[H] Yet again say SLI is smoother than crossfire

[bononos]

Thanks. And that was with 12.3 drivers. So After the 12.11's It must be getting more pronounced if now TR is reporting on it. I wonder who else will examine this phenomena.

Or not since TR mentioned that 12.11 is an improvement over 12.8 even if the latency issue isn't resolved. That to me sounds like the latency issue is not getting worse or possibly getting better with the new drivers.
http://techreport.com/review/23981/radeon-hd-7950-vs-geforce-gtx-660-ti-revisited/11

Vesku made the same point earlier in this thread when someone brought up the question that the 12.11 drivers could have improved framerates at the expense of latency (whereupon you concurred in your response).

 

[Keysplayr]

Bononos, what you just said isnt quite clear to me. I am maintaining
The position (for now) that from the 12.3's, the 12.11's have improved
framerate at the expense of latency (smoothness).
This goes for 12.8 to 12.11 as well.
So I am not clear on what your "Or not" statement means. Are you seeing what I mean?

 

[Vesku]

I haven't seen the kind of FRAPs results Techreport listed running 12.11 Beta 8 on Win 7 x64, but I run 1080P and they tested at 2560x1440. They really should do a more thorough study before drawing conclusions.

 

[NIGELG]

Wow...two sites out of dozens or hundreds worldwide are saying this crap.I go by the numbers not questionable conclusions from subjective editors who cannot even measure or quantify their 'findings' .

 

[Elfear]

From earlier comments it sounded like the only Nvidia card to get the frame-metering technology was the GTX 690. If that's the case, why would the other SLI solutions feel smoother as well?

 

[SirPauly]

There is software frame metering and there is hardware frame metering -- hardware frame metering may be used with the GTX 690, from my understanding.

 

[notty22]

There is software frame metering and there is hardware frame metering -- hardware frame metering may be used with the GTX 690, from my understanding.

And/ Or in the bios of all(some?) current GTX gpu's. I have also read about a metering system, but if a tear down of a particular card shows , no, unique lC (logic chip), then it would have to be handled by the gpu. In conjunction with drivers.

 

[SirPauly]

I have no idea! Personally asked a question to nVidia about hardware frame metering with Kepler in the PC Perspective Live Review and Chat of the GTX 690:

The response I did get was basically this, " SirPauly, good to see you here and sorry, we're not here today to discuss hardware frame metering!"

Hehe, at least I did get a response but they're really cagey on this and real tough to receive information about it.

 

[omeds]

I'd like to know this aswell, I was aware only the GTX690 uses hardware metering, but whats the details of software metering? My guess is through drivers, as after some driver updates certain games can seem better (or worse).

 

[SirPauly]

http://techreport.com/review/21516/inside-the-second-a-new-look-at-game-benchmarking/11

 

[Elfear]

http://techreport.com/review/21516/inside-the-second-a-new-look-at-game-benchmarking/11

Interesting read. Thanks for posting.

Sounds like both companies are aware of the problem and Nvidia has had some sort of frame-metering since the G80. If there really is a difference in smoothness between both teams multi-gpu solutions, I hope AMD has their own solution soon. Maybe that could be as easy as hiring the RadeonPro dev?

 

[Final8ty]

Interesting read. Thanks for posting.

Sounds like both companies are aware of the problem and Nvidia has had some sort of frame-metering since the G80. If there really is a difference in smoothness between both teams multi-gpu solutions, I hope AMD has their own solution soon. Maybe that could be as easy as hiring the RadeonPro dev?

AMD is sponsoring the RadeonPro dev.

I think the reason why such functionality like NVinspector, RadeonPro or not fully available in the control panels is because the average person does not use it, so they leave it to the enthusiasts, its constant badgering that gets things put in.

 

[omeds]

http://techreport.com/review/21516/inside-the-second-a-new-look-at-game-benchmarking/11

Thanks Pauly, you are the Yoda of video card info

 

[SirPauly]

Thanks Pauly, you are the Yoda of video card info

What's nice about that article is the site offers their thinking on offering more than just raw frame-rate:

http://techreport.com/review/21516/inside-the-second-a-new-look-at-game-benchmarking

 

[Lepton87]

Thanks Pauly, you are the Yoda of video card info

That is such an old article, I thought everybody interested in MS read it by now.

BTW. If GTX690 had some sort of HARDWARE metering technology it would be all over nv marketing just as PSYX. It would entice people to go with GTX690 instead of some much cheaper but just as fast solutions like OC 7950CF or even OC 670s. That would mean more money for NV. I still remember a rumor that 4970X2 would use some side port or something like that and be superior to 2x4970, but it never came to be even tough it really had some side port in hardware which was later removed from Evergreen. Apart from nV not missing any marketing advantage, the pictures of GTX690 PCB are widespread, show me which IC would be responsible for that?
UPDATE: It was Crossbar switch not side port.

 

[omeds]

That is such an old article, I thought everybody interested in MS read it by now.

Well I havent read it, and it was good he had the link almost instantly, as he does with almost any question I ask him concerning GPU's.

BTW. If GTX690 had some sort of HARDWARE metering technology it would be all over nv marketing just as PSYX.

You mean like this?

 

[Lepton87]

Well I havent read it, and it was good he had the link almost instantly, as he does with almost any question I ask him concerning GPU's.




You mean like this?

I didn't know they mention it, but where is the IC responsible for that? Now I want it compared to 2x680 at the same clocks, preferably with a high speed camera to see if it really is smoother than 2x680.

 

[omeds]

I didn't know they mention it, but where is the IC responsible for that?

No idea.

 

[SirPauly]

Here is an investigation with a GTX 690 and touches on MS:

http://www.tomshardware.com/reviews/radeon-hd-7990-devil13-7970-x2,3329-8.html

 

[Lepton87]

Here is an investigation with a GTX 690 and touches on MS:

http://www.tomshardware.com/reviews/radeon-hd-7990-devil13-7970-x2,3329-8.html

I wanted it compared to 2x680 not 7970CF as it doesn't cast any light on that hardware metering. I want to see if that hardware metering isn't some kind of marketing trick and that it is indeed some actual hardware and that it really has advantage over 2x680 in smoothness. Right now I can't see any special silicon on 690's PCB that could do that.

 

[Siberian]

I'm more interested in trying SLI now. It's good to see that it works smoothly. The early days of multi GPU had me keeping my distance.

 

[RussianSensation]

But looking at those TechReport data,

You must have not looked at the review objectively or read the comments at TR. Other people have noted this too. I'll just restate some of those comments briefly. If you did, you would have noticed something that simply cannot be true - GTX660Ti outperforming HD7950 in Skyrim in terms of frames per second at 2560x1440 with AA. A quick look at other reviews on the net shows that this is frankly impossible because HD7950 V2 not only faster than GTX670, but can beat 680 in Skyrim at high resolution and the performance gap is > 30% compared to GTX660Ti in this title at high resolution.

http://www.computerbase.de/artikel/grafikkarten/2012/test-amd-radeon-hd-7950-mit-925-mhz/19/

That's 6 reviews that all show HD7950 V2 smoking GTX660Ti in Skyrim at high resolution and I can find 20 more and none of them will show GTX660Ti achieving higher frames per second at 2560x1440 or 2560x1600 in Skyrim because it's one of those cases where GTX660Ti gets slaughtered. GTX660Ti simply cannot beat HD7950 950mhz in actual frames per second in Skyrim at 2560x1440 with AA, irrespective of smoothness discussion. Their review shows it garnered higher frames per second. :sneaky: At that point, you pretty much know that either there is a system wide hardware problem, or a Windows 8 driver problem, or a motherboard BIOS/chipset comparability issue with their Radeon cards. The rest of the review can no longer be taken seriously because of this result.

it kinda makes even more sense why H states SLI "feels" smoother than CFX.

This is not news. Out of the box SLI feels smoother than CF. If you test CF with Radeon Pro, what happens then? You have already been linked the Tom's hardware review where micro-stutter was eliminated effectively. 3rd party apps like MSI Afterburner, NV Inspector are fair game and so is Radeon Pro or its dynamic vsync feature.

Now it makes more sense when hearing people feel NV is smoother than AMD.

I went from 8800GTS to 4890 to GTX470 to HD6950 @ 6970 to HD7970. I felt no difference in micro-stutter. I did notice a huge improvement in frames per second in each case in certain games. My HD6950 OC crushed GTX470 @ 760mhz in Crysis/Warhead and F1 2010. If GTX470 felt smoother, I wouldn't have kept the 6950, instead I would have sold it and grabbed a GTX570 and I certainly wouldn't have gone with HD7970 if I felt NV cards were seriously smoother in single card configurations.

I am surprised NO ONE mentioned that in HardOCP's testing, the HD7970CF setup was providing higher image quality. For example, in Sleeping dogs, they were able to put AA+FXAA on high and enable SSAA to medium. GTX680 SLI could not render the game smoothly with those settings.

So on one hand, GTX680 felt "smoother" but it couldn't run the game at the same image quality? I see that GTX680 SLI had higher frames under the playability comparison:

And that's the point, if you level the playing field to apples vs. apples image quality, the HD7970CF crushes GTX680 SLI in performance:

If you want to improve smoothness, no problem, just get Radeon Pro!! You are getting 42% higher frames per second which means you have room to afford vsync here. You either end up with higher image quality than GTX680 SLI or higher performance, where you can use Radeon Pro to virtually eliminate micro-stutter. It's the same thing for all the other game tests they ran. HD7970 Ghz CF and 7970Ghz provided either more frames per second or higher image quality compared to GTX680 SLI or GTX680. It's not hard to imagine that GTX680 SLI would feel smoother in BF3 when it's doing 2xMSAA and HD7970 CF is subjected to 4xMSAA. So lower it to 2xMSAA and you get way higher frames on the 7970 CFX. The micro-stutter concerns can be easily fixed with Radeon Pro. You don't really have a good argument there.

 

[RussianSensation]

Bononos, what you just said isnt quite clear to me. I am maintaining
The position (for now) that from the 12.3's, the 12.11's have improved
framerate at the expense of latency (smoothness).
This goes for 12.8 to 12.11 as well.

:sneaky: The elephant in the room = frames per second. TR's review shows GTX660Ti outperforming HD7950 950mhz at 2560x1440 in titles where this is impossible and then shows higher fps average in the conclusion. Yet other reviewers show a 950mhz HD7950 being > 20% faster at high resolution in terms of frames per second. Please explain this.
http://www.techpowerup.com/reviews/HIS/HD_7950_X2_Boost/28.html

Smoothness is 1 factor of the discussion but frames per second still help us notice if something is completely illogical. Frames per second don't just disappear. In TR's review, they did. Your and tviceman arguments are contradictory: You both say that AMD is "cheating" by increasing frames per second at the expense of stuttering, but the TR review shows that GTX660Ti is providing higher frames per second, exactly contrary to basically every modern review on the planet!! I am sorry but GTX660Ti does not suddenly start outperforming HD7950 V2 in frames per second at 2560x1440 unless something is completely broken like drivers, OS.

"This plays out to AMD's advantage. At 1920x1200 the GeForce GTX 670 is 12% faster than the Radeon HD 7950 and just 2% faster than the 7950 Boost. However the GTX 670 is 27% more expensive than both cards, so whichever way you slice it the Radeon HD 7950 Boost is the better proposition and things just get worse for the GTX 670 as the resolution is increased." ~ Source This review was just done recently.

It could be even as trivial as running FRAPs in some of those games causing the actual stuttering. If you play videogames, there are instances where enabling FRAPs itself causes stutter and lower frames per second vs. running the game without it on. It happens.

Let me know though how GTX660Ti went from performing like a dog at 2560x1440 in frames per second to beating HD7950 950mhz. I guess NV added a magical 500mhz Boost function to the GPU core in the latest drivers?

You realize 2560x1440 or 2560x1600 is HD7900's forte compared to GTX600? Where did all those frame rates disappear to in TR's review for Skyrim for example?

Even GTX670 cannot beat HD7950 950mhz at 2560x1440 in GPU demanding games.

You know TR's review also shows GTX660Ti outperforming HD7950 950mhz in Medal of Honor at 2560x1440 with 4xMSAA? How did that happen, by magic?

http://gamegpu.ru/images/stories/Test_GPU/Action/Medal of Honor Warfighter/test/moh 2560 4x.png
http://www.guru3d.com/articles_page...ighter_graphics_vga_performance_review,6.html

It's like reading 20 reviews of Core i7 3770K beating FX8150 in CPU intensive games and 1 review that shows FX8150 dominating i7 3770K completely. That doesn't strike you as odd? Look how many anomalies in that review I pointed out already and the first thing you and tviceman talked about are "sacrificing game smoothness for the sake of frames per second" or basically cheating. But it didn't at all occur to you to check the frames per second first to make sure things actually make sense to begin with?

 

[Lepton87]

RussianSensation good finding, if that tech"review" really states that 660Ti is faster than 7950boost at 2560x1440 resolution across wide variety of games then that "review" cannot be trusted at all. It just doesn't align with every other site out there.

 

[Vesku]

In the comments "Damage" showed that if they drop the AC3 results the 7950 Average FPS is ahead about how you'd expect from the other reviews.

http://techreport.com/discussion/23981/radeon-hd-7950-vs-geforce-gtx-660-ti-revisited?post=692088

Keep in mind they are using a geometric mean for FPS.