GPU Mist microstutter analysis tool

[Gregster]

Just thought I would put this info out, as it seems a pretty dent little tool that measures microstutter. I have had a little play with it but still working out what is what, so will come back to this thread a bit later when I am more clued up.

Download it from here

http://www.5group.com/wordpress/2012/07/14/gpu-mist-pre-release-1-0-rc1/

Use FRAPS to measure your game and tick the box that has frametimes monitoring. Run your game, run the bench and see what is what

This was mine from COD: AW

Thoughts?

 

[shankly1985]

Could be a good tool.. But wasn't Fraps known to give wrong readings? reason for PCper to show more high end gear for checking...

Could be wrong! I wish more games had what Frostbite as a proper built in real time graph.

Edit
Just tried to open the CSV file from Frostbite but wouldn't open. shame

 

[Rroff]

^^ FRAPs (same with any other software tool that doesn't have access to hardware debugging on a dev board) can't show an entirely accurate picture but its close enough to give a rough idea for comparison purposes.

PCPer was using hardware capture so as to analyse what level of microstutter was actually being shown onscreen whereas FRAPs can only see what was sent to be displayed onscreen but not any delays that might happen between the frame being dispatched and actually appearing onscreen. (Which doesn't take into account things like nVidia's hardware smoothing).

 

[shankly1985]

^^ FRAPs (same with any other software tool that doesn't have access to hardware debugging on a dev board) can't show an entirely accurate picture but its close enough to give a rough idea for comparison purposes.

PCPer was using hardware capture so as to analyse what level of microstutter was actually being shown onscreen whereas FRAPS can only see what was sent to be displayed onscreen but not any delays that might happen between the frame being dispatched and actually appearing onscreen.

So we just going to be looking a false reading then.. Least with Frostbite its from real time, its raw date right from the rendering.. Its not hooked in like Fraps or any other software.

 

[Rroff]

Frostbite's data won't be any different* to what FRAPs can do - it can only see what has been dispatched to the hardware not what happens to it in the final stage on the hardware.


* Might be some low level timing accuracy differences but not so much that will massively skew the picture.

 

[shankly1985]

Frostbite's data won't be any different* to what FRAPs can do - it can only see what has been dispatched to the hardware not what happens to it in the final stage on the hardware.


* Might be some low level timing accuracy differences but not so much that will massively skew the picture.

I think if you was to test Fraps and then test BF4 own tool you find results will be completely different.

Who willing to test while I work?

 

[Gregster]

^^ FRAPs (same with any other software tool that doesn't have access to hardware debugging on a dev board) can't show an entirely accurate picture but its close enough to give a rough idea for comparison purposes.

PCPer was using hardware capture so as to analyse what level of microstutter was actually being shown onscreen whereas FRAPs can only see what was sent to be displayed onscreen but not any delays that might happen between the frame being dispatched and actually appearing onscreen. (Which doesn't take into account things like nVidia's hardware smoothing).

Yer, it certainly isn't anything like FCAT and can only read what goes to the GPU and not the screen. I will have a play some more and see if it picks up any stuttering.

 

[andybird123]

Frostbite's data won't be any different* to what FRAPs can do - it can only see what has been dispatched to the hardware not what happens to it in the final stage on the hardware.


* Might be some low level timing accuracy differences but not so much that will massively skew the picture.

+1
Anything that is software based wont be able to reflect what you see on screen, as there is physical hardware between the software and your eyeballs

Outside of driver tricks saying a frame has been completed when it hasnt, fraps is going to be no more or less accurate than any other software based output

 

[Rroff]

I think if you was to test Fraps and then test BF4 own tool you find results will be completely different.

Who willing to test while I work?

They will still both be limited in the same was as to not being able to see the whole picture. As an aside I wouldn't trust any numbers BF4 pulls out anyhow (even if FRAPs was inaccurate in this case) the game is a broken POS.

No software solution can give you a 100% accurate picture of microstutter unless it has access to hardware counters usually requiring a development board with a debugging setup or an external hardware capture device. (Not sure what is possible with perfkit or nsight with development/debug drivers as its a bit out of my area of experience).

FRAPs is close enough to get a general picture for comparison purposes.

 

[shankly1985]

Fair enough, BF4 works fine here but that's another discussion.

 

[Rroff]

I would certainly be wary of using FRAP's numbers as the be all and end all to judge it by but for approximate comparisons or to look for issues that would merit a deeper look it can be useful.

What I find it most useful for though is getting a real picture of the minimum frame rate situations as a single figure never does this justice. As an aside its also far more useful IMO for showing the quality of experience around 60fps than spitting out min/avg/max figures.

 

[Gregster]

I would certainly be wary of using FRAP's numbers as the be all and end all to judge it by but for approximate comparisons or to look for issues that would merit a deeper look it can be useful.

What I find it most useful for though is getting a real picture of the minimum frame rate situations as a single figure never does this justice and the consistency of frame rates around 60fps as both of these are far more meaningful on the quality of experience than a single minimum fps number or min/avg/max numbers.

Agreed. Fraps was the go to tool and this is the same thing with a different readout but it isn't a patch on FCAT. I remember the thing with Crossfire kicking off and FCAT was showing runt frames, which FRAPS was missing.

 

[tommybhoy]

Posted it aaaaaages ago somewhere, you've been reading too much Russian game BM's.