darktable speed (in general, and when using two monitors)

Dear Christian @pk5dark,

Actually I have both.
darktable 2.6.2 is installed via “emerge darktable” with the following USE-flags
media-gfx/darktable::gentoo opencl -colord geo libsecret webp linguas_de L10N_de

Also at the moment I am on git-master, which I install with this simple script:

#!/bin/sh

cd /root/darktable
git pull
git submodule update
./build.sh
cd build
make install

exit 0

If I want to keep that script, seems I have to learn some awk and sed to implement the changes you recommend :slight_smile:

Let me try that now manually first.

Cheers
Axel

@AxelG

That looks like the “normal” darktable git command commands.
See at the end of this page for additional info:

MfG
Claes in Lund, Schweden

1 Like

@Claes
cool, I just implement that bid in my script and all is fine for updates (I know, also the “make” it self):

### Building manually
 
$ mkdir $HOME/darktable/build 
$ cd $HOME/darktable/build 
$ cmake -DCMAKE_BUILD_TYPE=Release ..

There is no difference in speed for me between the build types. :innocent:

@pk5dark
for me an itsy bit

but not really for opencl activated

so, at the end not worthwhile to bother changing the script :slight_smile:

Did you notice the build logs of the opencl kernels if you start darktable -d opencl? They are compiled at runtime.

1 Like

@pk5dark
I’m lost :wink:
I just wanted to confirm, that --build-type=Release also didn’t do that much in terms of speed, when opencl is working

New clockings performed using darktable with/without openCL.
CPU = Ryzen 7 2700X; GFX GTX-1660Ti from base setting to overclocking CPU and RAM.

openCL
From these data I deduce that it is not so much to gain by overclocking the CPU (like going from 3.7 GHz to 4.0 GHz, which is a mild OC).

It was interesting to note that doubling the amount of RAM (from 8 GB to 16 GB) yields immediate benefits, with as well as without openCL.

Now, I wonder what would happen if I insert an additional 16GB of RAM.
Hm…

Have fun!
Claes in Lund, Sweden

Not apples to apples, @Claes, but my main editing rig has 32gb ram and Intel graphics and my play machine had 16gb ram and a gtx 970. The play machine is much faster with opencl. Its not even close.

Same here.
My first results were 21 seconds both from 32GB RAM machine with CPU i7-6700 and SSD.
For my 16GB i5-9600 with NVMe drive and Radeon RX480 8GB the results are:
OpenCL: 6,338431
CPU: 19,808357 (pixel pipeline processing took 19,445 secs (113,139 CPU))

If you do not run 2 or more VMs or several containers than 16 GB of RAM is a lot :slight_smile:

Are you using 8GB modules and going from single to dual channel setup?

1 Like

Moin, Christian!

You have a very valid point there! Thank you.
Normally, I have two 8 GB sticks in dual channel mode.
For the test (above) I just extracted one stick and
rebooted…

Would you have any learned guess as to what would
happen to dt speeds w/wo openCL if I inserted two more
8 GB sticks, i.e. in total 4 * 8 GB?
4 * 8 GB

MfG
Claes in Lund, Schweden

Why it should make a difference as long as there is no swapping on disk. If it would run out of memory on export, I guess it will get dam slow.

An i9-9900k and 4x16 GB 2666 are waiting for assembly her :grin:. You already tested single vs. dual channel memory setup. So there is nothing to add :thinking:

I will provide some benchmark figures once I’ve found time for assembly and moving the data …

1 Like

@pk5dark
Cannot wait to see your results, as I also have a core i9 9900k with NVMe and re-used my 32GB DDR4 2133. See my bench above…

Cheers
Axel

An i9-9900k and 4x16 GB 2666 are waiting for assembly her :grin:.

No fair!
Now I have not just a gear envy of the glass of others but the processing hardware of others.
:face_with_symbols_over_mouth:

1 Like

@martin.scharnke Allow me to paraphrase your famous countryman:
“That is not a computer…”
https://www.youtube.com/watch?v=POJtaO2xB_o
I mean there are also Threadrippers.

2 Likes

@Claes, you don’t have to rub it in! :wink:

[Then again, I’m a mere German-descended New Zealand born resident of Australia. Thus Mick Dundee is not truly my countryman!] :grinning:

So here are the new numbers for i9-9900k (4.8 GHz) with 64 GB 2666MHz:

  • OpenCl: 4.067 (was 6.6 s with the same GPU on i7-2600)
  • CPU: 14.354 s (was 32.8 s on i7-2600)

slight OC with 5.0 GHz:

  • CPU:13.636 s

Moinchen!

Interesting. Thank you.
If you have the time & feel like experimenting:
What would happen to the execution times
if you removed two of your memory sticks,
i.e. having “just” 32 gig inside?

Are you running them in XMP mode?

Have fun!
Claes in Lund, Sweden

Intersting, without OC your CPU seems 7% faster than mine… Just due to faster ram than mine? I do have 2133s (but latency of 13)

Which MoBo and which GPU do you have?

Cheers
Axel