Wednesday, April 24th 2013

AMD Makes a Breakthrough in Improving Frame Latency

AMD has been criticized by some tech publications over Radeon HD 7000 series posing higher frame-latency over NVIDIA GeForce chips, even in cases where AMD's chips offer higher frame-rates. "Frame latency" or "frame time" was purported as a metric of the same importance as frame-rates, in graphics card reviews. Various essentially identical methods were used to show that AMD Radeon GPUs yield higher frame latency (time taken for frames drawn by the GPU to make it to the display) than NVIDIA GeForce ones, even in cases where AMD's chips offer higher frame-rates. AMD has apparently made a significant breakthrough in improving frame latency.

In January, AMD made its first official response to early tests that showed Radeon GPUs to pose higher frame latency. In its defense, AMD stated that frame-latency issues are not a hardware design flaw, and can be ironed out by optimizing drivers to the redesigned memory controllers on GPUs based on its Graphics CoreNext architecture. Sources told us that AMD is ready with its first prototype drivers that fix frame latency issues. These drivers are pre-alpha, and are made available to select industry partners, with an adequate level of competence and expertise, since a week now. After AMD takes feedback from these partners, the company will begin rolling out the first beta drivers, followed by WHQL-signed ones.
Add your own comment

42 Comments on AMD Makes a Breakthrough in Improving Frame Latency

#2
cadaveca
My name is Dave
by: Ikaruga
Don't know who tested what and how, but it's very unlikely that the single chip Titan has more frame latency than the 690 with its SLI configuration.
It has more latency, because it renders each frame slower. Frame latency doesn't mean cards aren't working, it's all about how they display the work they've done.

Frame latency refers to the period of time between that work being done, and quite obviously lower FPS = "higher" latency.

80 FPS = 12.5 ms latency, avg.

40 FPS = 25.0 ms latency.


It's things like that that really need to be explained, discussed, etc, when it comes to this topic, and that's why I've taken a seat in the peanut gallery on this topic. It is far too complicated to simply say "OMG!!?!!". AMD's made a statement about what they are doing, and I'm happy with that.


That said, you really cannot compare frame latency from one solution to the next, directly, without first filtering the data because of things like this, and that's not happened as of yet.
Posted on Reply
#3
GamerGuy
by: blibba
7970CF isn't universally broken, it's just subjectively and quantitatively much weaker than FPS testing alone suggests. I doubt many of the people running 7970CF who also have 680/670 SLI on another comparable system, and if they do and still can't tell the difference, I'd suggest that they're not really taxing either system.
Well, I have a triple HD7970 CF system, and a dual GTX670's SLi'ed system......games can run pretty smooth. Sometimes, there may be lagging and microstutters, but I see it on both systems actually. May not be in the same game, but these issues exist for both.....in varying degrees.
Posted on Reply
#4
Ikaruga
by: cadaveca
It has more latency, because it renders each frame slower. Frame latency doesn't mean cards aren't working, it's all about how they display the work they've done.

Frame latency refers to the period of time between that work being done, and quite obviously lower FPS = "higher" latency.

80 FPS = 12.5 ms latency, avg.

40 FPS = 25.0 ms latency.


It's things like that that really need to be explained, discussed, etc, when it comes to this topic, and that's why I've taken a seat in the peanut gallery on this topic. It is far too complicated to simply say "OMG!!?!!". AMD's made a statement about what they are doing, and I'm happy with that.


That said, you really cannot compare frame latency from one solution to the next, directly, without first filtering the data because of things like this, and that's not happened as of yet.
I'm quite familiar with the subject, and no, frame time is != frame latency. Don't get me wrong please, I do not doubt that the actual measures are valid and legit on the picture of course, but this is not a "let's make rendering faster and everything will be suddenly fixed" kind of solution here. Don't forget that this story is about microstutters, and smoothness. Something needs to be fixed in the Titan's driver as well if it's slower than the 690.
Posted on Reply
#5
cadaveca
My name is Dave
by: Ikaruga
but this is not a "let's make rendering faster and everything will be suddenly fixed" kind of solution here. Something needs to be fixed in the Titan's driver as well if it's slower than the 690.
Well, you know, that does seem to be how AMD decided to take things. But of course, I agree with your slant here.


I think, personally, when it comes to frame latency, we need to only explore VARIATIONS, as this is what is upsetting to the end user(at least with myself and other users I have had the time to talk to at length).

And I'm not posting just for your benefit, as I said, this is a complicated subject, and how it's presented to the end user is of a critical nature. I don't mean to refute or deny or affirm anything..merely discuss.

I personally seem to be pretty sensitive to this, but I don't have ANY Nvidia cards, so I am not familiar with how their cards work, or how smooth of an image they present. Frankly, I don't really care too much either. I only care about AMD's solution, since that's the cards I bought. I am very very curious about Nvidia's cards though, let me tell you.
Posted on Reply
#6
blibba
by: GamerGuy
Well, I have a triple HD7970 CF system, and a dual GTX670's SLi'ed system......games can run pretty smooth. Sometimes, there may be lagging and microstutters, but I see it on both systems actually. May not be in the same game, but these issues exist for both.....in varying degrees.
Yes, I wouldn't claim that SLI is faultless in this regard or any other.
Posted on Reply
#7
Ikaruga
I'm 100% feel what you mean on this one. We are playing Quake a lot on CRT monitors with my mates (idtech3 locked @ 125fps/Hz and idtech1 locked @ 144fps/Hz), and we actually had to abandon AMD because of the OpenGL microstutter, and these were times long long before this whole story went viral with the current games, so it was not like that we red something and suddenly let our brains make it placebo happen to us. It was common knowledge among us for years that you have to go Nvidia if you want smooth quality gameplay, and I'm not an Nvidia fanboy at all, I respect and do like AMD as well (was importing ATI cards in the late 80's and early 90's because they had the best quality and fastest 2D graphics accelerators by far (with Tseng ofc, but they acquired them later), and I always knew and hoped for that they gonna be big one day.

But as a gamer, I trust Nvidia more for many years now, because playing on Geforce feels much more smoother for me, there is no other way around this, it just does, period (Last time I tested was two weeks ago when I had a 7850 for a few days). I hope these new drivers will bring change tho,:toast:
Posted on Reply
#8
cadaveca
My name is Dave
by: Ikaruga
I'm 100% feel what you mean on this one. We are playing Quake a lot on CRT monitors with my mates (idtech3 locked @ 125fps/Hz and idtech1 locked @ 144fps/Hz), and we actually had to abandon AMD because of the OpenGL microstutter, and these were times long long before this whole story went viral with the current games, so it was not like that we red something and suddenly let our brains make it placebo happen to us. It was common knowledge among us for years that you have to go Nvidia if you want smooth quality gameplay, and I'm not an Nvidia fanboy at all, I respect and do like AMD as well (was importing ATI cards in the late 80's and early 90's because they had the best quality and fastest 2D graphics accelerators by far (with Tseng ofc, but they acquired them later), and I always knew and hoped for that they gonna be big one day.

But as a gamer, I trust Nvidia more for many years now, because playing on Geforce feels much more smoother for me, there is no other way around this, it just does, period (Last time I tested was two weeks ago when I had a 7850 for a few days). I hope these new drivers will bring change tho,:toast:
I'm on Windows8, and I do have the new drivers. So far, I am pleased with progress made, but there's still some things left to be fixed(clearly).
Posted on Reply
#9
adulaamin
by: jigar2speed
Trust me you will never agree ever. Nvidia is the "BHEST" for you. :D
Look at my system specs before you type another word. :)
Posted on Reply
#10
Frick
Fishfaced Nincompoop
I just thought about the promising concept of Hybrid Crossfire. Would these drivers affect that? Because that is almost genuinely broken (depending on the games).
Posted on Reply
#11
blibba
by: Frick
I just thought about the promising concept of Hybrid Crossfire. Would these drivers affect that? Because that is almost genuinely broken (depending on the games).
Do you mean hybrid crossfire as in APU+GPU, or as in 7870+7850? Because aside from drivers, APU+GPU has an issue in that one of the GPUs is horrendously bandwidth starved.
Posted on Reply
#12
Frick
Fishfaced Nincompoop
by: blibba
Do you mean hybrid crossfire as in APU+GPU, or as in 7870+7850? Because aside from drivers, APU+GPU has an issue in that one of the GPUs is horrendously bandwidth starved.
APU+GPU. I wasn't even aware the other one also is called hybrid cf.
Posted on Reply
#13
blibba
by: Frick
APU+GPU. I wasn't even aware the other one also is called hybrid cf.
It probably isn't, I've just heard people refer to it as that.
Posted on Reply
#14
NeoXF
Curios about one thing... anyone ever measured frame latency and such on AMD's APUs? Especially curious about where there's any benefit for the system being so tightly packed... well, then there's the system RAM sharing/latency issue... but then again, as I understand system DDR memory has way better latency than video GDDR...

Oh and while I'm on the subject, some of those tests on Intel HD Graphics as well...
Posted on Reply
#15
james888
by: NeoXF
Curios about one thing... anyone ever measured frame latency and such on AMD's APUs? Especially curious about where there's any benefit for the system being so tightly packed... well, then there's the system RAM sharing/latency issue... but then again, as I understand system DDR memory has way better latency than video GDDR...

Oh and while I'm on the subject, some of those tests on Intel HD Graphics as well...
Make sure to test 1440p too :p
Posted on Reply
#16
NeoXF
by: james888
Make sure to test 1440p too :p
:p Yeah, but that won't be funny for long... It's something relevant to a lot more people than just enthusiasts, unless those measurements are considered some kid of luxury... or some sort of elitist crap. More and more PCs are shipping/being built with stronger and stronger integrated (or not) graphics horsepower and I can't even start to enumerate in how many ways that's a good thing...
Posted on Reply
#17
james888
by: NeoXF
:p Yeah, but that won't be funny for long... It's something relevant to a lot more people than just enthusiasts, unless those measurements are considered some kid of luxury... or some sort of elitist crap. More and more PCs are shipping/being built with stronger and stronger integrated (or not) graphics horsepower and I can't even start to enumerate in how many ways that's a good thing...
I do find it quite exciting what is happening in the integrated graphics, and that includes intels attempts.
Posted on Reply
Add your own comment