Frame Latency with GTX Titan?

[BababooeyHTJ]

Single GTX Titan is only 15% slower than HD 7970 GE 2-way CFX at 1600p (close to your 1440p); but could nearly halve the power draw. Plus the performance-consistency of single-GPU. I'd bite.

http://tpucdn.com/reviews/NVIDIA/GeForce_GTX_Titan_SLI/images/perfrel_2560.gif

It would be nice if TPU had used up to date drivers in this review. What kind of tri-fire user refuses to use beta drivers? The WHQL driver used in this review is pretty much a two and a half month old beta. Many of the newer games in this review see much better scaling with more recent drivers.

I've yet to see a single Titan go over 1200 Mhz, once you reach a certain power consumption the card will throttle itself.

If the gpu, memory, and vrms stay cooler they work more efficiently and will keep tdp down. So it may help. I'm also hearing that they may raise the tdp limit. +6% is pretty stingy.

 

[LAN_deRf_HA]

Wiz doesn't update the review drivers that often because he's done comparison tests before and the difference usually* isn't as big as claimed.... that and it's a lot of work.

 

[BababooeyHTJ]

We're talking about tri-fire here. I really think that it invalidates his results in that review.

 

[cadaveca]

We're talking about tri-fire here. I really think that it invalidates his results in that review.

Clearly you don't own a triplet of cards. In the past year, maybe 3 driver versions work properly with triple cards. Which invalidates your opinion on the subject.


And no, I do not have three cards. I have four.

Even before the 7-series, I was running three 6950's, and it was the same story there, too.

 

[the54thvoid]

Came across this.

https://forums.geforce.com/default/topic/533215/sli-titans-not-very-smooth/?offset=4


Far Cry 3

Crysis 3

The guys result on BF3 dont look bad and the latency for AC3 was minimal (well under 33.3ms)

I'm still keen but all the things I've seen so far across the forums is that it has throttling issues (due to the complex power/heat balance) and also - from the posters themselves - requires better drivers.
I'll bide my time until either Nvidia delivers tweaks that can improve it's performance by 10-20% or I'll just stick with my 7970's. Did a power consumption comparison and I'd save £8/month using a single card (assuming 300W load 4hrs/day 7 days a week).
Frankly my second card is off (zero core) 90% of PC time for light browsing and streaming.

That price tag is getting more and more ludicrous the more imperfect Titan seems to be

 

[BababooeyHTJ]

Clearly you don't own a triplet of cards. In the past year, maybe 3 driver versions work properly with triple cards. Which invalidates your opinion on the subject.


And no, I do not have three cards. I have four.

Even before the 7-series, I was running three 6950's, and it was the same story there, too.

I've been using 7950 tri-fire for months. Most games seem to scale very well for me. The only issue that I've run into was Far Cry 3 not scaling in tri.

Again, 13.1=12.11b11

Obviously newer games scale better with newer drivers. Maybe other tri-fire users only use WHQL drivers from windows update.

I await your next condescending, arrogant response.

I think that AMD moving away from the monthly whql and releasing more betas is a good move. I still think that the review is flawed.

 

[blibba]

Techreport test frame latency in every game, and they have 7970CF and Titan. Can't believe nobody's posted this link yet, maybe I just missed it:

http://techreport.com/review/24381/nvidia-geforce-gtx-titan-reviewed

Important results in spoiler:

Spoiler

TL;DR: Buy a 690?

Nah, if I was in the market for a £900 graphics setup, Titan would be it. Or maybe a 9800GT, a holiday and a used car

 

[LAN_deRf_HA]

I'd say it's a pretty even toss up. Latency seems to be a non issue on the 690. I'd say it's down to whether you want the max performance or the extra vram.

Any time I've had stutter with my SLI it's been vsync related and the solution is always to switch vsync methods for that game. Not a big deal.

 

[blibba]

Latency seems to be a non issue on the 690.

With one notable exception to my knowledge:

 

[cdawall]

Clearly you don't own a triplet of cards. In the past year, maybe 3 driver versions work properly with triple cards. Which invalidates your opinion on the subject.


And no, I do not have three cards. I have four.

Even before the 7-series, I was running three 6950's, and it was the same story there, too.

I haave run 3 to 4 cards on both brands since the 7 series with nvidia and 3 series with ati excluding the newest gernerations on boths and completely agree with you on this...

 

[radrok]

I haave run 3 to 4 cards on both brands since the 7 series with nvidia and 3 series with ati excluding the newest gernerations on boths and completely agree with you on this...

I'm actually going to give it another quote, I agree with you both, skipped this last generation tho, was kinda sick of the same problems over and over again.

 

[the54thvoid]

lol.

The frame latency stuff is hilarious. The more sites are doing it the more you see discrepancies. Some sites have Titan low, some high. Some sites have awful crossfire, other sites have excellent crossfire.

I think sticking with fps is the best metric and perhaps for reviewers, a (dangerously) subjective analysis of the smoothness.

Also, given the massive variance in frame latency results across review sites it really makes you wonder, is it worth the effort? And before you answer, based on Tech Reports testing from Blibba's post, the single GTX 680 is the worst in 4 out of 7 tests, crossfire 7970 the other 3.

Meh...

 

[blibba]

lol.

The frame latency stuff is hilarious. The more sites are doing it the more you see discrepancies. Some sites have Titan low, some high. Some sites have awful crossfire, other sites have excellent crossfire.

I think sticking with fps is the best metric and perhaps for reviewers, a (dangerously) subjective analysis of the smoothness.

Also, given the massive variance in frame latency results across review sites it really makes you wonder, is it worth the effort? And before you answer, based on Tech Reports testing from Blibba's post, the single GTX 680 is the worst in 4 out of 7 tests, crossfire 7970 the other 3.

Meh...

The reasons for discrepancy are that they're not testing the same thing - there's no standardised comparison yet.

If you look at time over 33ms or time over 40ms (which are the sensible things to look at), then in any one game, with any one set of drivers and at any one graphical settings, there are no discrepancies across sites. But that's not happening much, and because with frame latencies things tend to get out of hand very quickly when they do get out of hand, we see big differences. Particularly between driver versions and test-system setups.

As for going back to FPS, I think that we've fairly conclusively seen now that FPS is nearly useless as a metric. Look at the Dirt Showdown test above - the 690 is getting the highest FPS by some margin. Is it the solution you'd choose based on what you're seeing there? Even minimum-FPS testing doesn't reliably show this stuff up.

Back on topic, very few sites have Titan producing inconsistent latencies outside of maybe a single rogue result where there are driver issues.

 

[the54thvoid]

The reason raw fps is still rational is that latency is not always apparent to our eyes. The initial (well, recent initial) reports using a 7950/660 and skyrim had people pointing out both videos were a bit jittery.

Also, raw fps dictates the brute force of the machine which still counts for a lot as it relates to the underlying sum of it's architecture and processing power. As a point in hand, I could game with super dooper effects all on at 2560x1440 on my set up and although it was jittery, in parts, it was perfectly acceptable to me. I've not played it since the 13.2 drivers that fixed certain games .A lower power card may have been technically smoother but I may have had to dial down the effects to get that smoothness working, otherwise falling to fps lower than required.

So fps does play a role. I think some sites posting minimum fps and maximum fps is sound. But given that I'm meant to see issues on BF3 with my set up (according to one frame latency review) and I see NO shudder and no latency effects, it would be erroneous to include it in a review.

It's not an easy one to work out and I know there ARE problems out there but it does seem to affect most cards across many different games. The main issue being - who coded that game. The GTX 690 latency (for Dirt Showdown) is easy to explain away as when the full effects are switched on the Nvidia cards suffer badly. Parts are coded far better for GCN architecture.

Likewise AC3 was developed leaning towards Nvidia and so doesn't play well with AMD. I just hope for all persons who love tech and gaming that the next gen consoles with AMD inside dont have a detrimental effect on nvidia cards.

What would be nice is if the AMD console generation goes some way to addressing AMD's 'occasional' latency issue and also doesn't impact team green.

I think I'm tired of my own thread now It has the potential to go round in circles

 

[cadaveca]

I've been using 7950 tri-fire for months. Most games seem to scale very well for me. The only issue that I've run into was Far Cry 3 not scaling in tri.

Again, 13.1=12.11b11

Obviously newer games scale better with newer drivers. Maybe other tri-fire users only use WHQL drivers from windows update.

I await your next condescending, arrogant response.

I think that AMD moving away from the monthly whql and releasing more betas is a good move. I still think that the review is flawed.

Got some benches? I don't even get decent scaling with two 7-series cards, like 60%. 6-series AMD Crossfire was the best, efficiency-wise, but 7-series is garbage.

And I'll agree with you that the current way of doing drivers is nicer, and that the revie3w is a bit flawed, however, the data is sound, to me, since my own testing confirms that what they see, I see.


FYI, AMD beta drivers can crash, and then corrupt BIOSes. Look around, see reports of people with ASUS boards and stuck multis...unable to overclock, or, unable to go back to stock, and stuck with an unstable overclock. ASUS isn't the only brand with this problem, and for me, AMD's drivers crashing is the cause. WHQL certification ensures that drivers crash properly, prevent BSODs, and BIOS corruption from happening.

What makes it even more interesting is that my 7950's, clocked up to the same speeds as the 7970's....get exactly the same results, in all benchmarks, and all games, as a pair of 7970. That's right, with current drivers, a purchase of a 7970 was wasted money, since it performs EXACTLY the same as a 7950 at the same clocks. Good job AMD for not using all of the resources available to you.

With those aforementioned beta drivers you like.


And actually, since I include 3DMark in my memory reviews, I've got very recent betas is my reviews, too, 13-2 beta 5, actually. I try every driver, and ever 13.2 beta 5 doesn't work for 3-cards in 3DMark. 3DMark11 works fine, but 1 card gets 11.5k, two gets 16.5k, and 3 gets 17.8k.


Of course, I run these cards with either a 3960x @ 4.6 GHz, a 3770K @ 4.6 GHz, and a 8350 @ 5.0 GHz, so there's no real CPU limitations left.


In each config, the frame latency is horrible with two cards, adding a third adds nearly no performance, but does seem to smooth things out a bit in some apps.


And yes, I can be condescending and arrogant. This is the internet, and you don't hear the tone of my words as the come out of my mouth, you hear the words as your mind reads them.


IN the past week, I have run 1000's of benchmarks, with literally 100's of driver combinations, built from this list:

All these tests...say that triple-card is broken. This also means I have 1000's of benchmarks to show this. See, because this is what I do, find issue, and report them to OEMs. This ASUS BIOS issue... I reported back in September. Shamino just got around to confirming it, 6 months later. Fixing issues, replicating them, etc, takes time, and AMD is hard at work trying to fix things.



To me, this is all very basic. AMD admits that this framerating issue is a real one. End of discussion. Single-card...it's not a problem...and for AMD, Crossfire is horrible. They'll fix it, and soon, I imagine. The whole reason you see so many sites covering this problem now is that they've tested, confirmed problems, and together, are ensuring that people are aware of the issue, so that AMD is forced to deal with it, or take a pie in the face. NO big deal, really.


Nvidia, likewise, has some issues...but they have already taken big steps towards ensuring this is not a problem with their other cards...they'll fix it too, and soon, too.


Oh, one more important thing. I'm using Windows8. If you aren't...well..then our experiences aren't comparable, now are they? I could care less about Windows7...that's 2008 stuff.

 

[the54thvoid]

Single-card...it's not a problem...and for AMD, Crossfire is horrible

I'm on old Wndows 7, maybe that has a bearing but my experience doesn't match yours (crossfire - not tri-fire). The following lists what I've played recently and how I'd subjectively rate the crossfire performance:

Crysis 3 - Excellent, 99% gpu utilisation, very smooth to my eyes.
BF3 - as above
Hitman Absolution - as above.
ACM - No issues, smooth but didn't persevere with game.
Dishonored - good, slight loading latency lag, texture popping
GW2 - definite latency - now fixed from 13.2 (checked it out yesterday)
Far Cry 3 - as above but not checked yet with new drivers (finished game)
MOHW - bollocks - crash to desktop - only worked single card

Everybody knows crossfire isn't as good as sli. But it's not as bad as you're saying (Horrible). But maybe horrible is completely subjective. If you love Far Cry 3 (an AMD game) then you might find it unforgivable that crossfire was stuttery.

I'd totally agree that 3 cards is not really that great a jump as 2 is from 1. But it's not rational to decry crossfire as 'horrible', especially you say the 7-series. If anything from my experience (had 2 x 5850's in the past) crossfire is getting much better.

But i suppose that's why my last post said this is all very circular. Maybe a mod should shut this thread before it gets dirty?

 

[blibba]

The reason raw fps is still rational is that latency is not always apparent to our eyes.

That's just not the case. When a game feels sluggish, that's because frame latencies are consistently high. When it feels jittery, that's because occasional frame latencies are high. Latencies are all that we can detect, and FPS is just a way of measuring average frame latency. But averages aren't so interesting when things aren't smooth, which they aren't.

 

[the54thvoid]

That's just not the case. When a game feels sluggish, that's because frame latencies are consistently high. When it feels jittery, that's because occasional frame latencies are high. Latencies are all that we can detect, and FPS is just a way of measuring average frame latency. But averages aren't so interesting when things aren't smooth, which they aren't.

Minimal latency is meaningless if a game cannot render above 30 fps or as some prefer the refresh rate of their monitor. In that case, FPS absolutely does matter, especially where latency is not an issue. That's why i say latency can be overlooked for FPS.

Now, taken to the extreme, if the (somewhat sporadic) latency issue is fixed in the future, then fps becomes important again, as I say, if the delivery to our eyes is not above a certain criteria - we will notice that.

Let's agree that fps is not relevant if latency is 'consistently' smooth and the delivery of frames exceeds human visual processing parameters.
Conversely, latency is not relevant if a card cannot deliver the required framerate to exceed the human perception of 'movement' from the delivery of 'x' number of still frames.

Summary - latency matters where frame rate (fps) is high. Latency is irrelevant if the frame rate is low*

*Considering using maximal FX and not compromising IQ.

Can we drink beer now?

 

[blibba]

Summary - latency matters where frame rate (fps) is high. Latency is irrelevant if the frame rate is low

You seem to have things very muddled in general (no offence), but I agree with this bit I've quoted. But another way of saying that the frame rate is low is that the frame times are consistently high. So my point is that FPS tells us nothing that latency analysis does not.

But yeah, let's just drink beer. I'd much prefer that.

 

[the54thvoid]

You seem to have things very muddled in general (no offence), but I agree with this bit I've quoted. But another way of saying that the frame rate is low is that the frame times are consistently high. So my point is that FPS tells us nothing that latency analysis does not.

But yeah, let's just drink beer. I'd much prefer that.

I'm not muddled at all. I know where i'm coming from. If a card renders 10fps, it is shit. If a card renders 100fps but has consistent high latency - it is also shit.

If a card renders 100 fps with no latency it is excellent.

Run Crysis 3 on 30" screen with a 5 years old gfx card and it wont have enough power to render enough frames to make it smooth. If this means it has high latency then i get you.

I am already drinking beer.

 

[blibba]

If a card renders 100fps but has consistent high latency - it is also shit.

This is impossible. If it's rendering at 100FPS, its latency can be at most occasionally high.

If a card renders 100 fps with no latency it is excellent.

This is also impossible. No latency = infinite FPS.

I am already drinking beer.

I'm drinking tea. I'll move onto beer at the pub in about 50 minutes.

 

[TheMailMan78]

Single GTX Titan is only 15% slower than HD 7970 GE 2-way CFX at 1600p (close to your 1440p); but could nearly halve the power draw. Plus the performance-consistency of single-GPU. I'd bite.

http://tpucdn.com/reviews/NVIDIA/GeForce_GTX_Titan_SLI/images/perfrel_2560.gif

I agree man. Single GPU all the way!

 

[cadaveca]

I don't think this is any reason to NOT buy any cards. In the case of AMD Crossfire, AMD has said it's a problem, so they'll fix it. Nvidia already has routines in place to fix such issues, but AMD does not, and that's all that is really new about the information presented really.


Kinda makes you wonder if this wasn't brought about by nVidia as a selling point for their cards.


Either way, I still play games daily, and still enjoy them. I just happen to enjoy them most with a single card, and frame latency doesn't ever really seem to be an issue with single cards. A new card.. well, of course it's gonna have issues at launch. If Titan didn't, I'd really be shocked. Any card needs at least 6 months in the wild before drivers start to really work well overall.

 

[blibba]

Kinda makes you wonder if this wasn't brought about by nVidia as a selling point for their cards.

Yeah, but if so, I'm pleased. I think this stuff is a more valid concern than the tiny variance in FPS we normally get so hung up on.

 

[cadaveca]

Yeah, but if so, I'm pleased. I think this stuff is a more valid concern than the tiny variance in FPS we normally get so hung up on.

I agree. I've been talking about issues like this for years. Literally. Now people are taking it seriously.

The big issue is that there is not a lot of public documentation from AMD or Nvidia about what's really going on. I was complaining about 60+ FPS having stutter issues since 2900XT. I have a video of it up on my youtube channel, even. The number of people that said they saw no issue is really amazing to me.

Here we are now, many years later, and it's still problem, but more people see it. There's still tonnes of people who do not. To me, it's no different than some people getting motion sickness. And for me, that's what this problem causes, literal physical sickness. So it's a big issue for me.

Some drivers are worse than others. There's been times when I haven't been able to play BF3 for more than 10 minutes or so...turn off Crossfire, and it doesn't bother me.

They have warnings about games and flashing lights causing problems for people, and I'm one of those people that is subject to a form of that. Understanding that, through the help of a doctor, really makes this a small issue, really, but it's funny to me when people like above don't agree on this issue, because I really don't expect many to understand, at all.