AMD Framepacing Driver 13.8 review

[tential]

since the last post is about the start menu:
I've never one missed the start menu. When I want to load up something I don't click on the start menu. It's inconvenient to me now that I'm windows 8. It's quicker to hit Windows Key, Type XBM and then hit enter to launch XBMC.

If I want to launch utorrent,
Windows Key, utor then enter.

The transition was a little odd, but I have the start menu on my Kitchen PC, and I NEVER use it. It's just quicker to windows key, type 3 letters and hit enter. Or, to swing my mouse to the bottom right to show desktop, then click the shortcut I want.

 

[stahlhart]

The next member who posts in this thread about Nvidia hardware frame metering, Windows 8 start menus, the latest developments in Glide technology, or anything else besides the AMD 13.8 drivers unstickies the thread.

Stay on topic. Last warning.
-- stahlhart

 

[kakashi08]

When I install these new drivers 13.8b should I install the 13.5 CAP 1?
because I uninstalled the 13.6 drivers, from the add/remove program in the CP. then installed the 13.8b, idk if i'm suppose to install the CAP 1 also or anything else. help.

 

[railven]

PC reformatted, back on Win8 and 13.8Betas installed without a hitch.

Now to install a game (or two) and give them a whirl. My biggest interest (although I am on hiatus currently) would have to be WoW.

Any word of this frame pacing works with AFR or 1x1 CFX methods or just whatever "default" represents?

 

[Leadbox]

Any reviews with power consumption #s out there?

With peeps reporting lower temps and reduced gpu usage, I wonder were it stands now

 

[DiogoDX]

When I install these new drivers 13.8b should I install the 13.5 CAP 1?
because I uninstalled the 13.6 drivers, from the add/remove program in the CP. then installed the 13.8b, idk if i'm suppose to install the CAP 1 also or anything else. help.

13.6 and 13.8 don't need the cap.

 

[BallaTheFeared]

Just wanted to pull this back up and say that the release notes for the 13.8 beta specifically say that DirectX 10 games are supported for framepacing.

http://support.amd.com/us/kbarticles/Pages/AMDCatalyst13-8WINBetaDriver.aspx

"Supported for DirectX 10 and DirectX 11 applications, and resolutions up to and including 2560x1600 (single display)."

Yes that was cleared up back on post #33, what remains is DX9 and infinity I believe, correct?

 

[Shamrock]

For my lil 7850 OC, it feels like the minimum fps is a bit faster. I haven't really checked. Might be psychological.

 

[Stuka87]

Yes that was cleared up back on post #33, what remains is DX9 and infinity I believe, correct?

DX9, Eyefinity, and OpenGL do not yet have support.

 

[Leadbox]

After only having skimmed through the Ryan Smith's review of this driver, I finally gave it a full look yesterday.
It turns out that the slower your Xfire solution, the better your frame to frame variance. The 6990 has better delta %s than the 7990 in 5 of the six games they tested, Ryan reckons the slower the solution, the more time the phase 1 driver has to better pace the frames.
Pcper's review backs this up too:
Here's 7970 xfire in battlefield

And here's the 7870 xfire

The pattern is the same for the other 2 games they tested. Intresting, huh?

 

[VulgarDisplay]

I just wish they would try and do some sort of research to put a line on those graphs that shows where the average viewer can detect problems in the motion. Pretty graphs, but without that information they are kinda useless. I get that the more uniform the frametime is the better, but there has to be a point where it can't be seen.

 

[sushiwarrior]

I just wish they would try and do some sort of research to put a line on those graphs that shows where the average viewer can detect problems in the motion. Pretty graphs, but without that information they are kinda useless. I get that the more uniform the frametime is the better, but there has to be a point where it can't be seen.

Changes depending on each person though. Some people don't notice/care at all, regardless of how bad, some people notice the most minute differences (or at least think they do...). It's absolutely depends on the person.

 

[AnMig]

After only having skimmed through the Ryan Smith's review of this driver, I finally gave it a full look yesterday.
It turns out that the slower your Xfire solution, the better your frame to frame variance. The 6990 has better delta %s than the 7990 in 5 of the six games they tested, Ryan reckons the slower the solution, the more time the phase 1 driver has to better pace the frames.
Pcper's review backs this up too:
Here's 7970 xfire in battlefield

And here's the 7870 xfire

The pattern is the same for the other 2 games they tested. Interesting, huh?

Does this mean it would be better (for micro stutter) to not overclock your 7970? I guess as long as your are maintaining 60 FPS (my monitor refresh rate).

I have my 2 7970 @ 1100 and 1550 currently, maybe I should just use default clocks. Should run cooler, quieter and theoretically better for micro-stutter.


peace

 

[BrightCandle]

I just wish they would try and do some sort of research to put a line on those graphs that shows where the average viewer can detect problems in the motion. Pretty graphs, but without that information they are kinda useless. I get that the more uniform the frametime is the better, but there has to be a point where it can't be seen.

I have collected quite a lot of traces from people along with their impression of the stutter involved. I was hoping at the beginning of the year to have a more complete study but other projects came up, neitherless I do have some figures for you.

Around a 12ms swing from top to bottom of the variation frame to frame pretty almost everyone sees it and is affected by it.

Around 8ms a smaller number of people, around 10-20%, see it and find it affects playability, but a much smaller number (<5%) consider it unplayable.

Nvidia claims that no one can notice it below 2ms.

Personally I find around 6ms and less is normally good enough for the motion to fool my eyes and brain. I used to think it was around 8ms but some custom software I have that can test this (not yet in a state to release) tells me its more like 6ms.

So given that I would draw a top to bottom peak of 6ms to the graph, so if its varying around 16ms (60fps) then if the width of the line means it exceeds 19ms and the bottom descends lower than 13ms then a number of people will see it and be impacted by it.

That is about the best numbers I have. Its not wholy accurate yet, they are collected from a variety of games in a variety of ways and before frame pacing. We can't really use FCAT for such a study because home users aren't going to have it so we are stuck with fraps data from single card setups.

What I don't know is if that should be expressed as an absolute number or a percentage. As we move to 8ms and 120hz screens 3ms will be a 40% of the frame time not 20% like with 60 fps and then it might well become detectable again. I just don't know if its relatively or absolute.

 

[3DVagabond]

I just wish they would try and do some sort of research to put a line on those graphs that shows where the average viewer can detect problems in the motion. Pretty graphs, but without that information they are kinda useless. I get that the more uniform the frametime is the better, but there has to be a point where it can't be seen.

I've been asking for visual testing of this since day one. You are correct. Without us defining what is visible it's about as useful as 3DMarks. We can determine which card is faster and smoother at delivering frames, but little else. It appears that it's more useful as a marketing tool than any real useful information for the end users...

 

[3DVagabond]

Changes depending on each person though. Some people don't notice/care at all, regardless of how bad, some people notice the most minute differences (or at least think they do...). It's absolutely depends on the person.

Sorry, but this hasn't been proven. It's just excuses used to not determine the facts. While I'm not saying that their won't be outliers that are super sensitive to it, we need to know what the typical person's thresholds are. As far as others go, what is their tolerance and what % of users are they? I've seen plenty of claims by people who supposedly have super human perception but can't back it up when confronted with actual scientific testing.

 

[BoFox]

I have collected quite a lot of traces from people along with their impression of the stutter involved. I was hoping at the beginning of the year to have a more complete study but other projects came up, neitherless I do have some figures for you.

Around a 12ms swing from top to bottom of the variation frame to frame pretty almost everyone sees it and is affected by it.

Around 8ms a smaller number of people, around 10-20%, see it and find it affects playability, but a much smaller number (<5%) consider it unplayable.

Nvidia claims that no one can notice it below 2ms.

Personally I find around 6ms and less is normally good enough for the motion to fool my eyes and brain. I used to think it was around 8ms but some custom software I have that can test this (not yet in a state to release) tells me its more like 6ms.

So given that I would draw a top to bottom peak of 6ms to the graph, so if its varying around 16ms (60fps) then if the width of the line means it exceeds 19ms and the bottom descends lower than 13ms then a number of people will see it and be impacted by it.

That is about the best numbers I have. Its not wholy accurate yet, they are collected from a variety of games in a variety of ways and before frame pacing. We can't really use FCAT for such a study because home users aren't going to have it so we are stuck with fraps data from single card setups.

What I don't know is if that should be expressed as an absolute number or a percentage. As we move to 8ms and 120hz screens 3ms will be a 40% of the frame time not 20% like with 60 fps and then it might well become detectable again. I just don't know if its relatively or absolute.

I'd just present all the data (both absolute and percentage numbers). Don't worry too much about it being either relative or absolute - both become almost the same thing when it comes to higher refresh and lower frame time variations.. to the point where we can no longer discern the microstuttering. The more data - the merrier, I guess! It's impressive what you are doing!

BTW, if you guys see more reviews on 13.8 framepacing, please don't forget to share!

 

[TerryMathews]

I'd just present all the data (both absolute and percentage numbers). Don't worry too much about it being either relative or absolute - both become almost the same thing when it comes to higher refresh and lower frame time variations.. to the point where we can no longer discern the microstuttering. The more data - the merrier, I guess! It's impressive what you are doing!

BTW, if you guys see more reviews on 13.8 framepacing, please don't forget to share!

I'll chime in with my $0.02. FX-8150 and HD 6950s in Crossfire. 5600 3dMarks in Firestrike running Cat 13.4, 5550 under 13.8 beta. That said, 13.8 feels faster, in that it seems smoother without as much jerkiness.

1920x1080 resolution.

 

[OatisCampbell]

Changes depending on each person though. Some people don't notice/care at all, regardless of how bad, some people notice the most minute differences (or at least think they do...). It's absolutely depends on the person.

Have you used Crossfire?

I had two regular 7970s early this year, and probably still would have them if these drivers had been out. It was VERY choppy.

I had GTX670 SLi last year, and that was better, but I'm glad I'm back on single card with my GHz 7970.

One card is smoother than two, no matter who makes the drivers. AFR just seems to add a bit of lag, even when done "well" like the 670s.

In any case, for the people that need/want Crossfire, this is a very big deal. Anyone would have noticed the problem with what it was.

 

[omeds]

Yes that was cleared up back on post #33, what remains is DX9 and infinity I believe, correct?

DX9, Eyefinity, and OpenGL do not yet have support.

Aswell as 3-way and Quad.

 

[bystander36]

Sorry, but this hasn't been proven. It's just excuses used to not determine the facts. While I'm not saying that their won't be outliers that are super sensitive to it, we need to know what the typical person's thresholds are. As far as others go, what is their tolerance and what % of users are they? I've seen plenty of claims by people who supposedly have super human perception but can't back it up when confronted with actual scientific testing.

It seems that some people don't let it bother them, but when sitting next to two machines with or without frame metering, they do notice what is smoother.

To me, that means they'd still be better off with frame metering than without, regardless if they complain about it or not.

At least this is what I got from an earlier Tom's Hardware article where they had some blind testing done by gamers.

 

[3DVagabond]

It seems that some people don't let it bother them, but when sitting next to two machines with or without frame metering, they do notice what is smoother.

To me, that means they'd still be better off with frame metering than without, regardless if they complain about it or not.

At least this is what I got from an earlier Tom's Hardware article where they had some blind testing done by gamers.

I'm not saying there's no perceivable difference with and without frame metering. What I'm saying is we need to scientifically verify the parameters where it's visible. The intensity and frequency that it becomes statistically verifiable. I'd hypothesize that it would become apparent at lower intensities as the frequency increased. I've seen people absolutely convinced that they can perceive differences and when they were blind tested failed to do so. They weren't purposely being deceitful either. They truly believed they heard (these were listening tests) differences.

With this, people just want to make it absolute that if A is measurably "smoother" than B then A offers superior visual quality without considering other performance parameters.

 

[BrightCandle]

Many years ago they determined that 30 fps was an effective minimum frame rate for games. Nowadays 60 fps has most people happy but we have a lot of people seeing the difference from 60 to 120. Its far to say any such number would be just as flawed as the 30 fps number. That isn't to say I don't value such a thing, but one number is not going to be any more use than the 30fps one was. Simply put less variance is always going to be better, there isn't a cut off where it no longer matters (as far as I can tell), its going to be continuous improvement just as fps goes up.

 

[bystander36]

I'm not saying there's no perceivable difference with and without frame metering. What I'm saying is we need to scientifically verify the parameters where it's visible. The intensity and frequency that it becomes statistically verifiable. I'd hypothesize that it would become apparent at lower intensities as the frequency increased. I've seen people absolutely convinced that they can perceive differences and when they were blind tested failed to do so. They weren't purposely being deceitful either. They truly believed they heard (these were listening tests) differences.

With this, people just want to make it absolute that if A is measurably "smoother" than B then A offers superior visual quality without considering other performance parameters.

While there may be a point where people will not be able to notice much of a difference, and when people blind test the two, you'll get evenly split answers on what is smoothest, that does not mean that we are there yet, at least when comparing frame metering vs. no frame metering. Of course, that is assuming similar FPS as well. Higher FPS does play a part too.

 

[3DVagabond]

While there may be a point where people will not be able to notice much of a difference, and when people blind test the two, you'll get evenly split answers on what is smoothest, that does not mean that we are there yet, at least when comparing frame metering vs. no frame metering. Of course, that is assuming similar FPS as well. Higher FPS does play a part too.

Just as an example, we have websites claiming that spikes less than 40ms to 50ms don't matter and we have people claiming to see differences as low as 3ms. Both can't be correct.

...and yes, fps does matter. :thumbsup: