[Bf-cycles] still memory problem with cuLaunch on GTX780Ti 3GB with empty scene
matt.heimlich at gmail.com
Wed Sep 16 05:06:35 CEST 2015
Nope, same thing happens here on Win 7, Lukas.
On Sep 15, 2015 7:05 PM, "Lukas Stockner" <lukas.stockner at freenet.de> wrote:
> I noticed on Mint 17.1 that if you ran out of GPU memory before,
> subsequent renders will sometimes also fail, even if they usually work
> (like an empty scene). After rebooting or restarting the X server, the same
> scene works again.
> I suppose that this is a problem with the memory handling in the
> proprietary Linux kernel, which would mean it's outside of Cycles'
> On Di., Sep. 15, 2015 at 2:56 nachm., Thomas Dinges <blender at dingto.org>
> This is pretty much a known issue, since day 1 of the experimental kernel.
> We added the experimental kernel for features such as SSS, which are the
> cause of the memory peak (or rather SSS enabled as part of our mega kernel).
> The only real solution here will be the CUDA split kernel. That split
> might still happen later this year, it's on our todo list.
> Until then, as Brecht said, use it at your own risk.
> Am 15.09.2015 um 19:54 schrieb Piotr Arlukowicz:
> Thank you Brecht, I understand this and that's OK with memory, but...
> still 1.5GB is not enough even for empty scene... ? Come on! :)
> I was so desperate that I started to look which GTX should I buy. Surely
> 980Ti with 6GiB is a great graphics card, but WILL it work?
> Few months ago Cycles worked in experimental mode just very well and as I
> don't require it to be bulletproof, this memory lag just kicks me out :(
> Sorry for sharing personal mood on the list. I hope somebody can advice
> what to do. Because I don't know if 6GB will be enough, and then there is
> also Titan with 12GB...
> Arlukowicz, BFCT
> *YT: /user/piotao?feature=guide*
> *FB:* */polskikursblendera* *TW:*
> */piotao *
> *Blender Network:* *
> *Polski Kurs Blendera:* <http://polskikursblendera.pl>
> 2015-09-15 18:08 GMT+02:00 Brecht Van Lommel <brechtvanlommel at pandora.be>:
>> Experimental mode uses significantly more memory. I don't know if this
>> is more than expected at the moment, but it wouldn't surprise me.
>> This memory usage is one of the main reasons it's called experimental
>> mode, if all the features there worked well they would not be called
>> experimental. I know a lot of users enable this but it's really at
>> your own risk.
> Bf-cycles mailing list
> Bf-cycles at blender.org
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Bf-cycles