AMD, say it ain't so!

[BiggieShady]

If you understand what the graphs show(an increase in frame-to-frame render latency), that should be enough, really. Visually it amounts to a momentary pause.

When you add a second card, because this doubles that delay, the pause is even longer, and becomes even more noticeable, to the point that singlecard performance is perceptibly smoother, even though the framerate is significantly lower.

AMD has said, that with DX10 and DX11, this is due to memory management. With DX9, performance is adequate enough that it can be dealt with easily, as far as I can tell, but I'm not expert with that sort of stuff at all.

So, imagine playing at 60+FPS, and it's nice and smooth. FPS drops below 60 with some action, and it becomes less smooth. Staying above 60 FPS will keep that smoothness. Adding a second card, and increasing performance above 60 FPS should keep the same smoothness, right?

Unfortunately, what happens is that those stutters from going to less than 60 FPS overall still apply, but PER CARD, so that now to maintain a smooth experience, you need 120 FPS, negating the usefulness of the second card. Add in that the added driver overhead from managing both cards, and overall you end up with a worse experience than with one card.

Thanks for the clarification Dave.

Looking at these graphs, what surprised me is that FRAPS actually makes SLI look worse than it is (frame time fluctuation wise) and CrossFire looks better than it is.
Also it looks like there is some smoothing going on with SLI when looking at hardware capture graphs. If it's all due to memory management, no wonder Kepler team was so proud about it.

I agree there is no drama (except in the thread title :shadedshu what was I thinking?) people know what are the tradeoffs with multi gpu setups.

With AMD not producing any new GPUs this year, I would be surprised if they didn't improve the drivers significantly during that time. It should all be fixed soon, provided they didn't fire their driver team.

 

[cadaveca]

Looking at these graphs, what surprised me is that FRAPS actually makes SLI look worse than it is (frame time fluctuation wise) and CrossFire looks better than it is.
Also it looks like there is some smoothing going on with SLI when looking at hardware capture graphs. If it's all due to memory management, no wonder Kepler team was so proud about it.

I think the difference in visual example between the two methods is how they capture what is going on. Also, having two methods running at once is rather... um... silly.


It's funny because before the 7-series launch, Crazyeyesreaper and I were discussing what we knew about these AMD GPUs, and we both agreed that the 7870 seemed to have the best balance between all it's parts, shaders, TMUs, ROPs, memory bus, etc...and what AMD does to "hide" these mis-balances is essentially GPU virtualization.

To me, this translates to cache misses, or rendering a frame, and the needed data isn't present right where it needs to be, so a frame takes longer to render than it should. It's probably coming from local memory, rather than within cache.

This is called a "TLB", a buffer that helps find what data is needed. It translates virtualized memory addresses into real ones. IF it thinks something is in the right spot, and it isn't, things like what AMD's GPUs experience makes sense.

It also would explain why when you add another card, and get higher FPS, that higher FPS doesn't translate into actual performance gains, since it's technically a hardware problem, and contained within each GPU, rather than something to do with the actual output to the monitor.

OF course, I could be totally off here, but this is what I think is going on, and why it has taken so long for AMD to come up with a fix for this...

Nvidia designed their GPUs with CUDA in mind, not graphics, really. Virtualizing GPU resources makes the problem a software one, so NVidia concentrated on hardware built for the software, meanwhile, AMD is making drivers for their hardware. This different approach leads to a very different design philosophy, that might have created these problems in the first place. for AMD to say "we didn't look at smoothness", well, that means they don't care about anything other than FPS. So we got good FPS...

 

[RejZoR]

The whole micro stuttering problem is way overhyped. I'm not going to deny that it's not a problem, but not as big as many say. My eyes are very sensitive to any framerate issues and so far the only game that sort of bothered me was Skyrim, but even there the issue wasn't as big for the game not to be playable.

 

[ChristTheGreat]

but is anyone has seen the date of the article? xD

This is like a year past...

 

[the54thvoid]

...It also would explain why when you add another card, and get higher FPS, that higher FPS doesn't translate into actual performance gains,

That's not really accurate either. Case in hand, Crysis 3 had choppy fps - as it happened it was due to not having crossfire working. Downloaded the newer drivers (crossfire fix) and the frame rate went up by a long way and the smoothness is, well, smooth.

I know there are issues with crossfire but the latency issues are overhyped to the nth degree. I'm genuine entranced by Crysis 3 and I'm able to run it with 2x smaa at 2560x1440, getting 45-60 ish fps (albeit with 2 gpus). I am noticing NO stutter at all.

It's definitely not an endemic problem. It may occur but it is not universal. I still refute that we can see the "problems" all the time. It's game specific and even then, many people have stated they don't notice it.

Crossfire brings fps up and also in (I would say in my experience) most cases it smooths out the experience. To be fair, crossfired 7970's give 120fps+ in many titles so you wouldn't notice the lags. But again, even under 60fps, I'm not seeing it.

Maybe I've just got shit eyes.

And I'm definitely not arguing with you Dave - I know you know a shit load more about PC tech than I do and have way more experience of AMD problems but the issue is way overblown - not saying it isn't there - just way overblown.

but is anyone has seen the date of the article? xD

This is like a year past...

http://www.pcper.com/reviews/Graphi...art-3-First-Results-New-GPU-Performance-Tools

This is the OP article, from 22nd February 2013.

 

[ChristTheGreat]

That's not really accurate either. Case in hand, Crysis 3 had choppy fps - as it happened it was due to not having crossfire working. Downloaded the newer drivers (crossfire fix) and the frame rate went up by a long way and the smoothness is, well, smooth.

I know there are issues with crossfire but the latency issues are overhyped to the nth degree. I'm genuine entranced by Crysis 3 and I'm able to run it with 2x smaa at 2560x1440, getting 45-60 ish fps (albeit with 2 gpus). I am noticing NO stutter at all.

It's definitely not an endemic problem. It may occur but it is not universal. I still refute that we can see the "problems" all the time. It's game specific and even then, many people have stated they don't notice it.

Crossfire brings fps up and also in (I would say in my experience) most cases it smooths out the experience. To be fair, crossfired 7970's give 120fps+ in many titles so you wouldn't notice the lags. But again, even under 60fps, I'm not seeing it.

Maybe I've just got shit eyes.

And I'm definitely not arguing with you Dave - I know you know a shit load more about PC tech than I do and have way more experience of AMD problems but the issue is way overblown - not saying it isn't there - just way overblown.



http://www.pcper.com/reviews/Graphi...art-3-First-Results-New-GPU-Performance-Tools

This is the OP article, from 22nd February 2013.

Ok, I missed the post where it was linked, I just saw the techreport one. Thanks

 

[cadaveca]

That's not really accurate either. Case in hand, Crysis 3 had choppy fps - as it happened it was due to not having crossfire working. Downloaded the newer drivers (crossfire fix) and the frame rate went up by a long way and the smoothness is, well, smooth.

I know there are issues with crossfire but the latency issues are overhyped to the nth degree. I'm genuine entranced by Crysis 3 and I'm able to run it with 2x smaa at 2560x1440, getting 45-60 ish fps (albeit with 2 gpus). I am noticing NO stutter at all.

It's definitely not an endemic problem. It may occur but it is not universal. I still refute that we can see the "problems" all the time. It's game specific and even then, many people have stated they don't notice it.

Crossfire brings fps up and also in (I would say in my experience) most cases it smooths out the experience. To be fair, crossfired 7970's give 120fps+ in many titles so you wouldn't notice the lags. But again, even under 60fps, I'm not seeing it.

Maybe I've just got shit eyes.

And I'm definitely not arguing with you Dave - I know you know a shit load more about PC tech than I do and have way more experience of AMD problems but the issue is way overblown - not saying it isn't there - just way overblown.

Overall, yeah, I agree with you. I mean, I said it here earlier...there is no drama to be found here.

However, this is a real problem, and yes, it is something that is per-app, for sure. Overall, things are much better the past few months, so clearly drivers are getting better, but in all honesty, I don't expect them to be able to fix everything.

Which kinda sucks.

Overall, for me, Crossfire is rather useless. There are but three or four apps where it actually makes a tangible difference over a single card.

I've tried everything...fast drives, slow drives, high-end CPUs and low-end, every driver available under the sun, different OSes, different ram, different PSUs, different monitors...even different ways of connecting to monitors. If you enjoy Crossfire, that's great. For me, it's really hard to recommend.

Nobody really seems to understand this problem really well, but I will say, I do seem to have a better grasp than most as to what's going on.

So, what Ryan over at PCPer is doing is taking the output given over the cable to the monitor, and not using FRAPs, and still, on the monitor cable, framerate is noticed as incredibly unsmooth.

So, that makes me wonder...those that have no issues...do they have monitors with built-in scaling hardware? Is it active?


You are right, there's a lot of complications here in diagnosing the issue, even identifying the issue, but after having looked at this for years, literal years, I really do think this is a bigger issue than meets the eye. I also think that properly identifying the problem can go a long way to AMD making sure that newer hardware is not prone to the same issues. After it being an issue over countless generations of hardware, clearly just a driver isn't going to fix it. I hate to say it, but I actually mentioned that this might be an issue prior to the cards even being released. For me, there is no drama, because this is really old news.

With hardware out right now, I don't really expect much to change. Maybe if AMD can get this memory management stuff "fixed", it will get better. AMD agrees it's an issue...so for you to say it's not.. well... I'm glad you are enjoying your PC.

 

[Finners]

Finners is on the TPU forum occasionally so he might be able to shed more light but I genuinely haven't crashed in BF3 for a good few months, if not half a year. Though I've not played yet using 13.2 (6) beta's so maybe I'll come back and edit this post, lol.

The two persistent problems i have that comes and goes as the drivers change is flickering textures on demavand peak and my crosshair disappears when im revived.

I also had a terrible time trying to get bf3 smooth when i ran SLI 470's had to run it with vsync but then that made me have input lag. Good thing was you can see it in the performance overlay. top is with v-sync, second was without and third is my current 680. You can see how much smoother it is.
V-sync

v-sync off

GTX 680 with FPS capped at 65

 

[BiggieShady]

This is the OP article, from 22nd February 2013.

Ok, I missed the post where it was linked, I just saw the techreport one. Thanks

Damn, I was hoping you were a time traveler coming from +1 years in the future and that you'll share the news about AMD's 8000 series.

 

[the54thvoid]

Hate to drag this thread open again but I found this in the TPU reviews round up. Deserves some air time seeing as it directly contrasts the findings from the source OP.

It kind of backs up what I say when i say for me and my PC, BF3 runs very smooth indeed. I haven't even looked at the other graphs from the review but this clearly shows different reviewers are finding different results. This one favouring AMD for the smoother gameplay.

http://uk.hardware.info/reviews/399...force-gtx-680-frametimes-review-battlefield-3

 

[Whitestar]

I think TechReport is doing a good job of at least writing what these frame latency numbers translate into visually. They always include a subjective evaluation in their reviews, describing what their eyes see.
http://techreport.com/review/22890/nvidia-geforce-gtx-690-graphics-card/7
"If you're looking to avoid real slowdowns, consider the 50-ms results, where only the 6990 and the GTX 590 really show any notable issues. At 33.3 ms, the three multi-GPU solutions from the current generation are in a class by themselves, while the Radeon HD 7970 carves out a clear advantage over the GeForce GTX 680."

From what I have read it seems to me that the "smoothness threshold" in games in general is around 40-50 ms, so as long as the big majority of frames are rendered within that time, the results should be mostly smooth. Although it's not quite that simple either.

Looking at the frame times in Battlefield 3 I would think that it should be smooth, but of course mileage could vary, as will the eye of the beholder.

Like radrok says, trust your eyes. If you think it's smooth, then good for you.

Personally I have noticed stutter in Skyrim (pre 13.2) and Just Cause 2 (@1080p) with my 5970.

 

[W1zzard]

This is called a "TLB", a buffer that helps find what data is needed. It translates virtualized memory addresses into real ones. IF it thinks something is in the right spot, and it isn't, things like what AMD's GPUs experience makes sense.

Sorry, no.

Explaining TLB is kinda difficult, most articles I found, including Wikipedia are junk.
These slides http://www.ece.cmu.edu/~ece548/handouts/05vmarch.pdf do a decent job. Gotta go through the whole presentation though to understand the underlying concept.

 

[HammerON]

I switched recently from two GTX 580's to three 7970's (primarily for the purpose of crunching). I have to say that I have found two GTX 7970's in games such as BF3 and Crysis 3 run great. I have not seen any issues with stuttering, then again I have not used any software tools to monitor frame rates...
However running three 7970's has not improved performance. That is okay though as two 7970's allow me to game at 2560x1600 with almost all of the visual candy on

 

[mastrdrver]

I was just thinking, is it possible that the extra flat frame that Ryan is seeing in playing back the recorded output is due to the GPU not being limited by v-sync and thus just throwing up the an extra frame?

you will notice the impact of MS when you are playing as a sniper. drove me insane when i played as a sniper in BF3 with a couple of HD6950's. trying to lock on a moving object made my head explode!

Are you sure your not talking about the lag that happens when looking down sights and trying to peek around an object? The video is old but I still have the same problems. You have to be close to the object your trying to peak around.

 

[crazyeyesreaper]

W1zz layeth the smack down.

 

[Aquinus]

Sorry, no.

Explaining TLB is kinda difficult, most articles I found, including Wikipedia are junk.
These slides http://www.ece.cmu.edu/~ece548/handouts/05vmarch.pdf do a decent job. Gotta go through the whole presentation though to understand the underlying concept.

+1: Good stuff. Thanks for the link. I already knew a good amount of about virtual memory and paging so the level this was pretty good, at least for me. Pardon me if I do a bad job of explaining this, but I think there is a middle ground to describe what happens here.

Basically it's just a set of mappings of virtual memory relative to the cache. Keep in mind that cache does not reside in physical or virtual memory, it stands on its own and does it own thing. Whenever cache is accessed the real address that you're trying to access get translated and checked against the cache. The cache itself only stores the value of the data. Where it is in the cache is derived from the translation from the virtual address.

This is just specifics on how cache itself works as one of the faster tiers of memory (SRAM is fast stuff,) and doesn't tell you much more than what cache is supposed to do, and that's holding on to data that we think we will need again soon. The issue is, if I were to take a guess, is that memory access times are varying and since it's not getting data in the same amount of time every time an instruction is called it adds variability to the amount of time it will take for the GPU to fetch data process it and store it. So even though a particular set of instructions might execute in x amount of time in one circumstance, it could take y time in another instance because the composition of the cache is different and now you have a different number of hits and misses as you did before.

As I said, don't take my word on that, I'm speculating pretty hardcore right here and I could be very wrong, but it would describe why this happens.

 

[BiggieShady]

Deserves some air time seeing as it directly contrasts the findings from the source OP.

I believe you didn't read an article very thoroughly, single GPU performance was never an issue. Article is about CrossFire.

 

[cadaveca]

Sorry, no.

Explaining TLB is kinda difficult, most articles I found, including Wikipedia are junk.
These slides http://www.ece.cmu.edu/~ece548/handouts/05vmarch.pdf do a decent job. Gotta go through the whole presentation though to understand the underlying concept.

TLB isn't used for address translation?



As if my understanding of anything high-level is accurate. I'm mostly making stuff up here, since no one seems to have noticed these problems for some time, so no one can tell me what's up.


Not like this will ever go anywhere, but any other suggested reading?


Can o Worms, I tell ya, can o worms.

 

[the54thvoid]

I believe you didn't read an article very thoroughly, single GPU performance was never an issue. Article is about CrossFire.

Yeah fair enough - I knew that post was crossfire but i suppose i was also bringing in the other frame latency argument going around that is aimed at single gpu's. Point being, single gpu has been a topical issue but it is hard to replicate, system to system.

As Cadaveca has pointed out it's really hard to pin down and it also affect Nvidia cards (though perhaps not as much).

 

[cadaveca]

Yeah fair enough - I knew that post was crossfire but i suppose i was also bringing in the other frame latency argument going around that is aimed at single gpu's. Point being, single gpu has been a topical issue but it is hard to replicate, system to system.

As Cadaveca has pointed out it's really hard to pin down and it also affect Nvidia cards (though perhaps not as much).

Well, when it comes to AMD cards, and this issue, or whatever it is, single GPU performance is OK, but there is some very varied frame outputs, even in single card. What makes it an issue is that adding another GPU doesn't seem to smooth anything out, and actually makes that variation greater.

However, many users have reported that a third card fixes that, somehow. When I had triple 6950's, it did seem to be the case, but very few drivers work with triple cards.


Honestly, I don't know what is causing the problem, but I do know the symptoms very well. What is new is that the OP covers that this issues isn't just within the VGA, or software, but also in the output given to the monitor, and it seems to be worse over the cable than it is before it's sent out.

AMD's had issues with corrupt cursor, running multiple monitors, etc for some time now, since the 5-series. Maybe it's all related.

 

[crazyeyesreaper]

Im just gonna go out on a limb and blame the APIs that allow the graphics in the first place. Code to the metal you get more control and less overhead but Im not genius and i dont really care either lol.

 

[mastrdrver]

Well, when it comes to AMD cards, and this issue, or whatever it is, single GPU performance is OK, but there is some very varied frame outputs, even in single card. What makes it an issue is that adding another GPU doesn't seem to smooth anything out, and actually makes that variation greater.

However, many users have reported that a third card fixes that, somehow. When I had triple 6950's, it did seem to be the case, but very few drivers work with triple cards.


Honestly, I don't know what is causing the problem, but I do know the symptoms very well. What is new is that the OP covers that this issues isn't just within the VGA, or software, but also in the output given to the monitor, and it seems to be worse over the cable than it is before it's sent out.

AMD's had issues with corrupt cursor, running multiple monitors, etc for some time now, since the 5-series. Maybe it's all related.

I have a question: Is it possible for the monitor to cause, what Ryan is seeing in his video play back, not to show up at all in actual picture? I assume no as I'm under the impression that other then lag and things of that like, the monitor would put up the picture is gets sent over the cable, but I'm just wondering.

 

[cadaveca]

I have a question: Is it possible for the monitor to cause, what Ryan is seeing in his video play back, not to show up at all in actual picture? I assume no as I'm under the impression that other then lag and things of that like, the monitor would put up the picture is gets sent over the cable, but I'm just wondering.

Absolutely. some monitors only support native resolution, but others can do "scaling", or fitting a smaller image to their larger resolution. Then, there's a chance the output, if screwey enough, might screw with the monitor, and cause even more lag than is actually seen via this sort of testing(DisplayPort monitor dropout, for example, with messages about the monitor not supporting the resolution...that it was just working at fine).

Honestly, I can only assume that all these things I notice are related. It's all that makes sense to me at this point. But for al lI know, these are al ldiffernt problems, or problems related to other hardware...I just know there are issues, and no fixes, if these problems are even properly identified.


And that's the kicker. You need to identify, track, and be able to repeat issues like this, on a wide scale, because if the tech team cannot at whoever makes the product...good luck on getting a fix.


That's what makes this item new and interesting...Ryan Shrout finally created a way that AMD can track this issue, and work towards fixing it. There si a clear and concise way for them to see the problem. How or what causes it....and how they fix it...is all way beyond my pay scale. I just like seeing issues get resolved..or in this case, a problem I have been complaining about for years, finally repeated in a way that cannot be refuted.