Mantle and Frame Latency

Ian Evey · 4 Oct 2014 at 14:47

That's strange, I only see input lag with vsync on it certainly does not stutter just feels sluggish.

Again it's my own perception though, Maybe we all get used to it on our own pc's.

setter · 4 Oct 2014 at 14:55

Rusty0611 said:
That's the way I look at it but people will use whatever tools they can grasp at to make fallacious points in favour of their preferred GPU brand. Having tried both of them very recently there just wasn't any perceptible difference for me between the two brands Mantle included. Both DX through nVidia and Mantle through AMD worked well and felt smooth.

Ive only had a brief play with mantle in online bf4. Tbh it felt a bit smoother, but then for online the system on which i used mantle was connected to an 80mb fibre line. If i get on a decent server on my own adsl connection, bf4 under dx plays grand too.

CAT-THE-FIFTH · 4 Oct 2014 at 14:59

Its quite funny in late 2012 and early 2013 when the Never Settle driver came out,and the releases clustered right after them and websites found frametime issues with some of the AMD drivers,using both the TR methodology and the one that PCPER helped develop with Nvidia,people were all over it saying AMD crap drivers,etc. Runt frames(as Nvidia called them for the dual card results) and all that stuff.

Now that the same review sites,say the AMD drivers might even be better than Nvidia in a number of games,the frametime charts don't matter anymore.

LOL.

TheRealDeal · 4 Oct 2014 at 15:01

CAT-THE-FIFTH said:
Its quite funny in late 2012 and earlu 2013 when the Never Settle driver came out and websites found frametime issues with some of the AMD drivers,people were all over it saying AMD crap drivers.

Now the same review sites,say the AMD driver might even be better than Nvidia in a number of games,the frametime charts don't matter anymore.

LOL.

+1

Ian Evey · 4 Oct 2014 at 15:19

I think you will find that amd users said that all there games where smooth anyway and this info is meaningless, Not so meaningless now is it?

LOL

humbug · 4 Oct 2014 at 15:23

CAT-THE-FIFTH said:
Its quite funny in late 2012 and early 2013 when the Never Settle driver came out,and the releases clustered right after them and websites found frametime issues with some of the AMD drivers,using both the TR methodology and the one that PCPER helped develop with Nvidia,people were all over it saying AMD crap drivers,etc. Runt frames(as Nvidia called them for the dual card results) and all that stuff.

Now that the same review sites,say the AMD drivers might even be better than Nvidia in a number of games,the frametime charts don't matter anymore.

LOL.

Yes, and it speaks volumes.

Final8y · 4 Oct 2014 at 16:38

Ian Evey said:
I think you will find that amd users said that all there games where smooth anyway and this info is meaningless, Not so meaningless now is it?

LOL

Indeed there were comments like that as they did not have a first hand comparison beforehand, but lets not make out that they have changed there mind after the experience only recently, it started to matter quite some time ago.

Final8y · 4 Oct 2014 at 17:11

andybird123 said:
so his graph goes 5, 10, 16?
Dont think so, it goes 5,10,15,20,25 clearly, so he has only a few result over 16 and even where they are he has none ranging to 40+ unlike you

If results dont lie then the nvidia graph very clearly shows less skipping and skipping of a shorter duration

This is the closest to shankly1985 and jwrpalmer99 situation, only in reverse.

A thin line with the odd spike once in a blue moon is still more smoother over all than a fuzzy line with no spikes.

The 290X in CF had a fuzzy line but no spikes while the 980 in SLI had a few spikes but felt smoother over all.

For our multi-GPU testing,at 2560x1440, the AMD Radeon R9 290X cards in CrossFire showed much more frame time variance than the GTX 980 cards running in SLI. So even though the average frame rate of the 290X CF combo was higher, the better experience was clearly on SLI.

That isn't the case at 4K though where frame time variance is actually better on the AMD solution - things have been swapped!

http://www.pcper.com/reviews/Graphi...and-GTX-970-GM204-Review-Power-and-Efficiency

tommybhoy · 4 Oct 2014 at 17:26

Final8y said:
Indeed there were comments like that as they did not have a first hand comparison beforehand, but lets not make out that they have changed there mind after the experience only recently, it started to matter quite some time ago.

I held my hands up and ate the humble pie when first ran SD on the fixed frame latency drivers-night and day. :cool:

I also remember a (vocal)user in this very thread stating you couldn't mine because 'crossfire doesn't give you better results than a single card' and posted PCP's dual setup gaming expose as proof at the time.

humbug · 4 Oct 2014 at 17:45

tommybhoy said:
I held my hands up and ate the humble pie when first ran SD on the fixed frame latency drivers-night and day.

I also remember a (vocal)user in this very thread stating you couldn't mine because 'crossfire doesn't give you better results than a single card' and posted PCP's dual setup gaming expose as proof at the time.

You don't really know until you experience it, even then not everyone can tell

pgi947 · 4 Oct 2014 at 17:53

Some of you should have a read over this: http://international.download.nvidia.com/geforce-com/international/pdfs/FCAT_Reviewer's_Guide.pdf

Nvidias reviewer guide on using FCAT, going by the information in that PDF, the frame time tools we use here on this forum are largely irrelevant. Measuring the wrong thing at the wrong time.

If someone wants to pony up £1500+ on a capture card and raid SSD setup capable of accurately measuring frame times feel free

shankly1985 · 4 Oct 2014 at 18:00

pgi947 said:
Some of you should have a read over this: http://international.download.nvidia.com/geforce-com/international/pdfs/FCAT_Reviewer's_Guide.pdf

Nvidias reviewer guide on using FCAT, going by the information in that PDF, the frame time tools we use here on this forum are largely irrelevant. Measuring the wrong thing at the wrong time.

If someone wants to pony up £1500+ on a capture card and raid SSD setup capable of accurately measuring frame times feel free

But am not using FCAT? my results are coming right from the Battlefield Engine. Perfoverlay.framefilelog 1

humbug · 4 Oct 2014 at 18:03

pgi947 said:
Some of you should have a read over this: http://international.download.nvidia.com/geforce-com/international/pdfs/FCAT_Reviewer's_Guide.pdf

Nvidias reviewer guide on using FCAT, going by the information in that PDF, the frame time tools we use here on this forum are largely irrelevant. Measuring the wrong thing at the wrong time.

If someone wants to pony up £1500+ on a capture card and raid SSD setup capable of accurately measuring frame times feel free

So you have to look at Mantle vs DX FCAT results.

We know from PCper's DirectX FCAT comparisons that AMD are at least as good as Nvidia on DirectX, both are good.

Neither are as good as Mantle.

Almost perfect ^^^

layte · 4 Oct 2014 at 18:15

I see the thread has hit the petty bickering and wall of charts cherry picked benchmark time. Anyone interested in the full story of those chart posted on the previous page can find the full review here: http://www.pcper.com/reviews/Graphi...and-GTX-970-GM204-Review-Power-and-Efficiency

Final8y · 4 Oct 2014 at 18:26

I already posted the link.

pgi947 · 4 Oct 2014 at 18:30

shankly1985 said:
But am not using FCAT? my results are coming right from the Battlefield Engine. Perfoverlay.framefilelog 1

Read the PDF!

Capturing displayed frames is very difficult. If you control the driver, it can be done with software. But
drivers can be written to hide effects like dropped and runt fames. What is required is a method that
measures the actual output from the DVI connector before it’s delivered to the monitor.

Software tools, like Fraps, actually capture frames from the frame buffer off the GPU. However, this
data is not what you see: it is not representative of everything that is seen on the display. There is
actually a process called scan out, and that is the difference. What FCAT is measuring is what is
happening right at the display, as opposed to what is in the frame buffer.

Perfoverlay.framefilelogenable 1 will be reading from the frame buffer.

humbug said:
So you have to look at Mantle vs DX FCAT results.

We know from PCper's DirectX FCAT comparisons that AMD are at least as good as Nvidia on DirectX, both are good.

Neither are as good as Mantle.

snip

Almost perfect ^^^

I've never disputed Mantle gives more consistent (thus smoother) frame times. In fact me and matt did some testing many months ago and came to the same conclusion. Even in circumstances where min/max/avg were equivalent to DX, frame variance was sufficiently different.

andybird123 · 4 Oct 2014 at 18:34

CAT-THE-FIFTH said:
Its quite funny in late 2012 and early 2013 when the Never Settle driver came out,and the releases clustered right after them and websites found frametime issues with some of the AMD drivers,using both the TR methodology and the one that PCPER helped develop with Nvidia,people were all over it saying AMD crap drivers,etc. Runt frames(as Nvidia called them for the dual card results) and all that stuff.

Now that the same review sites,say the AMD drivers might even be better than Nvidia in a number of games,the frametime charts don't matter anymore.

LOL.

Your missing the point, the problem with crossfire at the time was that fraps was reporting frame times sub 16ms, where as fcat was showing consistently 32ms+ e.g. constant stutter, fraps was reporting solid 60fps but people were in effect getting 30fps

What this is now showing is frame time variance fluctuating from 5-10ms with a few frames at 15ms, well under the 16.6ms that would be perceived as a stutter at 60fps

So no, I am not saying that frame time variance is irrelevant, just that some people's interpretation of what acceptable variance is might be a bit wonky

shankly1985 · 4 Oct 2014 at 18:36

pgi947 said:
Read the PDF!

Perfoverlay.framefilelogenable 1 will be reading from the frame buffer.

I've never disputed Mantle gives more consistent (thus smoother) frame times. In fact me and matt did some testing many months ago and came to the same conclusion. Even in circumstances where min/max/avg were equivalent to DX, frame variance was sufficiently different.

No Perfoverlay.framefilelogenable isn't read from the buffer at all.. Its coming right from the engine just like having the graph on screen while playing only its outputting it into a easy to read file.

pgi947 · 4 Oct 2014 at 18:58

So you evidently haven't looked at the PDF then?

Skip to page 7, nice big picture for you (maybe easier to understand?) explaining it.

The line right after GAME ENGINE that says 'FRAPS measures this' is different to perfoverlay.framefilelogenable measuring from the GAME ENGINE how?

shankly1985 · 4 Oct 2014 at 19:17

The tool built into the frostbite engine isn't anything like fraps.
So basically you saying the performance graph on battlefield is giving wrong information? Because that's the data that is written to file.

What's the point in performance graph then?