AFR (Alternate Frame Rendering)

Pace

n00b
Joined
Jun 22, 2004
Messages
3
It's been a while, but I had an issue with the Rage Fury Maxx because it used AFR. Now that I was thinking of going dual GPU again, I noticed that AFR is still used. Can anyone post non-AFR results of the 9800GX2?

Why not AFR? Well, largely due to this - I think AFR is just a booster of framerate scores, not actual performance.
lag.gif

http://www.tomshardware.com/1999/11/08/preview_of_the_double_whopper_/page2.html

(wow, 1999!)
 
They are only the same in basic concept. Theres a large difference between the AFR techniques of 10 years ago and the AFR techniques today. For one there is no Z-buffer drop, and there are probably a lot of other differences as well. Technology has evolved somewhat in the last 10 years.

There is a very real boost in performance, and no strange half-assed hacks like you may have seen in the distant past. SLI-disabled scores for the GX2 will be nearly identical to an 8800 GTS 512MB, so you can use that as a reference.
 
I'm currently running (well was until last night) GTS512 SLI and there is a definate difference between having SLI enabled and disabled.
 
As long as there is a queue of frames, the actual fps scores aren't comparable with single-GPU scores, imho. Can you force AFR or SFR as a rendering mode? Can anyone do this and post a benchmark? :)
 
Pace:

The only problem with AFR is that it causes more latency in the loop when playing fast-paced games. You add an extra frame of latency, because the cards have to have the triangle and scene data one frame in advance of what you see.

This sounds like a small amount, but remember two things: 1. the "frame" of latency is actually at half your total framerate, because the speed at which each independent GPU renders a frame is the delay in question. Two, if your framerate is already low (30fps or less), your delay can escalate into the tens of milliseconds.

If your framerate isn't very high, you will most certainly feel the lag of AFR. Nothing will ever change this, but that's why it is the fall-back rendering mode for both SLI and Crossfire. SLI and Crossfire feature split-frame rendering, if the game supports it, which adds no lag. You may have to enable it explicitly in the control panel.
 
Pace:

The only problem with AFR is that it causes more latency in the loop when playing fast-paced games. You add an extra frame of latency, because the cards have to have the triangle and scene data one frame in advance of what you see.

This sounds like a small amount, but remember two things: 1. the "frame" of latency is actually at half your total framerate, because the speed at which each independent GPU renders a frame is the delay in question. Two, if your framerate is already low (30fps or less), your delay can escalate into the tens of milliseconds.

If your framerate isn't very high, you will most certainly feel the lag of AFR.

True, and then add in any input lag your LCD may have, for example. Then if you have a wireless mouse, etc. Of course it can be argued that losing a third or half of your framerate is more detrimental to gameplay than a frame of input lag.

That said, factoring in all your components can add a lot of lag.
 
Pace:

The only problem with AFR is that it causes more latency in the loop when playing fast-paced games. You add an extra frame of latency, because the cards have to have the triangle and scene data one frame in advance of what you see.

This sounds like a small amount, but remember two things: 1. the "frame" of latency is actually at half your total framerate, because the speed at which each independent GPU renders a frame is the delay in question. Two, if your framerate is already low (30fps or less), your delay can escalate into the tens of milliseconds.
This is my point exactly - so if you get 60fps on one card, and 90fps with AFR, then I'd take the 60fps any day of the week. I'm just curious how multi-GPUs fare in benchmarks when not using AFR, i.e. using split-frame render, SLI anti-aliasing etc. I think AFR is the fastest, so will be shown in most benchmarks, yet I'm not really interested in those ('fake') scores.
 
This is my point exactly - so if you get 60fps on one card, and 90fps with AFR, then I'd take the 60fps any day of the week. I'm just curious how multi-GPUs fare in benchmarks when not using AFR, i.e. using split-frame render, SLI anti-aliasing etc. I think AFR is the fastest, so will be shown in most benchmarks, yet I'm not really interested in those ('fake') scores.

There's nothing "fake" about it. It's a solid, measurable increase in frame rate. That is the goal of SLI.

The worst case scenario is you might have ~10ms (which is utterly negligible and completely undetectable) of latency introduced into the process. You're still getting many more frames per second and the overall performance is going to be superior to a single card in the same situation (driver issues aside).


Depending on the game, either AFR or slit-frame might be faster than the other method; But they will both be faster than a single card.
 
This is my point exactly - so if you get 60fps on one card, and 90fps with AFR, then I'd take the 60fps any day of the week. I'm just curious how multi-GPUs fare in benchmarks when not using AFR, i.e. using split-frame render, SLI anti-aliasing etc. I think AFR is the fastest, so will be shown in most benchmarks, yet I'm not really interested in those ('fake') scores.

You make a good point, but it could easily be argued that 10-20ms of input lag is well worth it in many situations. Obviously in the situation you pose, where it is 60fps vs. 90fps it probably wouldn't be worth it. But if somebody was already getting 60fps, they are probably not going to bother dealing with SLI to begin with.

I would think 20ms of input lag would be worth going from say, 25fps to 40fps average, or upping your minimum frame rate from 15fps to 25fps. At high resolutions and high settings often times a single card won't give people the frame rates they are comfortable with. Gaming at an unplayable or nearly unplayable framerate is far worse than a few milliseconds of input lag to most people.
 
Back
Top