Project

General

Profile

Bug #11528

Overall speed (actually not a bug)

Added by Paolo Astengo over 2 years ago. Updated over 2 years ago.

Status:
Incomplete
Priority:
Low
Assignee:
-
Category:
Darkroom
Target version:
-
Start date:
02/26/2017
Due date:
% Done:

20%

Affected Version:
git master branch
System:
Ubuntu
bitness:
64-bit
hardware architecture:
amd64/x86

Description

I've given a try to rawtherapee even if I'm very happy with Darktable and I've found a terrific difference in speed between twos.

There is anybody experienced the same?

Using the "-d perf" option, I've tried to change all possible settings (OpenCL on/off, different memory settings, etc) without any significant result.
I've tried also to switch to older versions of DT: no changes.

I'm using a XPS9550 with 6700HQ skylake and a Nvidia 960M. Any hint?

History

#1 Updated by Tobias Ellinghaus over 2 years ago

  • % Done changed from 0 to 20
  • Status changed from New to Incomplete

Could you give some concrete numbers? Not everyone here uses RT and could compare.

#2 Updated by Paolo Astengo over 2 years ago

Tobias, it is very difficult to give you some concrete numbers: RT does not have something like "-d perf".
From what I've seen, it seems that RT apply the iop in the order used by the user, not in a fixed order. This causes a smaller lag since when you modify a slider, the sw performs only the last operation. At least this is what I guess..

In any case, I feel my notebook should perform faster than it does. And I'd like to have some performance benchmark against other users/hw.

What do you think about posting a sample picture with a selected set of operations, asking the testers to post their outputs of "-d perf"?

Could this be a good starting point to tune and profile some parts of the code? I'm thinking about another ticked related to OpenGL device priority.

#3 Updated by Tobias Ellinghaus over 2 years ago

RT has a fixed order, just like dt does. Until someone with experience in both programs (or knows how to get timings from RT) comments there is nothing we can do I guess.

#4 Updated by Paolo Astengo over 2 years ago

Thank you Tobias, I've just posted a message on the Facebook groups to find someone interested in test the performances of different hardware.

I hope this can be useful also for the developing team.

Also available in: Atom PDF