Hi Jim
I just found time to run the mentioned benchmark on darktable v3.2.1 on
macOS 10.15.6 MacBook Pro (Touchbar, 13-inch, 2018, 4xTB3), 2.7 GHz Intel
Core i7, 16GB
CPU only: 25.2secs (significantly faster than v3.0.0 on the same machine
congrats and thanks to the coders!)
with Intel Iris Graphics 640 1536 MB: 18.2secs (same as 3.0)
with eGPU AMD Radeon Pro 580 Compute Engine 8192 MB: 5.2secs (same as 3.0)
so you should see much faster results on your macbook pro with dedicated
gpu vs CPU, since my eGPU is already dated and gives 3-5x faster results...
here's how I ran the bench
/Applications/darktable.app/Contents/MacOS/darktable-cli bench.SRW test.jpg
--core -d perf -d opencl
As Remco Vietor already said, make sure all opencl kernels are compiled
(needs many restarts), here's a dumb script to do this for you:
#!/bin/bash
DT="/Applications/darktable.app/Contents/MacOS/darktable -d opencl"
# first remove cached kernels
rm -r ~/.cache/darktable/cached_kernels_for_*
# re-run darktable until kernels are compiled
while true
do
$DT > /tmp/dt_opencl_compile.out &
DT_PID=$!
sleep 6
if grep --quiet "successfully loaded kernel"
/tmp/dt_opencl_compile.out
then
break
else
echo "opencl not successfully compiled, killing."
kill $DT_PID
fi
done
Am Fr., 14. Aug. 2020 um 13:39 Uhr schrieb Giuseppe Pagnoni <
gpagn...@gmail.com>:
> Thank you very much Jim, your suggestion was very helpful. The
> figures without and with opencl are 17.470 and 15.871 sec,
> respectively, corresponding to an opencl-related improvement in
> performance of around 9%. I have to say I am a bit underwhelmed by
> the benefits of opencl (or by the graphic card in the latest MacBook
> Pro, if that's the culprit).
>
> cheers
> giuseppe
>
> On Fri, Aug 14, 2020 at 1:24 PM Jim Robinson wrote:
> >
> > Giuseppe,
> >
> > I don't have a Mac myself so now idea how good the GPU is. How does the
> 15sec compare with the time taken with opencl disabled
> > (--disable-opencl)? With a high end GPU with lots of memory factors of
> 10 times faster seem to have been achieved though factors somewhat below 10
> (e.g. 4 or 5) seem more typical -see the examples in the link I originally
> included.
> >
> > Jim
> >
> > Jim
> >
> > On Fri, 14 Aug 2020 at 11:25, Giuseppe Pagnoni
> wrote:
> >>
> >> Hello Jim,
> >>
> >> thanks for the suggestion! I tried the following:
> >>
> >> CPU + GPU, using OpenCL:
> >>darktable-cli bench.SRW bench.SRW.xmp test.jpg --core -d perf -d
> opencl
> >>
> >> all the events were successful and the total time was 15 sec. Is that
> >> a reasonable value?
> >>
> >> giuseppe
> >>
> >> On Fri, Aug 14, 2020 at 11:13 AM Jim Robinson
> wrote:
> >> >
> >> > Giuseppe,
> >> >
> >> > You could try the benchmarks described here.
> >> > https://math.dartmouth.edu/~sarunas/darktable_bench.html
> >> >
> >> > A bit old now, but should still work.
> >> >
> >> >
> >> >
> >> > On Fri, 14 Aug 2020 at 10:03, Giuseppe Pagnoni
> wrote:
> >> >>
> >> >> Hi,
> >> >>
> >> >> I have just installed the latest 3.2.1 version on a 13" MacBook Pro
> >> >> (latest model) with an Intel Iris Plus Graphics 1536 GPU. I don't
> >> >> really see much of an increase in speed when I activate the OpenCL
> >> >> support. Is this the way it is supposed to be? Could you suggest a
> >> >> way to test it more rigorously?
> >> >>
> >> >> thank you in advance for any comments
> >> >>
> >> >> giuseppe
> >> >>
> >> >> --
> >> >> Giuseppe Pagnoni
> >> >> Dip. Scienze Biomediche, Metaboliche e Neuroscienze
> >> >> Sezione Fisiologia e Neuroscienze
> >> >> Univ. di Modena e Reggio Emilia
> >> >> Via Campi 287
> >> >> I-41125 Modena, Italy
> >> >> Tel: +39-059-205-5742
> >> >> Fax: +39-059-205-5363
> >> >>
>
> >> >> darktable user mailing list
> >> >> to unsubscribe send a mail to
> darktable-user+unsubscr...@lists.darktable.org
> >> >>
> >> >
> >> >
>
> darktable user mailing list to unsubscribe send a mail to
> darktable-user+unsubscr...@lists.darktable.org
> >>
> >>
> >>
> >> --
> >> Giuseppe Pagnoni
> >> Dip. Scienze Biomediche, Metaboliche e Neuroscienze
> >> Sezione Fisiologia e Neuroscienze
> >> Univ. di Modena e Reggio Emilia
> >> Via Campi 287
> >> I-41125 Modena, Italy
> >> Tel: +39-059-205-5742
> >> Fax: +39-059-205-5363
>
>
>
> --
> Giuseppe Pagnoni
> Dip. Scienze Biomediche, Metaboliche e Neuroscienze
> Sezione Fisiologia e Neuroscienze
> Univ. di Modena e Reggio Emilia
> Via Campi 287
> I-41125 Modena, Italy
> Tel: +39-059-205-5742
> Fax: +39-059-205-5363
>
>
> darktable user mailing list
> to unsubscribe send a mail to
> darktable-user+unsubscr...@lists.darktable.org
>
>
__
