Project

General

Profile

Bug #12308

Better Performance without OpenCL support

Added by Lorenzo Lu 9 months ago. Updated 24 days ago.

Status:
Closed: won't fix
Priority:
Low
Assignee:
-
Category:
-
Target version:
Start date:
09/02/2018
Due date:
% Done:

0%

Affected Version:
2.4.4
System:
other GNU/Linux
bitness:
64-bit
hardware architecture:
amd64/x86

Description

Hello,
I own an Acer Laptop with a Skylake i7-6500u and both an integrated GPU intel HD 520 and a NVIDIA GeForce 920m with 2GB of memory.
I've installed all the correct drivers to use OpenCL and everything worked fine by default, but I've always noticed that darktable was very slow to react to slider movements (it was not "real-time"), way slower than Lightroom for example.
Today I've tried to disable the OpenCL support into preferences and I've noticed that now the performance during the editing are way better, on par with Lightroom.
This happens with both the intel and the NVIDIA card, on both Arch Linux and Windows.
I know my GPU is not a monster of power, but shouldn't it give a little better performance anyway, instead of way worse ones?
And, just curious, should OpenCL impact all the operations in Darktable (the editing process and the export) or only the exporting time? (just to know if in the future I should buy a laptop with a good GPU or not, I don't really care about export time).

History

#1 Updated by Christian Kanzian 9 months ago

Hello,

Just some comments from an user.

Lorenzo Lu wrote:

I know my GPU is not a monster of power, but shouldn't it give a little better performance anyway, instead of way worse ones?

https://browser.geekbench.com/opencl-benchmarks lists the OpenCl performance of a 920m very low. I have a GTX 1060 Desktop GPU which gives much more performance. I would never expect high performance of mobile devices, as there is always a treade off btw. power consumption, cooling and performance.

And, just curious, should OpenCL impact all the operations in Darktable (the editing process and the export) or only the exporting time? (just to know if in the future I should buy a laptop with a good GPU or not, I don't really care about export time).

Hardware performance has an impact on editing and exporting performance. A good GPU will make editing more responsive. Here are some benchmarks for desktop GPU: https://www.phoronix.com/scan.php?page=article&item=darktable-opencl-gpu&num=3

I'm not aware of any mobile GPU benchmarks.

#2 Updated by Lorenzo Lu 9 months ago

Christian Kanzian wrote:

Hello,

Just some comments from an user.

Lorenzo Lu wrote:

I know my GPU is not a monster of power, but shouldn't it give a little better performance anyway, instead of way worse ones?

https://browser.geekbench.com/opencl-benchmarks lists the OpenCl performance of a 920m very low. I have a GTX 1060 Desktop GPU which gives much more performance. I would never expect high performance of mobile devices, as there is always a treade off btw. power consumption, cooling and performance.

And, just curious, should OpenCL impact all the operations in Darktable (the editing process and the export) or only the exporting time? (just to know if in the future I should buy a laptop with a good GPU or not, I don't really care about export time).

Hardware performance has an impact on editing and exporting performance. A good GPU will make editing more responsive. Here are some benchmarks for desktop GPU: https://www.phoronix.com/scan.php?page=article&item=darktable-opencl-gpu&num=3

I'm not aware of any mobile GPU benchmarks.

Thank you, I didn't realise that the 920m was so bad that it's actually weaker than the iGPU, what a nonsense. Well I guess this is the only reason, I'll try to use Darktable with a more powerful GPU. Thanks again!

#3 Updated by Aurélien PIERRE 9 months ago

  • Status changed from New to Closed: won't fix

#4 Updated by Roman Lebedev 8 months ago

  • Target version set to 2.6.0

#5 Updated by Sebastian Wagner 24 days ago

Hi there,

I've had similar problems and did some profiling which could further describe the problem. I've opened a new, more detailed and focused issue No #12664

Also available in: Atom PDF