Crossfire severely flawed?

[harmattan]

It seems there is a wave of reports coming out with evidence that Crossfire is severely flawed. In summary, some frames are not rendered completely although "normal" tests e.g., FRAPS will still record this frame. The end result is the Crossfire experience will actually be worse and less fluid in many cases than running with a single card, even with driver optimizations. This frame rendering issue may be the cause of the "stuttering" that [H] has been reported in several of their past reviews.

http://www.pcper.com/reviews/Graphi...ance-Review-and-Frame-Rating-Update/Frame-Rat

http://techreport.com/blog/24415/as-the-second-turns-frame-captures-crossfire-and-more

Hopefully, more reviewers (looking at you [H}) will start using this new "Frame Capture" method and we'll get confirmation on this. In my own experience with many Crossfire and SLI configs (some of them running side-by-side), there was definitely something "off" with Crossfire in terms of fluidity.

 

[jhokie]

I'm not privy enough about the specifics of framerate / frame capture technologies to make my own judgement on exactly how these technologies work.

I can offer some anecdotal evidence though.

In both BF3 and Crysis 3, I got good results with Crossfire. All framerates observed with FRAPS.

BF3: 1920x1200 Ultra with 2X MSAA and FXAA
1 6950: About 30 FPS. IMO not sufficient for multiplayer in this game.
2 6950: About 60 FPS. It definitely felt far smoother. Really didn't notice any significant stuttering issues.

Crysis 3: 1920x1200 High with Very High textures 2X SMAA (I really like SMAA, I think it is definitely superior to FXAA)
1 6950: About 28-40 FPS. Feels pretty sluggish at times.
2 6950: About 55-79 FPS Feels very smooth most of the time. Occassionally I do get stuttering issues when some large or new graphical effect happens.

Far Cry 3: 1920x1200 Ultra with 2X MSAA and FXAA (not 100% sure on these settings)
1 6950: About 30 FPS. Not really smooth enough for my liking.
2 6950: About 55-70 FPS. Ok big microstutter issues here. It was really annoying. Once I limited the frame rate to 45 in game it was a pretty good compromise. It felt alot better than 30 FPS, and mostly eliminated microstutter issues.

In conclusion, my 6950CF is definitely better than a single 6950. However, some tweaking has definitely been required and at times it has been sub-optimal. I would probably be better off with a current generation card.

If I could do a multi-card solution again, I would probably go with Nvidia, which is kinda sad to say.

 

[pandora's box]

We've had this discussion before (multiple times actually). Ends up with those of us actually running crossfire saying "no issues here".

 

[EPOQ]

I'll be happy to give my own personal results once my cards are in. Card number 1 comes today, and card number 2 next week. I will thoroughly test 1 and 2 card setups based on my own perception of how the games are performing rather than base it on FRAPS FPS numbers. Despite what some of the evidence may have suggested, it really comes down to the individual user and their perception of the cards performance. If you can tell it is obviously MUCH smoother and the frame rate is significantly higher with 2 cards (without even LOOKING at the actual FPS numbers) -- then the previous evidence sounds like a wash to me. Obviously, some tweaking is involved in this. I *WILL* be using a frame limiter and I suggest other CF users do the same, for the best possible experience.

 

[NukeDukem]

I'm sick of hearing this crap. Have run crossfire 5850's, crossfire and trifire 6970's, and now crossfire 7970's since they launched and while there have been a few hiccups along the way I have been quite satisfied with the performance these setups have offered me the past several years.

Right now running the 13.2's and my 7970 CFX smokes everything I throw at it , smooth as liquid butter on a 120Hz 1080p.

 

[pandora's box]

I'm sick of hearing this crap. Have run crossfire 5850's, crossfire and trifire 6970's, and now crossfire 7970's since they launched and while there have been a few hiccups along the way I have been quite satisfied with the performance these setups have offered me the past several years.

Right now running the 13.2's and my 7970 CFX smokes everything I throw at it , smooth as liquid butter on a 120Hz 1080p.

yup its getting old fast...

 

[harmattan]

I'm sick of hearing this crap. Have run crossfire 5850's, crossfire and trifire 6970's, and now crossfire 7970's since they launched and while there have been a few hiccups along the way I have been quite satisfied with the performance these setups have offered me the past several years.

Right now running the 13.2's and my 7970 CFX smokes everything I throw at it , smooth as liquid butter on a 120Hz 1080p.

Not crap. I have had dual desktop GTX 680s and dual 7970s running at the same time/in parallel, as well as dual 7970m (Pitcairn) and 680ms (Kepler) (as well as dual GPU setups from both camps every gen going back to 9800 XT vs 5800 days) -- there's a real noticable difference between how the two technologies render I've never been able to put my finger on.

When [H] consistantly calls out stuttering issues with Crossfire, and other respected publications are starting to get scientific data back proving the issue and its cause, it's worth discussion.

 

[DooKey]

Not crap. I have had dual desktop GTX 680s and dual 7970s running at the same time/in parallel, as well as dual 7970m (Pitcairn) and 680ms (Kepler) (as well as dual GPU setups from both camps every gen going back to 9800 XT vs 5800 days) -- there's a real noticable difference between how the two technologies render I've never been able to put my finger on.

When [H] consistantly calls out stuttering issues with Crossfire, and other respected publications are starting to get scientific data back proving the issue and its cause, it's worth discussion.

I completely agree with you. I got rid of my release day 7970s for a number of reasons and one of the reasons was the damn stuttering. Since I've gone back to Nv I've had a much better experience.

 

[GotNoRice]

One thing I would like to know, is if this is an AMD issue or a GCN issue. Does this occur with crossfire using VLIW4 cards? VLIW5 cards?

 

[DooKey]

One thing I would like to know, is if this is an AMD issue or a GCN issue. Does this occur with crossfire using VLIW4 cards? VLIW5 cards?

Prior to buying the 7970s I was running 6970 Lightnings in CF. I had better results with the 6970s in CF than the 7970s so I am leaning towards a GCN/Driver problem.

 

[Elmy]

I'm really sick and tired of this being posted. Its already been posted multiple times and even brought up in threads when a guy was talking about buying single GPU. I have never had any stuttering problems with my 6970's then 6990+6970 or my current 7970's. Either I'm not sensitive to it or its so small that it isnt really as big as deal as ppl are making it out to be.

 

[rennyf77]

as an owner of gtx680s and 7970s, the smoothness issue is more present on the crossfire setup, but it doesn't look as bad as these recent articles suggest. from the tone of the articles, the crossfire stuttering would make the games unplayable, but i don't find that to be the case. those who exclusively use crossfire really have no frame of reference to try to mute the concerns presented by the articles, but again from my perspective the problem is being embellished beyond what most 7970cf users are experiencing.

 

[Kwisatz Haderach]

Is everyone using FRAPS to capture this? Could it be a problem with the recording software, or something one company may have had added for a fee to get these results? I have had 5870's and now 7970's. It's smooth for me.

 

[trick0502]

this thread has some real promise!!

ive looked over the pcper article on this. if you look at their data there really seems to be an issue here. it also leads to a lot of questions. how much of a frame counts as a frame? looking at some of those little silvers that count as a frame in fraps i feel shouldnt count as a frame.

should that circle count as a frame???

Kyle, Steve, when are we going to see you guys take a [H]ard look at this?

 

[Dark12]

I just upgraded to a pair of tahiti le's from a pair of unlocked 6950s and i can say with certainty that it is not as smooth as it was before. I have cleaned drivers many times but i keep getting inconsistant frame rates. I never had an issue with xfire before so i think it is something to do with the new cards.

 

[Lord_Exodia]

I didn't notice this issues when I had my dual 6950 unlocked to 6970s then overclocked like mad but when I upgraded to my current setup my GTX 680s felt much smoother and at first I just thought that this was because they are much more powerful cards than overclocked 6970s but they do kind of seem unusally smoother even in times when I had similar fps e.g. battlefield 3 where I'm running at higher settings but getting the same fps. 75-85fps on AMD feels like 45-55 fps on SLi. I could never put my finger on it but I wonder if it has to do with these recent reports and the issues that people are suddenly noticing.

Hell microstutter was constantly a complaint back then, a issue that I never had to worry about because I dont seem to be sensitive to it.

I do wish the [H] would step up and investigate this to see if they can find anything interesting with all of this.

 

[Fooshnik]

I expect there'll be people linking to this article and telling others AMD is rendering little mini-frames for a quite a while now. Protip: AMD is not rendering "mini-frames".

I just upgraded to a pair of tahiti le's from a pair of unlocked 6950s and i can say with certainty that it is not as smooth as it was before. I have cleaned drivers many times but i keep getting inconsistant frame rates. I never had an issue with xfire before so i think it is something to do with the new cards.

I don't think the drivers for the Tahiti's are quite polished yet. Going from 6850 crossfire to 7870XT I was surprised to see the microstutter went from being intolerable on a couple games, to most all of them [that I've tried so far]. I was already using Afterburner's frame server for Skyrim and one other game, and after configuring it for the rest the issue went away. There's many tutorials on how to do this, takes about three minutes. Why AMD doesn't include Afterburner's option in their own driver is bewildering.

When [H] consistantly calls out stuttering issues with Crossfire, and other respected publications are starting to get scientific data back proving the issue and its cause, it's worth discussion.

This is hysterical. The dude's a scientist now.

 

[bendic]

One thing I would like to know, is if this is an AMD issue or a GCN issue. Does this occur with crossfire using VLIW4 cards? VLIW5 cards?

Its actually an issue for a while now but until quite recently there was no way to really quantify the problem. As of now the GCN cards have more stutter (non-sli) compared to Nvidia 6 series. AMD said that the problem was made worse by the older drivers which were based on VLIW hardware and beta drivers only offer tweaks which would improve stuttering on a game by game basis. The final driver would be a revamp which would offer a general solution for GCN cards.

TR was the first to explore the frame rate latency problem in detail.
http://techreport.com/blog/24133/as-the-second-turns-the-web-digests-our-game-testing-methods

 

[Final8ty]

I have seen it mentioned that around 20ms is unnoticeable.

You are reading the chart completely wrong.

The variation you see on the Crossfire solution is ALL below 20ms, making it literally imperceptible to the end user.

Honestly, without perspective people can take these charts and have them completely misrepresent the information being provided. A card can do the bloody hokey pokey all over the place below 20ms and it wouldn't make one lick of difference since the actual REAL WORLD experience would still read as smooth.

Again, you are making a blanket statement.

Do rapid fluctuations affect gaming? YES. However, the microstutter you see is likely NOT below 20ms but rather higher than that. Typically when deltas of 10ms or more quickly fluctuate at 25ms and above, you'll see microstutter. Anything below 20ms and you'd either completely miss it or be VERY picky and go looking for it.

Let me reiterate:

- Below 20ms: unnoticeable
- Rapid, cyclical fluctuations above 25ms: you'll notice it
- Sudden "spikes" (non-cyclical) above 40ms: you'll notice it
- Grey area between 21ms and 25ms: debatable. Some will notice it while others won't.

It isn't me who is defining "noticeable". Read any article or scientific paper about perceptive latency and you'll likely come to the same conclusion.

When I make a statement, don't think for a second that I haven't researched it to an extensive degree.

Links to some of said papers:

http://www.stuartcheshire.org/papers/LatencyQuest.html

http://www.ncbi.nlm.nih.gov/pubmed/16639613 (link to PDF @ source)

http://www.perceptionweb.com/abstract.cgi?id=p240749

If you have an account @ Sciverse / Science Direct, there are several other articles in their archives which discuss the same thing. Unfortunately, I can't post links to them due to copyright terms but you'll quickly get the idea after Googling a bit.

However, I will distill it down into plain speak: In extensive lab tests with and without human subjects, a time of 48 frames per second (ie: 20.83ms) was deemed to be the threshold at which the majority of people anbd test instruments saw completely smooth display images or refreshes.

One test actually went so far as to rapidly flash a lightbulb at variable speeds in order to demonstrate when people would see it "flicker". That threshold for 95% of the participants was at 50 flashes per second (ie: exactly 20ms).

Not sure how much more information anyone could be looking for here....

http://www.xtremesystems.org/forums/showthread.php?285164-PCPer-Frame-Rating-Part-3-First-Results

 

[Dark12]

Is there an ETA on AMD's fixed drivers? Will it help those of use with 7 series cards or only 8 and beyond?

 

[Fooshnik]

What fixed drivers?

 

[Dark12]

Oh I thought I read something about drivers being based on old architecture and they were going to revamp them.

 

[Relayer]

Its actually an issue for a while now but until quite recently there was no way to really quantify the problem. As of now the GCN cards have more stutter (non-sli) compared to Nvidia 6 series. AMD said that the problem was made worse by the older drivers which were based on VLIW hardware and beta drivers only offer tweaks which would improve stuttering on a game by game basis. The final driver would be a revamp which would offer a general solution for GCN cards.

TR was the first to explore the frame rate latency problem in detail.
http://techreport.com/blog/24133/as-the-second-turns-the-web-digests-our-game-testing-methods

What TR reported on is in no way connected to what PCPer is reporting on. TR even says it's not the same thing as stuttering caused by AFR with multi GPUs. Don't use one's findings to back the other's.

What [H] reports on when visually examining crossfire (the lack of smoothness and it appearing to run at lower FPS than what is reported by FRAPS) could possibly be explained by PCPer's findings. Keep in mind though that [H] doesn't report it across the board on all games. Many a perfectly fine. This does lead one to believe it's driver optimizations. We need more examination to really know anything, though.

To repeat though, "It has absolutely nothing to do with what TR reported on."

 

[BababooeyHTJ]

We've had this discussion before (multiple times actually). Ends up with those of us actually running crossfire saying "no issues here".

I'm not one of them. I clearly see more microstutter in crossfire than I ever did with sli. I have to use a framerate cap with any intensive game. If it weren't for the ability to do that I would have sold these cards a long time ago.

This is without a doubt something that AMD really needs to focus on.

To repeat though, "It has absolutely nothing to do with what TR reported on."

Yes and no, it may not be the same root cause but I really do think that a solution similar to Nvidia's framerate metering could be the solution for the both.

I think that memory optimization crap was a line of bs. Look at techreport's reviews with 6xxx and 5xxx series cards. Frametimes were awful in many games. The memory management system wasn't optimized for those cards either? Come on.

Microstutter has been a huge issue with crossfire forever. Its about time that they look for a real solution.

One thing I would like to know, is if this is an AMD issue or a GCN issue. Does this occur with crossfire using VLIW4 cards? VLIW5 cards?

Hell yes, microstutter was an issue on older cards. It was awful on 4870x2, just plain awful. In fact in crysis I though that my GTX280 was smoother. I ran into microstutter in a few games with 6950 crossfire. It tends to be an issue in games that are more demanding on the gpu. If you enable supersampling in the ccc you'll see microstutter in most games.

I have seen it mentioned that around 20ms is unnoticeable.

SkyMTL is so full of shit that its not even funny. I would love to see him posting that shit in the display section here. There are far more knowledgeable posters on the topic here than he is. He also didn't post a single link to back up those claims.

I expect there'll be people linking to this article and telling others AMD is rendering little mini-frames for a quite a while now. Protip: AMD is not rendering "mini-frames".

You're right, thats a full frame. Its just that the next frame is getting sent to the display too soon for much of it to be displayed. Thats why they need some sort of framerate smoothing similar to what Nvidia is doing.

 

[Final8ty]

SkyMTL is so full of shit that its not even funny. I would love to see him posting that shit in the display section here. There are far more knowledgeable posters on the topic here than he is. He also didn't post a single link to back up those claims.

That's all a matter of opinion.

 

[x3sphere]

Is there an ETA on AMD's fixed drivers? Will it help those of use with 7 series cards or only 8 and beyond?

This has been going on since at least the 5xxx, doubt it will be fixed anytime soon (if ever), or at least until there is a breakthrough in multi GPU tech. I think the technique Nvidia uses to smooth out frames comes at the cost of less FPS so AMD will end up looking worse in dual GPU benchmarks and it probably doesn't want that.

I used both 58xx and 69xx CFX setups in the past and both were horrible, 30FPS felt like 15 to put things in perspective. The single 7970 I have now dramatically improved my gameplay experience in a lot of games despite rendering less frames than my 6970x2 setup.

 

[BababooeyHTJ]

That's all a matter of opinion.

Its really not. He is wrong on that topic. Its like the people who say microstutter isn't there since they don't notice it.

 

[ShuttleLuv]

This is the real deal. I trust Ryan and Kyle on this bc I have seen it myself as well.

 

[NukeDukem]

I'll take the microstutter, because I know AMD cards produce better colors and crisper text than nVidia cards! If you can't see it doesn't mean it isn't true, right?

 

[Relayer]

Yes and no, it may not be the same root cause but I really do think that a solution similar to Nvidia's framerate metering could be the solution for the both.

I think that memory optimization crap was a line of bs. Look at techreport's reviews with 6xxx and 5xxx series cards. Frametimes were awful in many games. The memory management system wasn't optimized for those cards either? Come on.

Microstutter has been a huge issue with crossfire forever. Its about time that they look for a real solution.

You must not have read what I posted. Or didn't understand. HERE Scott Wasson of TechReport says that the frame latency issues and dual GPU microstutter aren't one in the same.

Also, a word on words. Although I'm reading a Google translation, I can see that they used the word "microstuttering" to describe the frame latency issues on the Radeon. For what it's worth, I prefer to reserve the term "microstuttering" for the peculiar sort of problem often encountered in multi-GPU setups where frame times oscillate in a tight, alternating pattern. That, to me, is "jitter," too. Intermittent latency spikes are problematic, of course, but aren't necessarily microstuttering. I expect to fail in enforcing this preference anywhere beyond TR, of course.

This is why I said don't use TR's reports when refering to dual GPUs. They weren't running dual GPUs when the tests were conducted. AMD's memory management rewrite isn't to address crossfire issues. To say that it's BS because crossfire has been messed up forever is wrong. The issue they said it addressed was with single GPUs, not crossfire.

 

[BababooeyHTJ]

You must not have read what I posted. Or didn't understand. HERE Scott Wasson of TechReport says that the frame latency issues and dual GPU microstutter aren't one in the same.

I don't think that you understood what I posted. Yes, they aren't the same issue but some sort of framerate metering similar to what Nvidia uses may be the solution to both issues. I don't know how else to word it.

 

[Flopper]

Is there an ETA on AMD's fixed drivers? Will it help those of use with 7 series cards or only 8 and beyond?

with 7 series card they need to tweak for each game in the driver.
I assume that wont be needed for the 8 series where a MORE GENERIC approach will work due to better design.

Its not just drivers, as the game engine and devloper also need to support things along the way. Multi monitor, multi gpu etc..
all in all it will become better with the consols going apu and amd down the line.

I find a single 7970 oc covers my need at 5040x1050.

 

[magoo]

The pcperspective article was interesting.
I saw all the nifty graphs.
Bottom line....
If its important to you, buy nvidia or whatever. If it isn't its just Internet debate fodder.

I have both company,s cards and I don't see a difference.

 

[maxius]

truthfully this microstutter debate smells like bullshit nvidia marketing to try to make amd's card look inferior cause they are loosing to them in almost every benchmark. Oh they beat us cause they dont handle frames the way we do... to the average joe you will hardly notice.

 

[BababooeyHTJ]

Yeah, except that the debate has been going on for years now. This isn't an issue that cropped up overnight.

Its in the consumer's best interest that AMD actually tries to address this issue.

 

[Ulysses 31]

I've had a lot of microstutter issues in the past but (forcing) v-sync fixed that usually, and now I use the fps limitter of Afterburner in Windows 8.

At the moment only Borderlands 2 still continues to feel very low fps despite running at 60 fps in Eyefinity + CFX.

 

[Elmy]

I have dual 7970's and never notice micro-stutter one bit. Its not as big as a deal as some ppl are making it out to be. Read thru these forums and see how many ppl are also not experiencing any issues at all like me. But then you read thru these forums and ppl say its there and when you look at their sig they are running Nvidia.... Makes me LoL

Maybe they did run AMD and switched to Nvidia because they are sensitve to the micro-stutter. IDK but alot of ppl here are happy as can be with crossfire.

 

[Fooshnik]

I've had a lot of microstutter issues in the past but (forcing) v-sync fixed that usually, and now I use the fps limitter of Afterburner in Windows 8.

At the moment only Borderlands 2 still continues to feel very low fps despite running at 60 fps in Eyefinity + CFX.

Afterburner+vsync worked well for me on B2 (didn't try with just Afterburner). Is FRAPS still showing framerate always 59/60?

I'm not one of them. I clearly see more microstutter in crossfire than I ever did with sli. I have to use a framerate cap with any intensive game. If it weren't for the ability to do that I would have sold these cards a long time ago.

This is without a doubt something that AMD really needs to focus on.

This. I didn't have much issue with xfire on 6850s, had to use Afterburner with two games, but 7870XT it's a problem for most games. If it weren't for Afterburner this wouldn't work for me. Why AMD spends $100 million+ on driver development and doesn't include an option that Afterburner has as a side effect is bewildering. They could even call it something newbish like "Smooth Crossfire Display". Personally I need xfire as no single card will run 2650x1600 with the settings I like to use at 60fps min.

 

[Relayer]

Yeah, except that the debate has been going on for years now. This isn't an issue that cropped up overnight.

Its in the consumer's best interest that AMD actually tries to address this issue.

This is true. It seems to be on a per game basis. Reading [H]'s reviews there are games that play fine and others that don't. Crossfire does seem to be more problematic than SLI, no doubt. Maybe PCPer's info can shine some light on it that AMD hasn't seen before.

I have dual 7970's and never notice micro-stutter one bit. Its not as big as a deal as some ppl are making it out to be. Read thru these forums and see how many ppl are also not experiencing any issues at all like me. But then you read thru these forums and ppl say its there and when you look at their sig they are running Nvidia.... Makes me LoL

Maybe they did run AMD and switched to Nvidia because they are sensitve to the micro-stutter. IDK but alot of ppl here are happy as can be with crossfire.

Yes, PCPer's response of not including any Crossfire results make it seem like this is an issue with all crossfire benches, period. Lot's of reviews and personal experience have said otherwise though, where 2 cards have visually performed better than one, and as good as SLI. Just not all of the time.

 

[Drangueos]

I have 2 7950 in crossfire for almost a year now, and I had Micro stuttering just in one game "APB Reloaded" and only if I use the catalyst profile, If I use FEAR profile the game runs OK. (and I blame the game, because it has problems in SLI as well.

I played bf3, Cod BO2, crysis 2 and 3, Dirt 3, Alan Wake, Dead Space 3, Mass Effect 3, FarCry3, Hitman Absolution, Sleeping Dogs and lots more with crossfire and eyefinity, never had Micro stuttering.

The only problem I had with crossfire is: if I leave ULPS enabled when the 2nd card kicks in sometimes my computer was freezing.