactic on this board. If the subject is one that people dislike, they kill the discussion of that subject through various means. It magically happens any time some site tries to look at performance using anything but raw FPS. Now, because of the hijacking, the discussion of the findings is gone, but something totally unrelated (and inherent to anything not using vsync) is magically "important".

 

[VulgarDisplay]

The chance is basically 1/1080 (on a 1080p) screen. So at 60 hz you get half the number of perfect frames per second of 120hz, but its very unlikely you will get that frame within a second, its only really going to happen once every 1080/60 seconds or once every 1080/120 seconds. That is such a rare event that it will have no impact on smoothness at all.

At >60fps on a 60hz monitor (or >120fps on 120hz) you will see more than one tear becoming increasing common on a screen. So to some extent you will notice more tearing, and the chance of a perfect frame decreases further but perfect or near perfect is so rare that it doesn't matter.

The other thing is that the faster refresh rate will result in less jump between the frames that are being rendered. I will use an extreme example:

Say there is a bar down the middle of the the screen. You pan the camera at a high rate and using the PCPer's capture method it shows the next rendered frame having the bar displays 1/2 inch away from the previous frame at the bottom of the screen.

On a 120hz screen this 1/2 inch difference would be reduced to 1/4 inch and more than likely be at a different spot on the screen, probably higher up since the screen is rendered top to bottom.

This is a very extreme example, but I think it illustrates the concepts my simple mind are trying to portray.

 

[VulgarDisplay]

Amazing how you've turned this thread about whether or not a specific line of cards is actually creating every frame that it reports it is in to something else entirely. I suggest that if you want to discuss this topic, you do it in a thread that you aren't mysteriously hijacking for some odd reason.

We were on topic until you posted this trash. Go away.

Discussions often times evolve as they progress. If no one talked about anything new in relation to what has been discussed every thread would die in less than one page.

I would also say that I'm not steering this topic away from it's origin because I didn't like what was being discussed, but I truly want to learn more and you're letting your own perception of my supposed bias make you irrational.

 

[BrightCandle]

The GPU is going to render at its pace, so if anything your bar will appear further down the screen or on the next frame because its scanning out twice as fast. But I don't think this effect in itself means we perceive the motion to be smoother, twice the hz is doing that all on its own as its showing more actual frames, more information is coming across as long as >60fps is maintained. It isn't the tearing placement that is impacting our feeling of smooth its the faster pixel refresh, faster scan out showing us more frames and younger frames.

 

[VulgarDisplay]

The GPU is going to render at its pace, so if anything your bar will appear further down the screen or on the next frame because its scanning out twice as fast. But I don't think this effect in itself means we perceive the motion to be smoother, twice the hz is doing that all on its own as its showing more actual frames, more information is coming across as long as >60fps is maintained. It isn't the tearing placement that is impacting our feeling of smooth its the faster pixel refresh, faster scan out showing us more frames and younger frames.

It gives less jump between each frame which is now think is the more important piece of the puzzle.

 

[notty22]

Can't wait for the follow up article where this investigation in crossfire is explained further. There was a dark time when scaling %'s were being tossed around in 'best card discussions' as a card pro, to the extent it was promoted even it was clear the buyer was never going multi-gpu. Yeah, made no sense.

 

[BrightCandle]

It gives less jump between each frame which is now think is the more important piece of the puzzle.

Indeed. I could be wrong of course but I cant think of a counter example except where 2 buffer swaps happen at nearly the same moment. In that case the first of the two swaps didn't impact smoothness at all, the frame wasn't really displayed but it also has a clear knockon effect on the next frame which will also have an older top and a tear further down than it should have done given the frame rate.

I really need to see the details of how pcper is determining which frames to ignore. On the one hand I suspect they are making a reasonable choice, but it might not be quite the right way to think about the problem, as latency is at play determining which frames are shared with others. A runt frame is worse than just loosing that frame, it causes a double jump between frames and a wider tear, both of which are more noticeable and its impacting the next frame as well as the bottom of this one is the top of the next.

 

[VulgarDisplay]

Indeed. I could be wrong of course but I cant think of a counter example except where 2 buffer swaps happen at nearly the same moment. In that case the first of the two swaps didn't impact smoothness at all, the frame wasn't really displayed but it also has a clear knockon effect on the next frame which will also have an older top and a tear further down than it should have done given the frame rate.

I really need to see the details of how pcper is determining which frames to ignore. On the one hand I suspect they are making a reasonable choice, but it might not be quite the right way to think about the problem, as latency is at play determining which frames are shared with others. A runt frame is worse than just loosing that frame, it causes a double jump between frames and a wider tear, both of which are more noticeable and its impacting the next frame as well as the bottom of this one is the top of the next.

Yeah, that's true too about the hidden frame increasing the jump between the frame before and after.

I hadn't thought of that because I'm mainly applying all of my conjecture to my own situation using a 120hz monitor and a single card.

 

[3DVagabond]

Yeah, that's true too about the hidden frame increasing the jump between the frame before and after.

I hadn't thought of that because I'm mainly applying all of my conjecture to my own situation using a 120hz monitor and a single card.

What PCPer showed was a screen capture of a game running crossfire, not a single card. Now, I have to assume that this situation PCPer is showing happens with regularity, or they wouldn't be making such a big deal out of it (removing crossfire results all together). I don't see where you would apply it to a single card, though. You would think if it applied to single cards PCPer would have mentioned that as well.

 

[BrightCandle]

What PCPer showed was a screen capture of a game running crossfire, not a single card. Now, I have to assume that this situation PCPer is showing happens with regularity, or they wouldn't be making such a big deal out of it (removing crossfire results all together). I don't see where you would apply it to a single card, though. You would think if it applied to single cards PCPer would have mentioned that as well.

The other chart of adjusted frame rate seems to suggest the runt frames are very common, in fact most frames show one. There are a few cases where crossfire outperformed a single card given the adjustment but also some moment where it hurt performance. This does show its a very common problem in bf3.

Question is does it happen in other games as well, it seems extremely likely given its cause and pcper decision not to show it through the review but as always I want more games with adjusted frame times shown, those charts are enormously telling.

I think pcper may finally have gotten to the bottom of my poor smoothness 7970 xfire problem, its great news that the problem has been found.

 

[3DVagabond]

The other chart of adjusted frame rate seems to suggest the runt frames are very common, in fact most frames show one. There are a few cases where crossfire outperformed a single card given the adjustment but also some moment where it hurt performance. This does day its a very common problem in bf3.

Question is does it happen in other games as well, it seems extremely likely given its cause and pcper decision not to show it through the review but as always I want more games with adjusted frame times shown, those charts are enormously telling.

I think pcper may finally have gotten to the bottom of my poor smoothness 7970 xfire problem, its great news that the problem has been found.

Seems reasonable. Of course we do need a broader range of games tested, but I agree that if it was game specific they likely wouldn't be making as big of a deal out of it as they are.

Assuming it holds up to more scrutiny, it's great that they found it. It would be nice if improved crossfire IQ came from this.

 

[Eymar]

Assuming it holds up to more scrutiny, it's great that they found it. It would be nice if improved crossfire IQ came from this.

I'm in total agreement with this. As a 7970 crossfire owner and potential whatever amd has next + crossfire owner, I would really like amd to address crossfire microstutter as least as much as Nvidia has (at minimum implement frame metering for vsync-off users and whatever Nvidia does for SLI for vsync-on users(myself)). The two main microstutter issues that I like to see fix is when microstutter causes a slower effective framerate (ie. Fraps reports 60fps, but subjectively looks like 45 or so FPS) and as PCPer videos show the slight jolt in smoothness when doing fast turns in 1st person shooters.

 

[3DVagabond]

I'm in total agreement with this. As a 7970 crossfire owner and potential whatever amd has next + crossfire owner, I would really like amd to address crossfire microstutter as least as much as Nvidia has (at minimum implement frame metering for vsync-off users and whatever Nvidia does for SLI for vsync-on users(myself)). The two main microstutter issues that I like to see fix is when microstutter causes a slower effective framerate (ie. Fraps reports 60fps, but subjectively looks like 45 or so FPS) and as PCPer videos show the slight jolt in smoothness when doing fast turns in 1st person shooters.

Keep in mind that frame metering slows down the quicker displayed frames to match up to the slower ones. It doesn't actually solve the face issue of uneven frame distribution. It covers it up. Don't anyone get their panties in a bunch, it's definitely the lesser of the two evils and a visual improvement, which is good.

IMO, it would be better if they looked at the games that don't exhibit this issue to find out why, and then make other games operate the same. Adding latency isn't really a fix anymore than frame capping is. It's another type of band-aid.

 

[SolMiester]

Keep in mind that frame metering slows down the quicker displayed frames to match up to the slower ones. It doesn't actually solve the face issue of uneven frame distribution. It covers it up. Don't anyone get their panties in a bunch, it's definitely the lesser of the two evils and a visual improvement, which is good.

IMO, it would be better if they looked at the games that don't exhibit this issue to find out why, and then make other games operate the same. Adding latency isn't really a fix anymore than frame capping is. It's another type of band-aid.

Maybe its just an issue when the game taxes the GPU in a certain way, which would explain why it doesn't appear to be across the board....which would then make NVs solution more palatable for you.....