Frame Rating and AMD crossfire issue

To those who don't want to watch the 20 minute video.

Basically this guy (Ryan) has used a method that 'marks' frames per second with coloured bars. There are a few types of FPS but the ones being highlighted here are Runts and Dropped frames.

A runt is a 5 pixel high frame that counts as a FPS in FRAPS but can actually hinder performance. A dropped frame is just that, a frame that does not count as it wasn't fully carried out.

A hardware device is used to record the game with the bars down the side. At that point a software app is used to read the bars. Once that is done it outputs to an XML file where the Runt frames and dropped frames are removed, showing the ACTUAL FPS that your eyes are actually seeing.

IE - Radeons in Crossfire are sending out two types of useless frames that don't do anything. Once removed you get the actual FPS count (or Observed as Ryan calls it).
 
AlienALX said:
To those who don't want to watch the 20 minute video.

Basically this guy (Ryan) has used a method that 'marks' frames per second with coloured bars. There are a few types of FPS but the ones being highlighted here are Runts and Dropped frames.

A runt is a 5 pixel high frame that counts as a FPS in FRAPS but can actually hinder performance. A dropped frame is just that, a frame that does not count as it wasn't fully carried out.

A hardware device is used to record the game with the bars down the side. At that point a software app is used to read the bars. Once that is done it outputs to an XML file where the Runt frames and dropped frames are removed, showing the ACTUAL FPS that your eyes are actually seeing.

IE - Radeons in Crossfire are sending out two types of useless frames that don't do anything. Once removed you get the actual FPS count (or Observed as Ryan calls it).
Click to expand...


That's is one of the easiest explanations of what is happening that, I have seen so far since I 1st heard about this issue.

AMD have said that there cards suffer quite allot from it, but that the latest drivers improve it but not to the standard that they want just yet, but that they are working on it.

Nvidia also suffers the same problem but not as much as AMD and I think Nvidia will be working on something as well to try and solve the issue, but I don't think as of yet Nvidia has actually stated that they suffer the issue being seen.
 
PANIC PANIC PANIC PANIC.

Meh. Put on vsync and stop moaning.

On the less sensationalist side of things:

For those of us who have been using xfire setups for a good while it has been known for a long time that AMD drivers are not as good as Nvidias in multiple card setups. I've always described it by saying that "AMD cards panic in xfire". You can stop the panic using vsync because the monitor forces the card to hold back the frame and do it properly.

This does not mean that xfire is broken. This does not mean that AMD suck the biggest balls known to man.

Firstly for single card uses it makes no difference. AMDs drivers remain good and smooth (certainly from what I've seen and including the evidence in that video). Secondly xfire performance is still a colossal boost, up to 200% as per usual. Thirdly many people have been running xfire for years without really noticing this so why are they panicking now? Anyone else, whose eyes were good enough to notice this, from experience uses vsync...

Interesting methodology although I think it's a little bit overdrawn. The FRAPS option to record frametimes does the same job in my opinion although having a second method which confirms the first is nice. This isn't a breakthrough in analysis - just that the market has suddenly decided that it now cares about frametimes.

We've already mentioned to TTL about adding frametimes to his reviews.(http://forum.overclock3d.net/showthread.php?t=51562)

I'm aware that saying all this with out much to back it up is not ideal. I will install BF3 and report back with frame time results.
 
Last edited:
i should my rig back up and running tonight, i dont mind running a few tests if anyone wants me to xfire'd 7970's
 
Master&Puppet said:
PANIC PANIC PANIC PANIC.

Meh. Put on vsync and stop moaning.

On the less sensationalist side of things:

For those of us who have been using xfire setups for a good while it has been known for a long time that AMD drivers are not as good as Nvidias in multiple card setups. I've always described it by saying that "AMD cards panic in xfire". You can stop the panic using vsync because the monitor forces the card to hold back the frame and do it properly.

This does not mean that xfire is broken. This does not mean that AMD suck the biggest balls known to man.

Firstly for single card uses it makes no difference. AMDs drivers remain good and smooth (certainly from what I've seen and including the evidence in that video). Secondly xfire performance is still a colossal boost, up to 200% as per usual. Thirdly many people have been running xfire for years without really noticing this so why are they panicking now? Anyone else, whose eyes were good enough to notice this, from experience uses vsync...

Interesting methodology although I think it's a little bit overdrawn. The FRAPS option to record frametimes does the same job in my opinion although having a second method which confirms the first is nice. This isn't a breakthrough in analysis - just that the market has suddenly decided that it now cares about frametimes.

We've already mentioned to TTL about adding frametimes to his reviews.(http://forum.overclock3d.net/showthread.php?t=51562)

I'm aware that saying all this with out much to back it up is not ideal. I will install BF3 and report back with frame time results.
Click to expand...

Do you actually understand the article? because from that ^ it doesn't sound like it.

Basically the FPS you are reading with FRAPS are completely false. The ACTUAL FPS you are seeing with your eyes is practically nothing over one card.
 
AlienALX said:
Do you actually understand the article? because from that ^ it doesn't sound like it.

Basically the FPS you are reading with FRAPS are completely false. The ACTUAL FPS you are seeing with your eyes is practically nothing over one card.
Click to expand...
I sure do - I'm talking about frame times not frame rates. The dude in the video shows the output of a frametime graph at 15:00. That's what I'm talking about.

Anyway I'll come back later with more detail dude. In the meantime did you see this?
http://forum.overclock3d.net/showpost.php?p=622405&postcount=16
What do you think?

Looks like I'm not going to be doing it with BF3 though which is annoying because I've just reinstalled it to find out about this "your data is corrupt" patch screw up and I'm not sure I cba to fix it myself. GG Dice. Anyone got a fix? Otherwise I'll try and use FarCry3.
 
Last edited:
Only using a single 7950... butter smooth to me and never had a lag/stutter issue(besides my tv which is awfully slow lol):)

But i will add i have only played Shogun 2 so far with my new rig so will see soon as i test more games. Hopefully nothing goes wrong but i will report back here if anything does.
 
Yeah dont worry if you only have a single card. The AMD video cards are a better bang for the buck if you ask me. The only time when i would not buy amd is when you want do go multi GPU.
 
Oh ok thanks for saving me testing time lol:)

Though i read somewhere the newest driver 13.4(i think) should fix the issues you guys proclaim:p
 
Myself and many others have thought the current min, average and max FPS way of portraying the performance of gpus was outdated. This new way of measuring and depicting the viewing experience is refreshing (no pun intended).

Does it does mean that EVERY review that used FRAPS to measure performance in the past is now wrong? At least in dual card benchmarks. I wonder how other benchmark programs fare? A frame is a frame, even if it is 5px tall!

Have AMD (who must have known about this) been a bit naughty? Shines a poor light on their fps claims.

I wonder if there is another way this can be captured and analysed without the need for all the gubbins pcper.com used. I would have thought a program could be written that captures the frame number and the amount of pixels (x-pixels X y-pixels) in that frame, and create a log file. Then the little runts could be detected, if for example 1920 x 5 = runt.

Interesting times ahead.
 
Mysterae said:
Does it does mean that EVERY review that used FRAPS to measure performance in the past is now wrong? At least in dual card benchmarks. I wonder how other benchmark programs fare? A frame is a frame, even if it is 5px tall!
Click to expand...

Yes, it does mean that techniques used in the past were not accurate and people were miss-sold dual GPU set ups. That's technology though I guess and with any luck from now on review procedure will change and become more accurate. I've been doing a lot of reading about this and it will definitely improve things IMO. As for a frame being a frame even if it's 5px tall?

I think you need to do more reading mate. Runt frames actually damage performance because those 5px frames are not accurate and cause jitters and tearing. They hurt performance and are of no increase whatsoever. Do some Googling on FCAT and you'll find articles that actually picture it. It's not pretty.

Mysterae said:
Have AMD (who must have known about this) been a bit naughty? Shines a poor light on their fps claims.
Click to expand...

Unless one of their employees or ex employees came forward then we have no way of knowing for sure. However, at their level of the industry and with the technology they have? of course they knew about it.

Either way now that this evidence has come to light and things evolve they're going to have to work to sort it out. Can you see many of the forthcoming 7990s selling? because I can't.

Mysterae said:
I wonder if there is another way this can be captured and analysed without the need for all the gubbins pcper.com used. I would have thought a program could be written that captures the frame number and the amount of pixels (x-pixels X y-pixels) in that frame, and create a log file. Then the little runts could be detected, if for example 1920 x 5 = runt.

Interesting times ahead.
Click to expand...

You can get the FCAT software from the internet. It's open source and it's free. The problem is the capture hardware, then, the hard drive set up needed to be able to capture it properly. It churns out 25gb of video in just 30 seconds :o

However, any decent reviewer that knows their salts will have to adopt this technology :)

Now obviously the easiest way to make this work would be to be able to read it on the fly. However, you would still need a RAID array of SSDs just to be able to work quickly enough to read it on the fly.

Either way it doesn't matter because even if it cost say £2000 to set this up (not researched the capture devices used !) it's worth it for reviewers in the long run. The better the information they provide the more readers they will get :)

Oh and I agree yes, this is very interesting indeed :)

Edit. Here, a runt frame 5px high.

vdub.png


Note how it has caused a tear when analyzed. Runt frames are imperceptible so they add nothing to the game. The end user will not see them so they are removed for the actual FPS.
 
Last edited:
AlienALX said:
As for a frame being a frame even if it's 5px tall?

I think you need to do more reading mate. Runt frames actually damage performance because those 5px frames are not accurate and cause jitters and tearing. They hurt performance and are of no increase whatsoever. Do some Googling on FCAT and you'll find articles that actually picture it. It's not pretty.
Click to expand...

I think you missed my sarcasm? I should have posed it as a question. I was referring to benchmark software detecting a 5px frame as a valid, legitimate value adding frame, which we know it is not. If, like FRAPS, benchmark software adds all the frames it detects as part of the score, then they will also be wrong to a degree? But I don't know how the likes of Unigine equate their scores.
 
Mysterae said:
I think you missed my sarcasm? I should have posed it as a question. I was referring to benchmark software detecting a 5px frame as a valid, legitimate value adding frame, which we know it is not. If, like FRAPS, benchmark software adds all the frames it detects as part of the score, then they will also be wrong to a degree? But I don't know how the likes of Unigine equate their scores.
Click to expand...

Sorry mate not the best at picking up sarcasm :lol:

I can't wait until things like 3DMark11 and Unigene are dissected. Especially Unigene because it always favoured AMD cards !
 
You must log in or register to reply here.

Similar threads

Kaapstad
Should Nvidia and AMD be broken up?
Replies
4
Views
290
KingNosser
K
Dawelio
Corsair iCUE Issue
Replies
1
Views
996
Warchild
W
AlienALX
For Robbie.
Replies
4
Views
94
AlienALX
AlienALX
J
Is LTT forum is choke full of spooks ?
Replies
5
Views
477
hmmblah
hmmblah
g0ggles1994
Kitchen Flooring
Replies
8
Views
533
g0ggles1994
g0ggles1994
Back
Top