On Fri, Jan 13, 2017 at 9:30 AM, Stefan Klinger
<dt-l...@stefan-klinger.de> wrote:
> Hello,


> what graphics card to get for using Darktable with OpenCL under Linux?
> I already have an Asus M5A97 evo R2 mainboard, with an PCIe 2.0 x16
> slot for the graphics card.  I want to stay in the price range below
> 200€.  I do not play games, nor do I use 3D graphics software of any
> kind.  Darktable will be the most demanding task for this piece of
> hardware.

I'm roughly in your same situation, my gaming days are long past
(sigh...) and the GPU is used only for photo edition (and hopefully
some numerical calculus in the future). To give you a reference point,
I run DT on an intel i7-4720HQ notebook with an Nvidia 960M 4GB, under
Ubuntu 16.10. CUDA is fully supported by the proprietary drivers, and
DT uses the GPU without issues.

In my case, the "boat photo" from the Phoronix test drops from 31s
processing time down to 8s going from CPU to GPU (and using less power
going by the coolers noise).  But I think that more important than
export time is the perception of real time edition in the darkroom:
you can leave the computer doing the export in the background or after
hours, but slow edition is what drives one crazy :). Even with this
second tier, previous generation, mobile version GPU, editing the
21Mpx images from a Canon 5D mkII is fast, with barely any lag in the
darkroom (real time modification of parameters, zooming, panning). It
could be faster (thing are not "instantaneous"), but everything runs
smoothly enough. You can check online benchmark comparisons between
your candidate GPU and this one, if its equivalent or better I think
you'll be fine.

My opinion, from what I understand about how CUDA works and what it's
mentioned in DT's manual, is that first and foremost you need a
supported GPU: it doesn't matter which, you'll see a massive
improvement right away (depending on the basis CPU, of course).
Regarding memory: get as much as you can, but there is also a little
quirk related to how the GPU manages it. For some reason even if my
GPU has 4GB it can not assign buffers larger than 1 or 1.5GB at a time
(I don't remember the exact number), resulting in very long export
times if the Mpx count is very large (I guess it's due to the
subdivision of the image during processing). For example, the other
two examples included in the Phoronix test, coming from a much higher
resolution Sony camera, don't show such a big improvement in my GPU.

darktable user mailing list
to unsubscribe send a mail to darktable-user+unsubscr...@lists.darktable.org

Reply via email to