Thanks for the explanation! I've been using an old single core computer
which has been just usable with a fast gpu, thanks to opencl, but have just
built a faster four core cpu system and I'm not sure if I'd get much
benefit from adding that nvidia 8800 - main gpu is much faster (gtx 560ti)
and the new cpu seems well able to do the previews and it all runs quite
smooth. I guess I'll just have to try and see =) One further question
though: with darktable -d opencl, I see that the preview is being processed
after each change in darkroom mode even when the left panel is hidden,
isn't that unnecessary? Perhaps that could be skipped to optimize speed?


On Tue, Apr 23, 2013 at 9:23 PM, Ulrich Pegelow <[email protected]
> wrote:

> Am 23.04.2013 09:19, schrieb Janne Kurri:
>
>  Hello all, a few questions:
>>
>
> Ah, then you're probably the only one besides of me who uses a multi-GPU
> setup with darktable :) At least you're the first one with a question.
>
>
>
>> -I have another gpu but it has less memory than the one I currently use,
>> if I install it as secondary gpu can I configure different amounts of
>> headroom for each? Presumably the secondary gpu - which isn't outputting
>> to any displays - isn't already using much if any of it's memory.
>>
> No, currently not possible. You only can set one value for most of the
> OpenCL config parameters. These should be the lowest common denominators.
>
>
>  -I can prioritize each gpu/cpu to process pixelpipes "center image",
>> "full view", "export" and "preview", but it's not clear to me what's
>> what, especially preview and center image?
>>
> "center image" = "full view" = the big zoomable image in the center of
> darkroom mode
>
> "preview image" = the small not zoomable view of the full image in the
> upper left navigation panel in darktable mode
>
> "export image" = quite obviously the images that gets exported by darktable
>
> "thumbnail image" = the small images that darktable generates as
> thumbnails so they can be shown in lighttable mode
>
>
>  -will multiple all processing devices get to work during export if I set
>> parallel >1 in core settings?
>>
> In theory yes. However, this option has been removed in current git
> master. Too many users have shot themselves into the foot, mainly due to
> the high memory demand. In fact multi-GPU setups are probably the only ones
> where it would make sense to have more than one export thread. Still you
> need big amounts of core memory as in-between darktable needs to copy image
> buffers between device and host.
>
> Ulrich
>
>
>
------------------------------------------------------------------------------
Try New Relic Now & We'll Send You this Cool Shirt
New Relic is the only SaaS-based application performance monitoring service 
that delivers powerful full stack analytics. Optimize and monitor your
browser, app, & servers with just a few lines of code. Try New Relic
and get this awesome Nerd Life shirt! http://p.sf.net/sfu/newrelic_d2d_apr
_______________________________________________
Darktable-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/darktable-users

Reply via email to