GTX 690 or GTX Titan?

[BallaTheFeared]

Sky is falling...

 

[blackened23]

It's not his fault people don't understand the difference between frametime and stutter...

CF and SLI have lower FRAMETIME, wow that's NEWS!

Although, it seems you can't grasp the concept either.

Let me help you out, frametime means basically nothing, it's just FPS in MS.

:whiste:

Oh thank you for stating something completely obvious. So you're agreeing that SLI delivers more frames at 20ms or less than a single card delivers frames at 20ms or less. 50ms or less isn't perceptible to the eye anyway, yeah?

Therefore, since single cards deliver fewer frames at 20ms or less, it has more microstutter. Thanks for agreeing. SLI delivers more frames at lower latency. That sounds like the textbook definition of less microsttuter to me.

 

[BallaTheFeared]

:whiste:

Oh thank you for stating something completely obvious. So you're agreeing that SLI delivers more frames at 20ms or less than a single card delivers frames at 20ms or less.

Therefore, since single cards deliver fewer frames at 20ms or less, it has more microstutter. Thanks for agreeing. SLI delivers more frames at lower latency. That sounds like the textbook definition of less microsttuter to me.

Yeah I edited, you so fast

Let me get out of game, see if I can have a go at it.

 

[blackened23]

those frame time charts aren't about showing microstutter. You would have to have a frame by frame comparison, not just the average. A graph showing averages might as well just give you fps because they are nearly one in the same.

Frametime charts are exactly about microstutter. 50-60 ms latency and below isn't perceptible to the human eye, therefore a higher number of consecutive frames you can get within that latency threshold will be stutter free.

Therefore, thanks to this great new performance metric we've had shoved down our throats for months now -- we learn that SLI is indeed better than single card, and that Titan serves no purpose over a 690 on the market except to those who absolutely need a small form factor single card or CUDA development.

 

[chimaxi83]

Looks like the green team is at a loss lol. I used 680 SLI for about 4 months before switching to Crossfire, and it was great. So, I must be blind, or my eyesight must be horrid, or I must be lying, or I just "don't know what it is".... All excuses the nvidia team uses ever since the frame time "expose" which, ironically, was supposed to reflect badly on AMD. Whoops.

 

[BallaTheFeared]

Looks like the green team is at a loss lol. I used 680 SLI for about 4 months before switching to Crossfire, and it was great. So, I must be blind, or my eyesight must be horrid, or I must be lying, or I just "don't know what it is".... All excuses the nvidia team uses ever since the frame time "expose" which, ironically, was supposed to reflect badly on AMD. Whoops.

I'm actually kind of dumbfounded by Tom's graphs...

They're basically running lowest/min/avg of frame times. They aren't actually showing anything that would relate to MS, such as inconsistent frame times. Hence their graphs often don't line up with their experience, like the 690 stuttering in FC3 with 20 ms 95%.

I don't get the point of their graphs from an actual stutter perspective.

 

[lavaheadache]

Frametime charts are exactly about microstutter. 50-60 ms latency and below isn't perceptible to the human eye, therefore a higher number of consecutive frames you can get within that latency threshold will be stutter free.

Therefore, thanks to this great new performance metric we've had shoved down our throats for months now -- we learn that SLI is indeed better than single card, and that Titan serves no purpose over a 690 on the market except to those who absolutely need a small form factor single card or CUDA development.

Dude, that's what I said. I said you need a frame by frame graph, and you said no, you need consecutive frames. Stop trying to fight with everyone and actually take the time to read some peoples responses before you play teacher.

I will repeat, those graphs DO NOT SHOW MICROSTUTTER. You need a frame by frame analysis.

 

[lavaheadache]

What are you talking about regarding less than 50ms being undetectable? That is crazy talk

 

[f1sherman]

Looks like the green team is at a loss lol. I used 680 SLI for about 4 months before switching to Crossfire, and it was great. So, I must be blind, or my eyesight must be horrid, or I must be lying, or I just "don't know what it is"....

I don't know... are you anything like blackend

Because he keeps on linking charts that prove exactly opposite of what he's trying to say. Look; He's out of control:

Quoting this :whiste: Now, lets look at how the GTX 690 *and* GTX 680 clobber the Titan in both framerates *and* both have less microstutter than the Titan:

 

[Grooveriding]

Stay on your 690, no reason to buy a $1000 card 20% slower than the $1000 card you already have.

 

[BrightCandle]

All this conversation continues to show me is there is a wide group of people who don't know the difference between stutters and microstutter and how you detect them in the graphs. The summary data points unfortunately mix multiple effects and if a graph shows more than one thing the summary graph gives misleading results and the only thing you can actually rely on is the raw frame time data.

So in the Toms hardware the issue is that the game is CPU limited for stages. What happens is that the CPU+GPU provides higher FPS (lower frame time) and then it jumps down to the same point as the lesser system. That causes a higher jump in the difference which in turn makes the frame time less consistent in this measure. That doesn't mean its microstuttering or inconsistent overall, its a one off where the frame rate is determined by a component that is a bottleneck and every system is limited to the same point but the high performance systems show more variation. Its fundamentally flawed as a summary of stutter and microstutter for this reason, it would only work to compare such things if that was all the graph showed.The data misleads you into thinking its less consistent, but in actuality it just had further to fall because it was more performant to start with. Toms hardware didn't show enough data to make a conclusion beyond their summary charts being obviously flawed. Which is the nature of the problem at Toms, they don't understand statistically what it is they measured and its flaw in this case.

What I look at and continue to look at is the frame time charts, everything else is basically useless. They are the only charts of any value to finding microstutter and determining the difference between stutter and microstutter. IE this one:

This chart is actual frame times as measured at the output with useless frames removed, that is frames that didn't show enough pixels to be worth even having. That shows severe microstutter in crossfire, it shows small amounts of microstutter in all the other configurations, more so in the 680 SLI system to the 680 on its own. I suspect if I had the actual frame time data I would classify the 7970 on its own about the same as the 680 SLI system in terms of microstutter, and the 680 is quite a bit better. But I don't have the raw data so I can only go from the graph they give whereas I would like one of my interframe difference charts to compare.

Stop looking at the summary charts, they don't tell you what you think they tell you, and more to the point your interpreting them wrong. One key thing to look at in the 50,75,90,99 charts is the spread between the points not the absolute values. The absolute values tell you nothing about consistency, its how much the points on the graphs change and as a percentage how far apart they are. But the issue with microstutter is because you have 2 levels the summary chart ends up averaging them, all it really shows is the stutter effect and somewhat a bit about the microstutter but both mixed together. These are different problems and they need to be isolated from each other, all I have ever noticed about these different summary charts that toms etc have produced is how flawed they are. At least techreport is being honest about the fact their summary for 99% is showing stutters but doesn't do a thing for microstutter, for that you need to zoom into the graph and look at the inter frame time.

Stop cherry picking the graphs and trying to make them say things they don't, you need only look at the frame time graph and the amount of variation between frame times reported to see how much a game microstutters, and you can also with your brain visually remove the stuttering or compare it separately.

I just think most of the arguments I hear about this are from a position of dishonesty. This data is a graphics card enthusiast dream, finally we can see what is going on at the actual monitor and have real data, far superior to just a high speed camera. Its good science and clearly an advancement over frame times as measured by fraps (which was clearly better than FPS measured by fraps which was clearly better than FPS as measured over a whole benchmark).

Only problem is we can't measure it at home.

 

[blackened23]

What are you talking about regarding less than 50ms being undetectable? That is crazy talk

That's actually not correct - 50ms and below is not perceptible. Feel free to throw SKYMTL at hardwarecanucks an email and ask him that same question about 50ms in terms of frametimes.

 

[blackened23]

Oh, BC. Cherry picking. LOL OK if you say so. You said Tomshardware was cherry picking when nearly every graph they have shows less MS on the 690 vs Titan. Then you suggested PCPer, where nearly every graph shows more MS on the Titan than the 690.. I would argue that you're cherry picking because you said toms is garbage and use PCper. SO I USED PCPER PER YOUR SUGGESTION.

I'm cherry picking based on your ever-shifting arguments I guess.

Then PCPer has every graph showing 690/680sli with less MS vs Titan. LOOK AT THE GRAPH YOU LINKED. All of the 680 sli frames are around 10ms. All of the titan frames are around 20-25MS. THEREFORE, Titan has more microstutter than the 690 and 680 sli. Furthermore, let us take a look of half of the graph up to where frame 3035 is. What's this? Titan has 2 spikes to 35ms while 680 sli has 1 spike to 35ms. Wow.


Thanks for showing us again, that the Titan has more microstutter than 680 SLI. Excellent detective work. Also, perhaps you should actually look at the graphs you're linking because you just defeated your own argument. The 680 sli has less MS than the Titan in your very own graph :awe:

I'll let you get back to work on an explanation, meanwhile maybe the PR crew can work on the development of a new performance metric that doesn't cast single cards in such a bad light. Wouldn't want to hurt sales of Titan.

 

[BallaTheFeared]

Oh, BC. Cherry picking. LOL OK if you say so.

There is only one site so far using both a specialized software program, and direct to screen output capturing...

"Cherry picking"

 

[Ferzerp]

Blackened, either you're trolling or you just have no clue.

Either way, you should stop. 50ms frametimes correspond to 20 fps. If you're happy with 20fps, go buy something lowend.

I'm not even going to address any of the other "points" you've "made".

 

[SirPauly]

If anything, it shows that nVidia is really trying to improve multi-GPU smoothness and I like these examples:

Titan, Titan Sli, GTX 680, GTX 680 Sli, HD 7970, Hd 7970 X-fire:

http://www.sweclockers.com/recension/16541-nvidia-geforce-gtx-titan/8#pagehead

I have to give credit to blackened23 even though he was being sarcastic.

 

[SirPauly]

50ms and below is not perceptible.

Are you sure?

 

[BallaTheFeared]

Are you sure?

 

[rgallant]

If anything, it shows that nVidia is really trying to improve multi-GPU smoothness and I like these examples:

Titan, Titan Sli, GTX 680, GTX 680 Sli, HD 7970, Hd 7970 X-fire:

http://www.sweclockers.com/recension/16541-nvidia-geforce-gtx-titan/8#pagehead

I have to give credit to blackened23 even though he was being sarcastic.

only asking
-hows that work if the frames are capped as in msi-ab @ 60-70 fps

 

[SirPauly]

I can't express how wonderful it is to see all these discussions, reviews, articles that go beyond frame-rate! All kinds of awareness and sites investigating smoothness!

 

[lavaheadache]

Blackened, either you're trolling or you just have no clue.

Either way, you should stop. 50ms frametimes correspond to 20 fps. If you're happy with 20fps, go buy something lowend.

I'm not even going to address any of the other "points" you've "made".

I'm going to second this. Not that it makes him a bad guy or anything but telling me that a 50 ms swing is undetectable is insane. Like you said, 50 ms = 20 fps intervals if it is at an even pace. This is why we need a frame by frame chart to tell us if we are getting even frame delivery or handful different frame rates.

I will agree that minor swings are nothing to be concerned with and to be expected but 50ms? Nope, not good.

 

[SirPauly]

only asking
-hows that work if the frames are capped as in msi-ab @ 60-70 fps

Frame limiters probably help but shouldn't gamers deserve smooth frame-rate without them?

 

[blackened23]

Yeah, I went overboard with the sarcasm. Still, I have found consistent results across a wide array of frametime results indicating that overall, SLI has lower consecutive frametimes than single cards do. Let's be clear that this wasn't an isolated chart, it was across a wide array of benchmarks at multiple websites showing SLI has lower consecutive frametimes than that of single cards.

For instance, a graph from PCper above shows the Titan hanging around 25ms. Meanwhile, 680 sli consistently stays in the 10ms range over 6000 frames indicating that it is smoother than the Titan.

I'll just leave it alone now and let you guys make of it what you will.

 

[Ferzerp]

Yeah, I went overboard with the sarcasm. Still, I have found consistent results across a wide array of frametime results indicating that overall, SLI has lower consecutive frametimes than single cards do. Let's be clear that this wasn't an isolated chart, it was across a wide array of benchmarks at multiple websites showing SLI has lower consecutive frametimes than that of single cards.

For instance, a graph from PCper above shows the Titan hanging around 25ms. Meanwhile, 680 sli consistently stays in the 10ms range over 6000 frames indicating that it is smoother than the Titan.

I'll just leave it alone now and let you guys make of it what you will.

Of course on average frametime is lower. That's just the reciprocal of FPS. You have to get far more granular to see "microstutter".

 

[blackened23]

Of course on average frametime is lower. That's just the reciprocal of FPS. You have to get far more granular to see "microstutter".

We need to define microstutter then in the context of frametimes. Let's say SLI 680 looks like this across 6000 frames:

10ms 7ms 13ms 15ms 15ms 10ms 7ms 15ms 10ms (repeating)

While Titan is:

25ms 25ms 30ms 20ms 22ms 22ms 22ms (repeating)

That's what most of the consecutive frametime charts are showing. A couple of you stated that consecutive frametimes as charted over a graph can define microstutter - call me crazy but that would indicate that SLI has less microstutter. How do you define "microstutter" in the context of these frametime latency charts?

Also, you say that 50ms = 20fps. I would really like to know how you arrived at that number, and what you extrapolate 25-30ms to be because that's where Titan happens to land. SKYMTL of hardwarecanucks mentioned that less than 50ms is imperceptible to the human eye and I would tend to agree but regardless, I would like to know how one equates ms frametime to a framerate, i'm just curious - not being facetious.