[H] Yet again say SLI is smoother than crossfire

[Keysplayr]

Could explain in part the big performance improvements out of the "never settle" drivers. AMD chose to optimize throughput at the expense of orderly output. Optimize drivers for what most reviews show - absolute frame rates - and hope that the more frequently occuring hiccups aren't noticed or are overlooked because only a small handful of websites investigate beyond averages and minimums

Agreed. It could explain in part the big boost from the 12.11's was it? Mondo increase in framerates and mondo increase in erratic framerate throughput.
Brent mentioned that higher framerates helped alleviate the symptom on the AMD Crossfire setup. Requiring about 60 to 70 fps to feel smooth where as Nvidia would need only 40, depending on game.

 

[tviceman]

Any plans to use a high speed camera and replay things in slow-mo to get an idea of what's really happening in terms of frame time? Even a Nikon J1 ought to be sufficient and that's $250 or so, so it should be possible.

This would be the absolute best way to test this out. Techreport is also showing some big latencies with a single hd7950 in their latest "review". I know I said this a few posts back, but if this is measurable think AMD's never settle driver increased throughput at the expense of consistency.

 

[Will Robinson]

Man the performance increase from the Never Settle drivers really stings doesn't it...:whiste:
What luck you and Keysplayr have worked out the real reason for the boost....they musta crippled their IQ and thrown smoothness out the door to achieve it...damn...who knew...

 

[BoFox]

Some seem to be forgetting NVIDIA does employ technology to smooth SLI, whether this be frame rate, or frame time, I do not know, but the fact is NVIDIA uses technology to do some sort of smoothing. That is simply a fact.

Some people are more sensitive to detecting smoothness/consistency differences between SLI and CrossFireX. Like any other human sense, there are varying degrees of sensativity. Some people can detect it, while others cannot. It is analogous to FPS, some people are sensative to FPS over 60 FPS, and some simply don't feel it.

But the fact remains that NVIDIA is doing something to make SLI smoother in gameplay experience, this is known information.

So when I say I "feel" it, perhaps feel is a bad descriptive word (but it is lack of a better description that exists), it is a combination of my eyesight/brain/mouse/keyboard input, all wrapped up together that translates to "feeling" different. Descriptive word aside, you put it all together, and it all points to the same conclusion.

There is still microstuttering with all GTX 6xx SLI setups. GTX 690 ameliorates the microstuttering severity in most cases by reducing the "latency" of microstuttering by roughly 50% compared against dual GTX 680 SLI.... yet GTX 690 has not yet completely resolved microstuttering.

It'd be nice if [F]laccidOCP could test 25 games like the other site, rather than 3-5 games like usual and then reaching a sweeping conclusion! :sneaky:

 

[Vesku]

Could explain in part the big performance improvements out of the "never settle" drivers. AMD chose to optimize throughput at the expense of orderly output. Optimize drivers for what most reviews show - absolute frame rates - and hope that the more frequently occuring hiccups aren't noticed or are overlooked because only a small handful of websites investigate beyond averages and minimums

Techreport said in the conclusion that 12.11 actually improves on latency from 12.8

"I can tell you that we've seen the same spiky frame time plots in most of these games from three separate revisions of AMD's drivers—and, yes, Catalyst 12.11 is an improvement over 12.8, all told, even if it doesn't resolve the latency issues. " - http://techreport.com/review/23981/radeon-hd-7950-vs-geforce-gtx-660-ti-revisited/11

Not sure why they used 2560x1440 on some games but 1920x1080 for others. Those two cards are mainstream 1080P gaming cards.

 

[VulgarDisplay]

Agreed. It could explain in part the big boost from the 12.11's was it? Mondo increase in framerates and mondo increase in erratic framerate throughput.
Brent mentioned that higher framerates helped alleviate the symptom on the AMD Crossfire setup. Requiring about 60 to 70 fps to feel smooth where as Nvidia would need only 40, depending on game.

That makes no sense because crossfire had more stutter long before 12.11 was released.

 

[ICDP]

Techreport said in the conclusion that 12.11 actually improves on latency from 12.8

"I can tell you that we've seen the same spiky frame time plots in most of these games from three separate revisions of AMD's drivers—and, yes, Catalyst 12.11 is an improvement over 12.8, all told, even if it doesn't resolve the latency issues. " - http://techreport.com/review/23981/radeon-hd-7950-vs-geforce-gtx-660-ti-revisited/11

Not sure why they used 2560x1440 on some games but 1920x1080 for others. Those two cards are mainstream 1080P gaming cards.

Something is wrong somewhere then. Here is Techreports results from June showing Skyrim.

http://techreport.com/review/23150/amd-radeon-hd-7970-ghz-edition/7

7950 back in June had 0 (zero) time spent beyond 50ms. Somehow it has now gone to 82.

7950 back in June had a 99th percentile frame time of 18. Now its 25.4

If this review is accurate something between then and now has changed. If it is worth anything I notice no such latency when comparing a GTX 680 and a HD 7970. If anyone wants to test their frame latency they can use fraps to get results for comparison.

 

[Silverforce11]

Since TR did their first report on microstutter with frame time measurements, NOBODY has shown with actual proof that the monitor draw rate is linked to frap's measure of frame time rendering. Actual output at the monitor is the ONLY thing that matters since its what we see.

Even in their original report, they noticed that the 6870 in some games had much worse frame time but was smoother in gameplay. Huh?

If they want to get serious, get a good camera and capture high speed monitor outputs.

 

[Vesku]

Something is wrong somewhere then. Here is Techreports results from June showing Skyrim.

http://techreport.com/review/23150/amd-radeon-hd-7970-ghz-edition/7

7950 back in June had 0 (zero) time spent beyond 50ms. Somehow it has now gone to 82.

7950 back in June had a 99th percentile frame time of 18. Now its 25.4

If this review is accurate something between then and now has changed. If it is worth anything I notice no such latency when comparing a GTX 680 and a HD 7970. If anyone wants to test their frame latency they can use fraps to get results for comparison.

"We did, however, decide to mix things up by moving to a new test area. Instead of running around in a town, we took to the open field, taking a walk across the countryside. This change of venue provides a more taxing workload than our older tests in Whiterun. "

I don't notice any issues on my 7950 @ 1080P even though I seem to be a bit on the sensitive side when it comes to microstutter.

Since TR did their first report on microstutter with frame time measurements, NOBODY has shown with actual proof that the monitor draw rate is linked to frap's measure of frame time rendering. Actual output at the monitor is the ONLY thing that matters since its what we see.

Even in their original report, they noticed that the 6870 in some games had much worse frame time but was smoother in gameplay. Huh?

If they want to get serious, get a good camera and capture high speed monitor outputs.

I agree, considering there are relatively inexpensive high speed capture options nowadays.

 

[BrightCandle]

Techreport also noted in their original review what it was that frame time measured, and it isn't everything that needs to be in there. Ideally what you want is the start of a games calculation for a moment in time to be the beginning and the moment its displayed to be the end. Currently you can't measure either point. What TR do measure is hand of to the GPU and when it syncs to the monitor.

So while frame times sometimes gives a good indication of stuttering it isn't actually as scientific as we need. I think its time someone asked microsoft to include a hook to understand when a game frame started so we can track everything up to the point it leaves the GPU. The latency of monitors can be measured separately so we can eliminate that part of the overall picture.

 

[Jacky60]

I knew it would be brought up. Hey if you are happy with your setup, that's all that counts. This idea that Nvidia 'markets' their way to success is lame. They don't do Apple like commercials, they sponsor gaming LAN events, fund game development, proactive driver support.

Marketing shenanigans! You Nvid-sheeple! /sarcasm

Apple's marketing is far more sophisticated than simply broadcasting commercials. 'Below the line' advertising/marketing including sponsorship, promotions and public relations (for example giving free product to recognised industry 'players') works at a subtle but very tangible level.
The whole point of such marketing is that the recipients of the persuasive message don't realise they're being sold to. This has been around since the 1960's or earlier and is nothing new. A huge number of companies do market their way to success, Nvidia supplies kit-gaming machines etc to industry game launches etc for free. Do you think a computer journalist sitting at an Nvidia supplied Alienware branded desktop at a game launch doesn't subtly absorb the message. I understand this because I've helped organise and produce such events and have also been a tech journalist.
Clearly they can't just use marketing but its a very important part of their success. To believe marketing isn't important for the huge majority of today's successful companies is naive in the extreme or just wilfully ignorant.

 

[Keysplayr]

Man the performance increase from the Never Settle drivers really stings doesn't it...:whiste:
What luck you and Keysplayr have worked out the real reason for the boost....they musta crippled their IQ and thrown smoothness out the door to achieve it...damn...who knew...

What probably stings more is that the public is finding out how they got a "fps" increase. Let's talk about "performance" as a separate topic.

I know you hate this stuff, Will, but I know you have the strength to suffer through it. Hang in there bud. :thumbsup:

 

[Keysplayr]

That makes no sense because crossfire had more stutter long before 12.11 was released.

Then why didn't we hear about all this "smoothness" talk back then?
That makes more sense than your comment. Doesn't it?

 

[The Alias]

Then why didn't we hear about all this "smoothness" talk back then?
That makes more sense than your comment. Doesn't it?

I don't think many people besides nvidia shills should listen to you . It's been proved long ago that the increases from 12.11 were indeed legit after being tested thoroughly . It seems you take your job very seriously ad bash amd every chance you get, even if what you're bashing isn't that important

 

[TakeNoPrisoners]

It just seems to me that any form of Multi-GPU isn't really a good idea. Considering how driver dependent it is and no matter what it will have microstutter regardless of what anyone does.

It just seems to me that it is best to avoid monitor configurations that require multi-GPU such as eyefinity. Partly because of all this I now think it is best to go for a 2560x1440 monitor over an eyefinity configuration. High end single GPU setups can handle that resolution reasonably well while they fall flat when multi-monitors are taken into consideration.

 

[KCfromNC]

Then why didn't we hear about all this "smoothness" talk back then?

You're saying no one mentioned micro-stutter before the middle of 2012?

 

[Hitman928]

Something is wrong somewhere then. Here is Techreports results from June showing Skyrim.

http://techreport.com/review/23150/amd-radeon-hd-7970-ghz-edition/7

7950 back in June had 0 (zero) time spent beyond 50ms. Somehow it has now gone to 82.

7950 back in June had a 99th percentile frame time of 18. Now its 25.4

If this review is accurate something between then and now has changed. If it is worth anything I notice no such latency when comparing a GTX 680 and a HD 7970. If anyone wants to test their frame latency they can use fraps to get results for comparison.

As was mentioned, they switched scenes, however, they also switched from Windows 7 to Windows 8 which may be an issue as well. Personally, I'd like for them to go back and test the old scene under Windows 8 and see if they measure drastically increased latency again. If they do, AMD cards have an issue in Windows 8. It's hard for me to imagine that changing from a town scene filled with people to a vacant outdoor scene would have such a huge impact on latency, but I guess anything's possible. My money's on something else going on though,

edit: Also, it kind of makes me sad that they don't follow up with this at all (or haven't yet). Any respectable review site should see this huge discrepancy and try to figure out what's going on, especially since this was a revisit review and so not nearly as much of a time crunch. Most places I would hope would realize that there may be more than meets the eye going on and try to get to the bottom of it before coming to a conclusion. That would be what I expect from a tech journalist anyway. I guess they tried a few things but didn't try to replicate their original test conditions? That would be the easiest way to see if its legit or an issue with how things were setup this round.

 

[Arzachel]

Then why didn't we hear about all this "smoothness" talk back then?
That makes more sense than your comment. Doesn't it?

Because Nvidia is now doing much worse from a performance standpoint, it's clearly a conspiracy!

...or maybe you should try taking off that tinfoil hat of yours. While the way [H] conduct their tests is perfectly valid, it's subjective by definition. The only data point you have is TechReport and they use frame data provided by Fraps, which we're not sure is accurate or even worse - have a fair reason to doubt its accuracy.

Can you in good faith say that:

• Fraps reports frame data accurate to what's actually being output by the monitor
• The fluctuations in frame display are more pronounced for GCN compared to Kepler GPUs, outside the margin of error
• These fluctuations have increased in 12.11 drivers compared to previous revisions

Because otherwise you're talking out of your ass.

Seeing some/more actual investigative journalism on this would be nice because this field is sorely lacking some (journalism as a whole is too, but that's a discussion for another day).

 

[felang]

It just seems to me that any form of Multi-GPU isn't really a good idea. Considering how driver dependent it is and no matter what it will have microstutter regardless of what anyone does.

It just seems to me that it is best to avoid monitor configurations that require multi-GPU such as eyefinity. Partly because of all this I now think it is best to go for a 2560x1440 monitor over an eyefinity configuration. High end single GPU setups can handle that resolution reasonably well while they fall flat when multi-monitors are taken into consideration.

+1

Especially if you like to play newly released games as soon as they come out. It´s a PITA to have to wait for updated drivers or forcing semi-working sli or crossfire profiles.

I currently have a 2560x1440 monitor and GTX 680 and Couldn´t be happier, if I feel I need a bit more performance I´ll just lower AA until I achieve the FPS I´m looking for. Also, 2GB seems to be just enough for this resolution, eyefinity or 30 inch monitors would probably need more.

 

[Lepton87]

I don't think many people besides nvidia shills should listen to you . It's been proved long ago that the increases from 12.11 were indeed legit after being tested thoroughly . It seems you take your job very seriously ad bash amd every chance you get, even if what you're bashing isn't that important

Why do you people even bother to talk to nv focus group members about competitor's products? Focus group members should be banned from such topics by everybody ignoring them from discussions about cross-vendor things relating to AMD GPUs. If nobody answered to them I think they would stop bashing AMD cards quite fast.

 

[Silverforce11]

High end single GPUs can't even handle recent games at 1080p.. forget about 1440p or 1600p. To play at 1080p smoothly, you already have to tone down the settings a lot.

SLI already has a frame time limited in their software, it stutters less as claimed by a lot of users, i have no reason to doubt them. But CF with Radeon Pro is stutter free, im gonna get it and test if its features improve smoothness of single cards too later when i get home from work. This program seems to work miracles, that and it as a fully functional OSD so no need to load afterburner anymore.

 

[BrightCandle]

+1

Especially if you like to play newly released games as soon as they come out. It´s a PITA to have to wait for updated drivers or forcing semi-working sli or crossfire profiles.

I currently have a 2560x1440 monitor and GTX 680 and Couldn´t be happier, if I feel I need a bit more performance I´ll just lower AA until I achieve the FPS I´m looking for. Also, 2GB seems to be just enough for this resolution, eyefinity or 30 inch monitors would probably need more.

I have run a 4870X2, a 5970, a pair of 7970's and a pair of GTX 680's and to some extent I have to agree. The dual cards are a lot of aggravation compared to a single one, its not a technology that just works. But microstutter isn't limited to dual cards, its just so happens to be worse on them. I remember a time when I was happy with 30 fps, I am no longer fooled by the image motion going below 45 on a 7970, yet on a 680 I am happy down to 30 still. But with NVidia SLI I am only happy at 50 and above. There is definitely more microstutter today than there was 10 years ago and until its exposed and measured we can't hope to get rid of it.

We need a genuine way to measure this, we need to know when a frame was started to be calculated by the CPU and ended by syncing out to the monitor. Once we have the combined picture of CPU + GPU, latency and overall frame render time only then can we understand why NVidia's cards produce a smoother movement than AMDs current cards.

I want ideas about how to go about getting this. I can write the software that measures it given the appropriate hooks exist, but I don't think they currently do.

 

[Joje]

It's annoying :/

 

[VulgarDisplay]

High end single GPUs can't even handle recent games at 1080p.. forget about 1440p or 1600p. To play at 1080p smoothly, you already have to tone down the settings a lot.

SLI already has a frame time limited in their software, it stutters less as claimed by a lot of users, i have no reason to doubt them. But CF with Radeon Pro is stutter free, im gonna get it and test if its features improve smoothness of single cards too later when i get home from work. This program seems to work miracles, that and it as a fully functional OSD so no need to load afterburner anymore.

I'm really interested to hear how it works with a single card.

Also, LOL at microstutter not being talked about until July 2012. You are either completely not paying attention to computer hardware, or just trying to be inflammatory.

 

[Keysplayr]

You're saying no one mentioned micro-stutter before the middle of 2012?

Is that what you're going to interpret my comments as? Have at it. You'll have fun conversing alone.
I'm saying EXACTLY what my words say. As I usually do.