Software for diagnosing bottlenecks?

[DeadlyPilot]

I noticed that XPlane had this cool thing where it would show how much time both the CPU, and the GPU had each spent working on a single frame. I figured this was great for testing whether you where CPU or GPU limited. Is there a name for this? Are there any other applications that do this? (Look in the top left)

 

[DrMacintosh]

Yeah, MSI Afterburner. 

[DeadlyPilot]

Just now, DrMacintosh said:

Yeah, MSI Afterburner. 

Do you mean just CPU and GPU usage, or does it actually show the frame times?

[DrMacintosh]

Just now, DeadlyPilot said:

Do you mean just CPU and GPU usage, or does it actually show the frame times?

It will tell you how much of each is or isnt being used

 

If CPU is at 100% but GPU is at like 50% then you know its the CPU. 

 

If the GPU is at 100% but the GPU is at 50% you know its the GPU. 

[Mira Yurizaki]

I just run Task Manager in the Performance tab and GPU-Z in the sensors tab. If I need to do any more in depth analysis I fire up Performance Monitor to log CPU activity and log the sensors on GPU-z

 

Theres nothing that will go "you have a bottleneck!" You have to determine that yourself.

 

Though it would be nice to trace threads going across the scheduler

[Jurrunio]

MSI Afterburner and RivaTunerStatisticsServer bundled with it (though you can update it independently). RTSS is required to show these info in games, but it grabs these data from MSI Afterburner

[ARikozuM]

It'll display frame-times as well. 

[DeadlyPilot]

3 minutes ago, DrMacintosh said:

It will tell you how much of each is or isnt being used

 

If CPU is at 100% but GPU is at like 50% then you know its the CPU. 

 

If the GPU is at 100% but the GPU is at 50% you know its the GPU. 

That makes sense. However, since many applications will not utilize every thread, even if the CPU wasn't the bottleneck you would still get less than 100% usage. So this really only works for checking GPU bottlenecks right?

[ARikozuM]

3 minutes ago, DeadlyPilot said:

That makes sense. However, since many applications will not utilize every thread, even if the CPU wasn't the bottleneck you would still get less than 100% usage.

You should never measure a CPU's performance by it's total usage. You need to look at the individual cores to find what the actual performance is.

 

This is one of the biggest misconceptions to an i7 being better than an i5 or, more generally, an octa-core being better than a quad-core since 25% looks better than 50% when using two cores fully while the others idle.