Frame Rating: AMD plans driver release to address frame pacing for July 31st

Final8y · 20 Jun 2013 at 22:35

Since that April release AMD has been very quiet about its driver changes and actually has refused to send me updated prototypes over the spring. Either they have it figured out or they are worried they haven't - but it looks like we'll find out at the end of next month and I feel pretty confident that the team will be able to address the issues we brought to light.

For those of you that might have missed the discussion, our series of Frame Rating stories will tell you all about the issues with frame pacing and stutter in regards to AMD's CrossFire multi-GPU technology.

AMD gave the media a prototype driver in April to test with the Radeon HD 7990, a card that depends on CrossFire to work correctly, and the improvements were pretty drastic.

So what can we expect on July 31st? A driver that will give users the option to disable or enable the frame pacing technology they are developing - though I am still of the mindset that disabling is never advantageous. More to come in the next 30 days!

http://www.pcper.com/news/Graphics-...Driver-Release-address-frame-pacing-July-31st

https://twitter.com/AMDRadeon/status/347803712930070529

PorkPieCat · 20 Jun 2013 at 22:52

Erm... I wonder what would be the benefit of disabling something that people have been calling a "fix"? I wonder if it's going to be something like a choice between low latency (input lag) and low frame latency...

pgi947 · 20 Jun 2013 at 23:11

Better late then never, but 31st of July is a funny 'June'

Confused Stu · 20 Jun 2013 at 23:17

31st July? But that's like a billion years away, you can't expect us to wait that long!!!!

I am wondering if they've made any progress on frame pacing for Eyefinity + Crossfire. In the previews in April they stated the fix was only for single screen setups at the moment, so curious if this is still the case or if they've been able to incorporate a multi-screen fix too. Fingers crossed!

pgi947 · 20 Jun 2013 at 23:23

Haha check the twitter comments, appears nvidia are getting a little revenge on Roy

Rroff · 20 Jun 2013 at 23:28

PorkPieCat said:
Erm... I wonder what would be the benefit of disabling something that people have been calling a "fix"? I wonder if it's going to be something like a choice between low latency (input lag) and low frame latency...

In some cases you actually want frames to be rendered as fast as possible - if you know what your doing there are cases where you don't get the side effects (or not enough to bother about) while gaining a decent framerate increase.

Overall its generally better to have controlled frame output tho, the induced latency is pretty low - peaking around 5ms IIRC and mostly in the region of 0-3ms. (EDIT: Thats assuming your rendering at around 60fps).

Thracks · 20 Jun 2013 at 23:42

The tradeoff for modulating frame pacing in software is a small delay in user input. That is always the tradeoff. Some users want the most responsive user input possible (e.g. competitive twitch FPS gamers), while most will prefer the even frame pacing in games where it has been apparent.

Whatever the user's preference, we strongly believe in giving the user a choice, and we're pleased to be delivering the driver in the June/July timeframe we promised.

Locky · 20 Jun 2013 at 23:45

Thracks said:
The tradeoff for modulating frame pacing in software is a small delay in user input. That is always the tradeoff. Some users want the most responsive user input possible (e.g. competitive twitch FPS gamers), while most will prefer the even frame pacing in games where it has been apparent.

Whatever the user's preference, we strongly believe in giving the user a choice, and we're pleased to be delivering the driver in the June/July timeframe we promised.

Why can Nvidia deliver pretty much both at the same time?

Thracks · 21 Jun 2013 at 00:21

NVIDIA's solution also increases input latency in exchange for modulating frame times, but in their case, the user cannot choose whether or not this feature is enabled.

//EDIT: I want to be perfectly clear, here, so let me expand my answer....

It is always the case that modulating frame pacing introduces a small input delay at the user end. You smooth out frame pacing by establishing a consistent cadence between when the frame is rendered, and when the DirectX present() call puts that frame on the screen. There is no mechanism to deliver even frame pacing that avoids the small delay. But when I say small, I mean small: lower than the reaction time of even the fastest record humans... just a few milliseconds. The average human reaction time is 215 milliseconds, and the fastest recorded human is in the range of 10ms. The delay is even smaller than that.

Rroff · 21 Jun 2013 at 00:30

Locky said:
Why can Nvidia deliver pretty much both at the same time?

If the results for the prototype driver are anything to go by the results are pretty much identical for nVidia's software solution and only very slightly behind the nVidia cards with hardware assistance (in terms of latency/performance trade off).

Giving the user the choice is the better implementation imo tho its only in fringe cases where not having metering in play is useful.

PorkPieCat · 21 Jun 2013 at 01:20

Ok, understand now. Good thing there's a choice. Also 200ms reaction time still doesn't completely negate a small 5ms input lag. You're still getting a 5ms disadvantage (2.5%?) over yourself without input lag, and you're also still seeing a less responsive input (albeit after an 80ms delay, or however long it takes the human eye to process something). Whether 5ms lower input lag or smoother frames gives you better aim is another matter though. I get the feeling you'd aim better with smoother frames. If I had CFX I'd probably choose low frame latency both for casual and competitive gaming...

humbug · 21 Jun 2013 at 01:28

Thracks said:
NVIDIA's solution also increases input latency in exchange for modulating frame times, but in their case, the user cannot choose whether or not this feature is enabled.

//EDIT: I want to be perfectly clear, here, so let me expand my answer....

It is always the case that modulating frame pacing introduces a small input delay at the user end. You smooth out frame pacing by establishing a consistent cadence between when the frame is rendered, and when the DirectX present() call puts that frame on the screen. There is no mechanism to deliver even frame pacing that avoids the small delay. But when I say small, I mean small: lower than the reaction time of even the fastest record humans... just a few milliseconds. The average human reaction time is 215 milliseconds, and the fastest recorded human is in the range of 10ms. The delay is even smaller than that.

Keep up the good work

Rroff · 21 Jun 2013 at 01:32

PorkPieCat said:
Ok, understand now. Good thing there's a choice. Also 200ms reaction time still doesn't completely negate a small 5ms input lag. You're still getting a 5ms disadvantage (2.5%?) over yourself without input lag, and you're also still seeing a less responsive input (albeit after an 80ms delay, or however long it takes the human eye to process something). Whether 5ms lower input lag or smoother frames gives you better aim is another matter though. I get the feeling you'd aim better with smoother frames. If I had CFX I'd probably choose low frame latency both for casual and competitive gaming...

Much of the time the effective difference in latency is really really small definitely having smoother rendering is a far bigger benefit even for competitive shooters.

PorkPieCat · 21 Jun 2013 at 01:40

Yeah I thought so... but there's still no harm in offering a choice

. Though speaking of latency and giving a choice, I think of all the awesome RadeonPro tweaks, "flip queue size" is one of the things that ought to be a CCC option. It actually sounds fairly similar to this. If I understand correctly it also lets you choose between frame smoothness and input lag?

humbug · 21 Jun 2013 at 01:57

PorkPieCat said:
Yeah I thought so... but there's still no harm in offering a choice . Though speaking of latency and giving a choice, I think of all the awesome RadeonPro tweaks, "flip queue size" is one of the things that ought to be a CCC option. It actually sounds fairly similar to this. If I understand correctly it also lets you choose between frame smoothness and input lag?

RP is fantastic, if I can get it to work with Steam Games..... lots of really good features and enhancements.

I have the same feeling about it being part of CCC, but to play the devils advocate, it is work in progress, it seems to me, and i'm no expert, it digs so deep into the API to give it all those nice features inevitably it does not work consistently, the program perhaps needs back end profiles for some individual titles for it to work, so its a never ending work in progress.

Its a deep tweakers tool that is an AMD application partly developed by AMD, or at least they help John Mautari develop it. just not officially.

I don't think AMD want official responsibility for something that works in the way that RP does, IMO they are right to take that stance as something like this would probably not exists with an official stamp of approval on it.

PorkPieCat · 21 Jun 2013 at 02:14

I suspect you could be onto something, but at the same time I would personally have nothing but respect for the AMD driver team if they officially acknowledged RadeonPro as the fantastic tool that it is and offered to hire japamd or incorporated RP into CCC. Actually I heard they sent him some monitors to help him out with Eyefinity, which was really good of them.

humbug · 21 Jun 2013 at 02:21

PorkPieCat said:
I suspect you could be onto something, but at the same time I would personally have nothing but respect for the AMD driver team if they officially acknowledged RadeonPro as the fantastic tool that it is and offered to hire japamd or incorporated RP into CCC. Actually I heard they sent him some monitors to help him out with Eyefinity, which was really good of them.

Yeah, They do acknowledged him and his App, just not officially.

What politics or reason behind that, only they know. but support him on it, I think they do, your cited monitors is probably just one of many ways.

Thracks · 21 Jun 2013 at 02:41

Tools like RadeonPro (or RivaTuner, going back further) play an important role by giving users more options than we could safely provide directly in CCC. The truth is that RP offers a dizzying array of options, some that can seriously impair the performance of a game, if not directly cause games to crash. With millions of users worldwide, we cannot implement RP's featureset into CCC, as not every user has the experience necessary to set the right options.

However, we can continue to enable japamd with hardware, games, SDKs, APIs and more--which we do--to make sure that RP is the amazing tool it deserves to be for the role it plays with enthusiasts.

shankly1985 · 21 Jun 2013 at 02:45

RadeonPRO rocks so much you can add to your games.

humbug · 21 Jun 2013 at 02:47

Thracks said:
Tools like RadeonPro (or RivaTuner, going back further) play an important role by giving users more options than we could safely provide directly in CCC. The truth is that RP offers a dizzying array of options, some that can seriously impair the performance of a game, if not directly cause games to crash. With millions of users worldwide, we cannot implement RP's featureset into CCC, as not every user has the experience necessary to set the right options.

However, we can continue to enable japamd with hardware, games, SDKs, APIs and more--which we do--to make sure that RP is the amazing tool it deserves to be for the role it plays with enthusiasts.