[PCPER]Frame-rating, what it is, what is does, and why it's important.

[cadaveca]

For those that don't want to read, and appreciate video:

Ryan has done a good job, and at the same time, I'd like to point out the cost of the systems used to test and show this, and the huge amount of work involved.


Shows real well how Crossfire is useless right now, FPS be damned, nearly any review showing FPS in Crossfire shows useless numbers, potentially faked, and it's not any reviewer's fault....technically, AMD has been cheating.


Yeah, all of you that spent money on a second card for Crossfire, did it for nothing...really nothing...for now.


And THAT is why AMD is going to fix it.




I said a while ago, I give them 4 months...if they don't, I smell a big huge lawsuit, and AMD's demise as a company.

For real. This could kill AMD's graphics division.

Now, the question is..did AMD really know about it? I say yes. Proving that, on the other hand...

 

[radrok]

Man I really hope they get this sorted out, they must.

I've been loyal to ATI for years but seriously it's been nothing short of headaches.

 

[Nordic]

Is it bad if this makes me wonder how cheap amd gpu's will become?

 

[cadaveca]

Man I really hope they get this sorted out, they must.

the runt frames are every second frame

= real performance = 50% reported FPS = singlecard FPS. Wow, me reporting for years that the second card in Crossfire seemed useless really was true. Go figure. I apologize, but I cannot help but feel a bit vindicated here.

The behavior is present on 50% of titles

Yeah, it's a big issue, and yeah, they'll fix it. I do have faith. Maybe this is why they fired so much of their driver staff.

 

[MxPhenom 216]

Wow! That was pretty interesting video. Will definitely sit down and read the article as well. Maybe during my flight home on Saturday from Hawaii. Thanks for posting this Dave. AMD needs to get this fixed!

 

[Dillinger]

What a great post, very informative

 

[BiggieShady]

Ah, finally an update! I've been following PCPER articles about frame latencies for some time now ...

 

[Outback Bronze]

I can see they wernt using vsync. I saw screen tearing. Would this make a difference?

Interesting article although im happy playing crysis 3 with my 7950 xfire. I was getting 20fps Vsyncd all maxed and now i get 50/60 coz of new 13.3 beta.

 

[MxPhenom 216]

I can see they wernt using vsync. I saw screen tearing. Would this make a difference?

Interesting article although im happy playing crysis 3 with my 7950 xfire. I was getting 20fps Vsyncd all maxed and now i get 50/60 coz of new 13.3 beta.

They said they will have vsync tests in later parts of the article. Should be in a few weeks.

 

[Divide Overflow]

So the extractor program that nVidia helped write shows problems with ATI systems? Well it's not like nVidia software would do something like hamper performance, disable ATI features or anything else like that. Oh wait....

 

[Outback Bronze]

They said they will have vsync tests in later parts of the article. Should be in a few weeks.

Yeah doh i should have finished watching the entire video. Cheers.

 

[BiggieShady]

Well it's not like nVidia software would do something like hamper performance, disable ATI features or anything else like that. Oh wait....

That particular software is analyzing video captured data (what gamer sees) and it is not in position to affect anything. As you can see in this image, FRAPS can't get this data because it's a software process operating on a different OS layer (than GPU driver) :

Even if measured at GPU driver level, it would not be as accurate as video capture analysis.

 

[LAN_deRf_HA]

So the extractor program that nVidia helped write shows problems with ATI systems? Well it's not like nVidia software would do something like hamper performance, disable ATI features or anything else like that. Oh wait....

If you'd been following along you'd know all this tool does is confirm what AMD themselves admitted. Their focus has been on fixing stuttering on single cards first, and they say they'll shift focus to multi gpu in a few months. My guess is to coincide with the reference 7990 launch. I don't think they knew they had these problems before very recently and them prioritizing the more common configurations is to be expected. They'll get there.

 

[BiggieShady]

Now, the question is..did AMD really know about it? I say yes. Proving that, on the other hand...

This image proves it for me, it would be hard for AMD driver team to miss all these missed frames

 

[cadaveca]

This image proves it for me, it would be hard for AMD driver team to miss all these missed frames

http://www.pcper.com/files/review/2013-03-25/run.stats__3.png

5760x1080 is the biggest offender, for sure. I've mentioned many times that it is odd that that res with 50% more pixels than 2560x1600 performs better than 2560x1600 does, and not @ 66% of the performance of 2560x1600, as you'd expect.



It is an interesting topic for discussion. I know AMD can fix it. 1000%. only since the 7990 is launching. I mean, it explains why they refused to launch a dual GPU card for so long...they new dual-GPU wasn't working right.

They did the same with the 5970.


They even mentioned that Crossfire and Eyefinity were not supported together with 5-series cards.


I don't think it's as bad as even I have portrayed it.. it's a huge issue, in that it makes me physically ill, and unable to enjoy playing my games when using multi-GPU configurations. In a way, I'm a physical meat and bone representation of Ryan's fancy software and hardware.

Be that as it may, I am now more interested in the why, the when, and not much else. Fixing it isn't a big deal, really. I mean, I fix it daily...by not using my 2nd, third, and fourth GPUs.






The real unfortunate part is the cost in testing such things is still prohibitive for nearly everyone. SSD Raid 0, TB's of data, Thunderbolt...multiple GPUs...oh look, thing's I've put on my shopping list for W1zz to get me in the past that I ended up buying myself. Oh well.


OH, and Happy Easter, everyone! Look at the shit the bunny left behind!

 

[Rowsol]

Fascinating.

 

[cadaveca]

Fascinating.

Meh. Most people seem to not care.

 

[the54thvoid]

Meh. Most people seem to not care.

I think it's because this is definitely an area where logical arguments win and juvenile brand loyal sabre rattling is pointless. I posted a link to an Anandtech article (in another thread) that also looked at the issue that coincided with an AMD conference call. Amd have hand on heart admitted it all and that they were relying on a tool for the FPS numbers. While Nvidia started a few years back trying to argue for other measuring tools.

It's pretty black and white. The usual 'fan'atic posters wont come here.

 

[tokyoduong]

and there's people here that will swear up and down that they never have problems with CF lol.

Just the slight microstutter CF produce alone already gives me headaches. I know others don't experience it but I do. I really hope they fix this soon because it doesn't look like a hard fix.

 

[Xzibit]

Atleast PCPerspective got AMD and Nvidia to agree on something.

That this isnt a good way to measure frame latency.

 

[tokyoduong]

Meh. Most people seem to not care.

I do and hence why I don't own any CF systems. I always felt weird using CF since the 2k series running 2850. That's not even accounting all the bugs associated with it. It always gives me headaches and makes me not want to play games. Any gaming past an hour starts becoming torturous. I always thought it was probably just my eyes being weird so i avoid it and didn't complain.

I've set up many CF systems for many people and it works like it should. Testing it was always a pain for me. Most people didn't think the microstutter was a big deal and didn't bother them. So to each his own but this new discovery just makes my view about CF/SLI just a little stronger. It is just too many problems to deal with. I would rather just update my card more often than double up the GPU.

BTW i don't think this will kill their graphics division. There's very few people that actually do CF. As long as they fix it in time then it shouldn't be much of a problem. Worst case, they send everyone a $50 check that can prove they bought CF set up.

 

[DRDNA]

Well this is really good news as things will get better and performance will increase and we will be happy!

 

[Sasqui]

Meh. Most people seem to not care.

Certainly those with 7xxx CrossFire must.

If all that analysis is correct, I think you're spot on about the 7990 delays.

 

[DRDNA]

Certainly those with 7xxx CrossFire must.

If all that analysis is correct, I think you're spot on about the 7990 delays.

Can you speculate how long this issue has been going on? Is it all the way back to the X800 days?

 

[Xzibit]

The only real issue i have with this is Ryan has gone on record saying Nvidia wrote the overlay for the frame-grabbing for there software which is also nvidias.

So it throws a wrench into the testing..

It would have been better if a 3rd-party would be the one writing the software to grab the frames.

They way they are grabbed after the renderer could benefit or cater specific to nvidia since they wrote it.