AMD Framepacing Driver 13.8 review

Page 4 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

SPBHM

Diamond Member
Sep 12, 2012
5,066
418
126
FC3 is just a stutter mess, even on single cards so its inclusion in these reviews is pointless.

Anyway its good to see PoV of users of both brands. For me, single card wise, there's no difference. I like the SMAA inject in Radeon Pro, it removes aliasing with minimal blurry compared to FXAA.. but I also like the adaptive vsync global or profiles via NV control panel. Gameplay wise its not any different.


at least now it's as bad as the rest, before it was much worse

FarCry3_2560x1440_PLOT.png
 

BrightCandle

Diamond Member
Mar 15, 2007
4,762
0
76
Kyle over at HardOCP took some FRAPs frame time captures (presumably he doesn't have FCAT) and the results are actually quite interesting. 3 of the 4 games showed a marked improvement in the frame times, with tomb raider being the biggest gainer.
1375377273oIx1f99egS_4_9.gif

1375377273oIx1f99egS_4_10.gif


Many people will ask why this matters when FCAT captures what is actually output. The reason is that FCAT measures frames, not the contents of the frames. FRAPs frame times tell us how smooth the input to the contents of the frames are. This is another source of stuttery gameplay that actually AMD and Nvidia are ignoring right now. In this case introducing frame pacing has also improved the "backpressure" to the CPU and the game engine via DirectX. Thus we have better more consistent frame contents production as well as better output to the monitor. This is good all around, AMDs fear that it might produce odd backpressure and worse frame rates are both evidently unfounded.

They have solved much of the problem although not all of it. I want the exact scan line data as there is clearly a pattern to the fps data with discrete steps and I am wondering if they are only adjusting periodically verses every frame. The single frame runts are also a good indication that AMDs frame pacing is reactive not predictive. That is it waits for the cards to get out of sync and then corrects it. But I am surprised they still have runts as I would expect them to correct the moment they move away from each other, its very likely it takes several frames to get so far out of sync to produce a runt/dropped frame. So speculatively it looks like they aren't correctly continuously (like Nvidia) and they are doing so reactively (rather predictively as Nvidia has claimed in the past). They also seem to be correcting only at certain thresholds which may explain the thicker bars of variance.
 

Leadbox

Senior member
Oct 25, 2010
744
63
91
Kyle over at HardOCP took some FRAPs frame time captures (presumably he doesn't have FCAT) and the results are actually quite interesting. 3 of the 4 games showed a marked improvement in the frame times, with tomb raider being the biggest gainer.
1375377273oIx1f99egS_4_9.xgif

1375377273oIx1f99egS_4_10.xgif


Many people will ask why this matters when FCAT captures what is actually output. The reason is that FCAT measures frames, not the contents of the frames. FRAPs frame times tell us how smooth the input to the contents of the frames are. This is another source of stuttery gameplay that actually AMD and Nvidia are ignoring right now. In this case introducing frame pacing has also improved the "backpressure" to the CPU and the game engine via DirectX. Thus we have better more consistent frame contents production as well as better output to the monitor. This is good all around, AMDs fear that it might produce odd backpressure and worse frame rates are both evidently unfounded.

They have solved much of the problem although not all of it. I want the exact scan line data as there is clearly a pattern to the fps data with discrete steps and I am wondering if they are only adjusting periodically verses every frame. The single frame runts are also a good indication that AMDs frame pacing is reactive not predictive. That is it waits for the cards to get out of sync and then corrects it. But I am surprised they still have runts as I would expect them to correct the moment they move away from each other, its very likely it takes several frames to get so far out of sync to produce a runt/dropped frame. So speculatively it looks like they aren't correctly continuously (like Nvidia) and they are doing so reactively (rather predictively as Nvidia has claimed in the past). They also seem to be correcting only at certain thresholds which may explain the thicker bars of variance.

Contents of frames? What is that and how can fraps possibly measure that?
 

Leadbox

Senior member
Oct 25, 2010
744
63
91
^ but doesn't the AT review say that there are now '0' runts with frame pacing enabled?

Guru 3d had runts, but the hardwarecanuks guy reckons their setup has not been updated so it counts everything at below 10ms as runts. A bug corrected in FCAT 1.8
 

BrightCandle

Diamond Member
Mar 15, 2007
4,762
0
76
^ but doesn't the AT review say that there are now '0' runts with frame pacing enabled?

Guru 3D actually has detailed dropped pictures whereas pcper and anandtech don't at this point. They are seeing just occasional single frame drops:

index.php


Correction - I suspect this is the 6 pixel high previous frame problem at the top of the screen not the same runt frame. FCAT sees that as a runt, its clearly an artefact but its not the same sort of problem as crossfire previously had.
 
Last edited:

SickBeast

Lifer
Jul 21, 2000
14,377
19
81
I commend AMD with what they have done with this new driver. Hopefully over time they will get even better.
 

railven

Diamond Member
Mar 25, 2010
6,604
561
126
Welps, guess I'll try the "Hack Method" again which listed the "OpenGL" driver as the most recent 8.1 driver (go figure, haha.)

Otherwise, a format and back to Windows 8. Not really seeing anything special in 8.1, however, I do like that I can right click the Windows logo and get access to almost all the system options.

First world problems.
 

n0x1ous

Platinum Member
Sep 9, 2010
2,574
252
126
Welps, guess I'll try the "Hack Method" again which listed the "OpenGL" driver as the most recent 8.1 driver (go figure, haha.)

Otherwise, a format and back to Windows 8. Not really seeing anything special in 8.1, however, I do like that I can right click the Windows logo and get access to almost all the system options.

First world problems.

Start8 for $5 = win
 

blackened23

Diamond Member
Jul 26, 2011
8,548
2
0
Having read some of the reviews, it looks like a good foundation for AMD to start with going forward. I can't overlook the late nature of this driver, but this at the very least won't be a pronounced issue with the upcoming R-9xxx series (which is strongly rumored to take the single gpu crown again). But, they need to fix eyefinity and D3D9. Hopefully that will happen in short order.

I can't help but think that the software development is a money issue. Well, that IS a money issue - nvidia has more money, personnel and resources to throw at this while AMD does not. AMD has to make do, presumably, since they don't have tons of excess cash to hire software engineers (and driver development is an incredibly tough thing to do from what I understand.) Since the consoles will be fairly lucrative for AMD (75-90$ per console sold IIRC) i'm hoping that situation changes. I'm using a 780 right now that I love but if AMD offers a compelling r9xxx option down the road, i'd consider it.

Also, railven, one does not simply turn off crossfire. You will use crossfire and you will like it :D
 

BallaTheFeared

Diamond Member
Nov 15, 2010
8,115
0
71
I commend AMD with what they have done with this new driver. Hopefully over time they will get even better.

As Apoppin said, you're really only get so far with software, 9xxx series or the one after that will most likely offer hardware frame metering to complement the software, AMD is four(?) years late to this party.

I'm just glad they finally joined it.
 

jackstar7

Lifer
Jun 26, 2009
11,679
1,944
126
Of course it wouldn't be that easy...

Installed drivers and am now having my second card idle at 500/150.

Yes, there is only one monitor connected.

I don't even know where to start to figure this one out. My second non-used card is running hotter than my primary card.

Good job AMD! :thumbsup:

Not sure if related, but I wound up finding out that Firefox was running with GPU acceleration on and maxing out all of my cards whenever I used it. Once I unchecked that box in the preferences, things went back to normal.
 

VulgarDisplay

Diamond Member
Apr 3, 2009
6,188
2
76
Its not that odd considering that this amd driver thread is now about nvidia hardware frame metering.

Here's an idea anyone have any comments on the improvements these drivers bring? I would rather hear about that.
 
Last edited by a moderator:

Kenmitch

Diamond Member
Oct 10, 1999
8,505
2,250
136
Its not that odd considering that this amd driver thread is now about nvidia hardware frame metering.

Here's an idea anyone have any comments on the improvements these drivers bring? I would rather hear about that.

I only have a single card but looking at the results in the reviews it does look like they've made a huge improvement. Guessing this will improve with future drivers.

This sub-forum is a strange one at times.
 

DiogoDX

Senior member
Oct 11, 2012
757
336
136
Tested this in my single 7970. Don't have Metro LL so no changes as expected.
 

Stuka87

Diamond Member
Dec 10, 2010
6,240
2,559
136
As Apoppin said, you're really only get so far with software, 9xxx series or the one after that will most likely offer hardware frame metering to complement the software, AMD is four(?) years late to this party.

I'm just glad they finally joined it.

Looks like some post were lost after the database restore.

Anyway, to answer the post you made there and refer to here. Fermi did NOT have hardware frame metering. Kepler was the first to do it in hardware.
 

BigChickenJim

Senior member
Jul 1, 2013
239
0
0
Out of curiosity, has anyone else noticed drastically different GPU usage percentages since applying the drivers? My experience does seem smoother, but my slave card now runs at nearly 100% all the time while the master sits at 30-50%. Before the drivers those percentages were almost always roughly equal.

Temperatures haven't appeared to change, though, which leads to believe that the issue may be with my monitoring software (MSI Afterburner) after the driver update. Thoughts?
 

Red Hawk

Diamond Member
Jan 1, 2011
3,266
169
106
Looks like some post were lost after the database restore.

Anyway, to answer the post you made there and refer to here. Fermi did NOT have hardware frame metering. Kepler was the first to do it in hardware.

No, stalhart deleted them because things were getting off topic. No discussion of whether or not Fermi has hardware metering, etc. here, unless you want this thread unpinned and locked.

On topic, any idea when an "official" Windows 8.1 preview driver incorporating 13.8's improvements will be released, rather than trying to hack a driver titled for OpenGL
 

BallaTheFeared

Diamond Member
Nov 15, 2010
8,115
0
71
Looks like some post were lost after the database restore.

Anyway, to answer the post you made there and refer to here. Fermi did NOT have hardware frame metering. Kepler was the first to do it in hardware.


I went looking for the post to delete it, I will say Fermi didn't have hardware metering. I can't even say for sure if G80 did or not.

I hope we can drop this and continue talking about AMD, their driver, and where they're heading from here.
 

VulgarDisplay

Diamond Member
Apr 3, 2009
6,188
2
76
Honestly I think all this hardware metering talk is actually just marketing nonsense and I fully expect AMD to jump on that bandwagon with the 9xxx cards.

People assume that there is some magical chip on nvidia cards that actually does the metering. Since no one has ever been able to look at the PCB (to my knowledge) and said here is the chip that performs this duty I think the drivers (software) just tell the hardware (gpu core) how to meter the frames. They called it hardware metering just to make it seem better.

As I said I fully expect AMD to come up with this "hardware based" frame metering marketing bullet point with the 9xxx series. It is seriously a sad state of affairs that it took this long. They truly shot themselves in the foot by not releasing these drivers until rumors of their next generation were creating buzz. If they would have released these back in March I would already have another 7970 in my PC, but since they didn't come out until now I will just wait to see what happens to prices and the performance crown after AMD releases their 28nm refresh cards.

I believe many other people would have picked up a second Radeon if they had their crossfire drivers in order rather than ditching them entirely to pick up Titan's and gtx780's for that extra horsepower. I do hope for the sake of enthusiasts they can get their house in order, stop their bungling, and get relevant again. They came a long way with Tahiti, and I full expect them to either match Titan, or get withing 10% of it with their upcoming GPU's.

A good old fashioned halo GPU duel would be a great thing to see at this point, and could generate a lot of positive press for AMD.
 

3DVagabond

Lifer
Aug 10, 2009
11,951
204
106
Honestly I think all this hardware metering talk is actually just marketing nonsense and I fully expect AMD to jump on that bandwagon with the 9xxx cards.

People assume that there is some magical chip on nvidia cards that actually does the metering. Since no one has ever been able to look at the PCB (to my knowledge) and said here is the chip that performs this duty I think the drivers (software) just tell the hardware (gpu core) how to meter the frames. They called it hardware metering just to make it seem better.

As I said I fully expect AMD to come up with this "hardware based" frame metering marketing bullet point with the 9xxx series. It is seriously a sad state of affairs that it took this long. They truly shot themselves in the foot by not releasing these drivers until rumors of their next generation were creating buzz. If they would have released these back in March I would already have another 7970 in my PC, but since they didn't come out until now I will just wait to see what happens to prices and the performance crown after AMD releases their 28nm refresh cards.

I believe many other people would have picked up a second Radeon if they had their crossfire drivers in order rather than ditching them entirely to pick up Titan's and gtx780's for that extra horsepower. I do hope for the sake of enthusiasts they can get their house in order, stop their bungling, and get relevant again. They came a long way with Tahiti, and I full expect them to either match Titan, or get withing 10% of it with their upcoming GPU's.

A good old fashioned halo GPU duel would be a great thing to see at this point, and could generate a lot of positive press for AMD.

There doesn't actually have to be a separate chip it could be in the firmware/bios and use resources in the GPU. Face it, there's always software involved or else the hardware wouldn't know what to do.

As far as AMD's upcoming refresh, if they want to make an impact they need to be faster than Titan. When you come along this much later with your new design it better be able to beat what the competition has had out for the last 6 months. If they don't, then they are going to need to be at a substantially lower price point and I don't think AMD wants to do that (Lower than the 780. No way they can come in at $1K like Titan. AMD's customers won't accept a $1000 GPU.) If they get a clean kill on Titan they can come in at $650, which would be a huge premium for AMD.
 

Red Hawk

Diamond Member
Jan 1, 2011
3,266
169
106

railven

Diamond Member
Mar 25, 2010
6,604
561
126
Start8 for $5 = win

I tried the trial, and honestly don't miss the start-menu. But...

You can do the same thing in vanilla 8.

Orly...? I never tried to even right click it on Vanilla 8 but if it serves the same function, then 8.1 has nothing for me! Thanks for the tip.

Not sure if related, but I wound up finding out that Firefox was running with GPU acceleration on and maxing out all of my cards whenever I used it. Once I unchecked that box in the preferences, things went back to normal.

I was only using Chrome, and just sitting on the desktop with CCC open my second card was 500/150. Either way, I'm running some kind of Frankenstein setup now which is rather amusing haha.

Gonna start the reformat and fall back to Win8.