Inference Pattern: App Uses GPU-->App is Running As Efficiently As Possible on rMBP?

[rhyzome]

If an application/process shows up in Activity Monitor as making use of the GPU, can one safely assume that that app/process is coded in a way optimized for the GPU (CUDA support for 750M on the current 15"?), and that, if the process were performed instead on the integrated Iris Pro, it would have taken more time?

(For example, Adobe Acrobat Pro seems to make use of my GPU when I OCR my PDFs...is it operating more efficiently using the GPU for this, or should I force it to use the integrated Intel GPU?)

Is the inference pattern 'if an app or process automatically makes use of the GPU it must be performing the given task the most efficient way' valid?

If it isn't valid, how do you know if/when apps are making efficient/wise use of the dGPU?

 

[flybass]

No, some code may just be poorly written and trip up the automatic switching. The dGPU is much better performance-wise, but if you're running on battery it is best to stay on integrated. Try gfxcardstatus - you can use it to force the iGPU before opening the app.

 

[leman]

No, its actually the opposite. If an app uses the GPU for no apparent reason (such as when not performing a demanding workload), then its not coded properly and is wasting system resources. An app should trigger the discrete GPU only for brief periods of time and only when actually needing the performance (except when its a game, where the GPU usage is justified at any time).

 

[dusk007]

Sometimes it just so happens to be true but that is really by accident don't by design. More often than not the opposite is true as the other post have written.
In all either case it is usually the fault of bad software.
With the iGPU the system can power down much lower when no load is there. The dGPU stays active the whole time an app is open even if it just sits there but if there actually is something to do it sometimes does the same work using less power. Usually a fault of bad drivers or badly written code for the Intel GPU. Sometimes not even the programmers fault but Apple not giving enough access to the GPU APIs in a decent way.

Under windows things work very different and much better but that is another story.