[QGIS-Developer] Some initial OpenCL feedback

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

[QGIS-Developer] Some initial OpenCL feedback

Andreas Neumann-4

Hi Alessandro,

I just tested if the hardware acceleration (OpenCL) helps in the live hillshade renderer.

The good news is that it works stable so far and substantially improves rendering time of the live hillshade renderer. My notebook on my workplace is rather slow. Without hardware acceleration it renders the hillshade in approx 8 seconds (large window on two screens). With OpenCL acceleration set to on, it renders in about 1 second or even less. I assume quite some time is also spent on reading the file. This with resampling on (average for zooming out and bilinear for zooming in beyond native resolution).

Hardware is an Intel HD Graphics 4600. The CPU is also listed as supported.

Visually, with or without HW accelation on, there still are some rectangular artefacts in both versions.

Tested on Windows.

--------------

One more thing: on my machine there is no restart of QGIS necessary for the changes (HW acceleration on of off) to be applied. This is contrary to the statement in the settings dialogue, which says, that a restart of QGIS is necessary.

---------------

Thanks a lot for your work!

Andreas


_______________________________________________
QGIS-Developer mailing list
[hidden email]
List info: https://lists.osgeo.org/mailman/listinfo/qgis-developer
Unsubscribe: https://lists.osgeo.org/mailman/listinfo/qgis-developer
Reply | Threaded
Open this post in threaded view
|

Re: Some initial OpenCL feedback

Alessandro Pasotti-2

On Tue, Sep 4, 2018 at 2:21 PM Andreas Neumann <[hidden email]> wrote:

Hi Alessandro,

I just tested if the hardware acceleration (OpenCL) helps in the live hillshade renderer.



Hi Andreas, 

thanks a lot for your feedback, I honestly am a little worried about the possible problems on different platforms, graphic cards and driver versions, I choose to stay on the safe side and there is a safeguard that disables the OpenCL support at the first issue in the renderer.

The processing algs looks generally more stable (never got a single crash myself).


The good news is that it works stable so far and substantially improves rendering time of the live hillshade renderer. My notebook on my workplace is rather slow. Without hardware acceleration it renders the hillshade in approx 8 seconds (large window on two screens). With OpenCL acceleration set to on, it renders in about 1 second or even less. I assume quite some time is also spent on reading the file. This with resampling on (average for zooming out and bilinear for zooming in beyond native resolution).


Yes some time is spent on reading the file and some time is spend on I/O tranferring data to/from the graphic card, this means that the performance gain is usually higher on larger images than on smaller images.

Hardware is an Intel HD Graphics 4600. The CPU is also listed as supported.


Can you also try the CPU? I'm curious about the performances.
 

Visually, with or without HW accelation on, there still are some rectangular artefacts in both versions.


Yes: that does not change, the algorithm is pretty much the same in both implementations.

 

Tested on Windows.

--------------

One more thing: on my machine there is no restart of QGIS necessary for the changes (HW acceleration on of off) to be applied. This is contrary to the statement in the settings dialogue, which says, that a restart of QGIS is necessary.


The text can probably be improved: what needs a restart is if you change the device, enabling and disabling OpenCL support does not require a restart.
 

---------------

Thanks a lot for your work!



Thank you for the feedback!

 


--
Alessandro Pasotti
w3:   www.itopen.it

_______________________________________________
QGIS-Developer mailing list
[hidden email]
List info: https://lists.osgeo.org/mailman/listinfo/qgis-developer
Unsubscribe: https://lists.osgeo.org/mailman/listinfo/qgis-developer
Reply | Threaded
Open this post in threaded view
|

Re: Some initial OpenCL feedback

Andreas Neumann-4

Hi Alessandro,

Thanks for the clarification that only the change of the OpenCL-device needs a restart.

I tried both - acceleration in CPU and in GPU. I had the impression that the acceleration in the CPU was slightly faster than in the CPU - but not a lot faster. Maybe the GPU (Intel HD Graphics 4600) is not too powerful. It is a ThinkPad notebook, not a desktop.

Greetings,

Andreas

On 2018-09-04 15:56, Alessandro Pasotti wrote:


On Tue, Sep 4, 2018 at 2:21 PM Andreas Neumann <[hidden email]> wrote:

Hi Alessandro,

I just tested if the hardware acceleration (OpenCL) helps in the live hillshade renderer.

 
 
Hi Andreas, 
 
thanks a lot for your feedback, I honestly am a little worried about the possible problems on different platforms, graphic cards and driver versions, I choose to stay on the safe side and there is a safeguard that disables the OpenCL support at the first issue in the renderer.
 
The processing algs looks generally more stable (never got a single crash myself).
 
 

The good news is that it works stable so far and substantially improves rendering time of the live hillshade renderer. My notebook on my workplace is rather slow. Without hardware acceleration it renders the hillshade in approx 8 seconds (large window on two screens). With OpenCL acceleration set to on, it renders in about 1 second or even less. I assume quite some time is also spent on reading the file. This with resampling on (average for zooming out and bilinear for zooming in beyond native resolution).

 
Yes some time is spent on reading the file and some time is spend on I/O tranferring data to/from the graphic card, this means that the performance gain is usually higher on larger images than on smaller images.
 

Hardware is an Intel HD Graphics 4600. The CPU is also listed as supported.

 
Can you also try the CPU? I'm curious about the performances.
 

Visually, with or without HW accelation on, there still are some rectangular artefacts in both versions.

 
Yes: that does not change, the algorithm is pretty much the same in both implementations.
 
 

Tested on Windows.

--------------

One more thing: on my machine there is no restart of QGIS necessary for the changes (HW acceleration on of off) to be applied. This is contrary to the statement in the settings dialogue, which says, that a restart of QGIS is necessary.

 
The text can probably be improved: what needs a restart is if you change the device, enabling and disabling OpenCL support does not require a restart.
 

---------------

Thanks a lot for your work!

 
 
Thank you for the feedback!
 
 

 
--
Alessandro Pasotti
w3:   www.itopen.it



_______________________________________________
QGIS-Developer mailing list
[hidden email]
List info: https://lists.osgeo.org/mailman/listinfo/qgis-developer
Unsubscribe: https://lists.osgeo.org/mailman/listinfo/qgis-developer
Reply | Threaded
Open this post in threaded view
|

Re: Some initial OpenCL feedback

Alessandro Pasotti-2


On Tue, Sep 4, 2018 at 4:57 PM Andreas Neumann <[hidden email]> wrote:

Hi Alessandro,

Thanks for the clarification that only the change of the OpenCL-device needs a restart.

I tried both - acceleration in CPU and in GPU. I had the impression that the acceleration in the CPU was slightly faster than in the CPU - but not a lot faster. Maybe the GPU (Intel HD Graphics 4600) is not too powerful. It is a ThinkPad notebook, not a desktop.


Thanks for the info!

Yes, maybe that the lower I/O (there is no need to transfer data to/from the graphic card) compensates the slower processing on the CPU
 
Cheers

Greetings,

Andreas

On 2018-09-04 15:56, Alessandro Pasotti wrote:


On Tue, Sep 4, 2018 at 2:21 PM Andreas Neumann <[hidden email]> wrote:

Hi Alessandro,

I just tested if the hardware acceleration (OpenCL) helps in the live hillshade renderer.

 
 
Hi Andreas, 
 
thanks a lot for your feedback, I honestly am a little worried about the possible problems on different platforms, graphic cards and driver versions, I choose to stay on the safe side and there is a safeguard that disables the OpenCL support at the first issue in the renderer.
 
The processing algs looks generally more stable (never got a single crash myself).
 
 

The good news is that it works stable so far and substantially improves rendering time of the live hillshade renderer. My notebook on my workplace is rather slow. Without hardware acceleration it renders the hillshade in approx 8 seconds (large window on two screens). With OpenCL acceleration set to on, it renders in about 1 second or even less. I assume quite some time is also spent on reading the file. This with resampling on (average for zooming out and bilinear for zooming in beyond native resolution).

 
Yes some time is spent on reading the file and some time is spend on I/O tranferring data to/from the graphic card, this means that the performance gain is usually higher on larger images than on smaller images.
 

Hardware is an Intel HD Graphics 4600. The CPU is also listed as supported.

 
Can you also try the CPU? I'm curious about the performances.
 

Visually, with or without HW accelation on, there still are some rectangular artefacts in both versions.

 
Yes: that does not change, the algorithm is pretty much the same in both implementations.
 
 

Tested on Windows.

--------------

One more thing: on my machine there is no restart of QGIS necessary for the changes (HW acceleration on of off) to be applied. This is contrary to the statement in the settings dialogue, which says, that a restart of QGIS is necessary.

 
The text can probably be improved: what needs a restart is if you change the device, enabling and disabling OpenCL support does not require a restart.
 

---------------

Thanks a lot for your work!

 
 
Thank you for the feedback!
 
 

 
--
Alessandro Pasotti
w3:   www.itopen.it




--
Alessandro Pasotti
w3:   www.itopen.it

_______________________________________________
QGIS-Developer mailing list
[hidden email]
List info: https://lists.osgeo.org/mailman/listinfo/qgis-developer
Unsubscribe: https://lists.osgeo.org/mailman/listinfo/qgis-developer