Re: [Bf-committers] split cycles cuda kernels from BF_NUMJOBS
I'm not sure how easy it is to force scons building parts single-threaded in-process. BF_NUMJOBS is just a way to be able to set the number of threads to be used using your user-config.py. It does what -j does on command-line. /Nathan On Mon, Dec 5, 2011 at 7:10 PM, Yousef Hurfoush ba...@msn.com wrote: usually each kernel thread needs 1.5 GB RAM, that means in case of 4 threads i need 6GB only for compiling, still there are the OS, etc. so you need at least 8GB of RAM in any machine that want to compile blender with cycles! Per default, we only build 3 kernels though, sm_13, 20 and 21. You can only build the one kernel for your card, that should use less memory. that is not the case if you are building for the masses can i hard code it in SConscript ? Regards Yousef Harfoush ba...@msn.com ___ Bf-committers mailing list Bf-committers@blender.org http://lists.blender.org/mailman/listinfo/bf-committers -- Nathan Letwory Letwory Interactive http://www.letworyinteractive.com ___ Bf-committers mailing list Bf-committers@blender.org http://lists.blender.org/mailman/listinfo/bf-committers
Re: [Bf-committers] split cycles cuda kernels from BF_NUMJOBS
I've committed a trick now so it compiles one kernel at a time, inserting dependencies between the kernels. Brecht. On Tue, Dec 6, 2011 at 9:54 AM, Nathan Letwory nat...@letworyinteractive.com wrote: I'm not sure how easy it is to force scons building parts single-threaded in-process. BF_NUMJOBS is just a way to be able to set the number of threads to be used using your user-config.py. It does what -j does on command-line. /Nathan On Mon, Dec 5, 2011 at 7:10 PM, Yousef Hurfoush ba...@msn.com wrote: usually each kernel thread needs 1.5 GB RAM, that means in case of 4 threads i need 6GB only for compiling, still there are the OS, etc. so you need at least 8GB of RAM in any machine that want to compile blender with cycles! Per default, we only build 3 kernels though, sm_13, 20 and 21. You can only build the one kernel for your card, that should use less memory. that is not the case if you are building for the masses can i hard code it in SConscript ? Regards Yousef Harfoush ba...@msn.com ___ Bf-committers mailing list Bf-committers@blender.org http://lists.blender.org/mailman/listinfo/bf-committers -- Nathan Letwory Letwory Interactive http://www.letworyinteractive.com ___ Bf-committers mailing list Bf-committers@blender.org http://lists.blender.org/mailman/listinfo/bf-committers ___ Bf-committers mailing list Bf-committers@blender.org http://lists.blender.org/mailman/listinfo/bf-committers
Re: [Bf-committers] split cycles cuda kernels from BF_NUMJOBS
That's a nice trick indeed! On Tue, Dec 6, 2011 at 2:31 PM, Brecht Van Lommel brechtvanlom...@pandora.be wrote: I've committed a trick now so it compiles one kernel at a time, inserting dependencies between the kernels. Brecht. On Tue, Dec 6, 2011 at 9:54 AM, Nathan Letwory nat...@letworyinteractive.com wrote: I'm not sure how easy it is to force scons building parts single-threaded in-process. BF_NUMJOBS is just a way to be able to set the number of threads to be used using your user-config.py. It does what -j does on command-line. /Nathan On Mon, Dec 5, 2011 at 7:10 PM, Yousef Hurfoush ba...@msn.com wrote: usually each kernel thread needs 1.5 GB RAM, that means in case of 4 threads i need 6GB only for compiling, still there are the OS, etc. so you need at least 8GB of RAM in any machine that want to compile blender with cycles! Per default, we only build 3 kernels though, sm_13, 20 and 21. You can only build the one kernel for your card, that should use less memory. that is not the case if you are building for the masses can i hard code it in SConscript ? Regards Yousef Harfoush ba...@msn.com ___ Bf-committers mailing list Bf-committers@blender.org http://lists.blender.org/mailman/listinfo/bf-committers -- Nathan Letwory Letwory Interactive http://www.letworyinteractive.com ___ Bf-committers mailing list Bf-committers@blender.org http://lists.blender.org/mailman/listinfo/bf-committers ___ Bf-committers mailing list Bf-committers@blender.org http://lists.blender.org/mailman/listinfo/bf-committers -- Nathan Letwory Letwory Interactive http://www.letworyinteractive.com ___ Bf-committers mailing list Bf-committers@blender.org http://lists.blender.org/mailman/listinfo/bf-committers
Re: [Bf-committers] split cycles cuda kernels from BF_NUMJOBS
Date: Tue, 6 Dec 2011 13:31:14 +0100 From: brechtvanlom...@pandora.be To: bf-committers@blender.org Subject: Re: [Bf-committers] split cycles cuda kernels from BF_NUMJOBS I've committed a trick now so it compiles one kernel at a time, inserting dependencies between the kernels. Brecht. thanks, one last thing, how much often it gets updated to recompile? Regards Yousef Harfoush ba...@msn.com ___ Bf-committers mailing list Bf-committers@blender.org http://lists.blender.org/mailman/listinfo/bf-committers
Re: [Bf-committers] split cycles cuda kernels from BF_NUMJOBS
Subscribe to bf-cvs mailing list and you know :) Am 06.12.2011 16:45, schrieb Yousef Hurfoush: Date: Tue, 6 Dec 2011 13:31:14 +0100 From: brechtvanlom...@pandora.be To: bf-committers@blender.org Subject: Re: [Bf-committers] split cycles cuda kernels from BF_NUMJOBS I've committed a trick now so it compiles one kernel at a time, inserting dependencies between the kernels. Brecht. thanks, one last thing, how much often it gets updated to recompile? Regards Yousef Harfoush ba...@msn.com ___ Bf-committers mailing list Bf-committers@blender.org http://lists.blender.org/mailman/listinfo/bf-committers -- Thomas Dinges Blender Developer, Artist and Musician www.dingto.org ___ Bf-committers mailing list Bf-committers@blender.org http://lists.blender.org/mailman/listinfo/bf-committers
Re: [Bf-committers] split cycles cuda kernels from BF_NUMJOBS
Commited an option in 42465, to not use that trick if enabled, for those who have much RAM and don't want to wait longer than needed. Enable WITH_BF_CYCLES_CUDA_THREADED_COMPILE to still build with several flags. Am 06.12.2011 15:36, schrieb Nathan Letwory: That's a nice trick indeed! On Tue, Dec 6, 2011 at 2:31 PM, Brecht Van Lommel brechtvanlom...@pandora.be wrote: I've committed a trick now so it compiles one kernel at a time, inserting dependencies between the kernels. Brecht. On Tue, Dec 6, 2011 at 9:54 AM, Nathan Letwory nat...@letworyinteractive.com wrote: I'm not sure how easy it is to force scons building parts single-threaded in-process. BF_NUMJOBS is just a way to be able to set the number of threads to be used using your user-config.py. It does what -j does on command-line. /Nathan On Mon, Dec 5, 2011 at 7:10 PM, Yousef Hurfoushba...@msn.com wrote: usually each kernel thread needs 1.5 GB RAM, that means in case of 4 threads i need 6GB only for compiling, still there are the OS, etc. so you need at least 8GB of RAM in any machine that want to compile blender with cycles! Per default, we only build 3 kernels though, sm_13, 20 and 21. You can only build the one kernel for your card, that should use less memory. that is not the case if you are building for the masses can i hard code it in SConscript ? Regards Yousef Harfoush ba...@msn.com ___ Bf-committers mailing list Bf-committers@blender.org http://lists.blender.org/mailman/listinfo/bf-committers -- Nathan Letwory Letwory Interactive http://www.letworyinteractive.com ___ Bf-committers mailing list Bf-committers@blender.org http://lists.blender.org/mailman/listinfo/bf-committers ___ Bf-committers mailing list Bf-committers@blender.org http://lists.blender.org/mailman/listinfo/bf-committers -- Thomas Dinges Blender Developer, Artist and Musician www.dingto.org ___ Bf-committers mailing list Bf-committers@blender.org http://lists.blender.org/mailman/listinfo/bf-committers
Re: [Bf-committers] split cycles cuda kernels from BF_NUMJOBS
Hi, well it's normal that Render Kernel compile takes some resources. Per default, we only build 3 kernels though, sm_13, 20 and 21. You can only build the one kernel for your card, that should use less memory. Regards, Thomas Am 05.12.2011 07:02, schrieb Yousef Hurfoush: hi as currently added support cuda binaries for scons, specifying BF_NUMJOBS=4 will start 4 cuda kernels compile processes and that consumes more than 4GB RAM to do so i think there should be a separate build option for these. i use scons and msvcsp1 on windows 7 x64 i5 4 cores, 4GB DDR3 Regards Yousef Harfoush ba...@msn.com ___ Bf-committers mailing list Bf-committers@blender.org http://lists.blender.org/mailman/listinfo/bf-committers -- Thomas Dinges Blender Developer, Artist and Musician www.dingto.org ___ Bf-committers mailing list Bf-committers@blender.org http://lists.blender.org/mailman/listinfo/bf-committers
Re: [Bf-committers] split cycles cuda kernels from BF_NUMJOBS
usually each kernel thread needs 1.5 GB RAM, that means in case of 4 threads i need 6GB only for compiling, still there are the OS, etc. so you need at least 8GB of RAM in any machine that want to compile blender with cycles! Per default, we only build 3 kernels though, sm_13, 20 and 21. You can only build the one kernel for your card, that should use less memory. that is not the case if you are building for the masses can i hard code it in SConscript ? Regards Yousef Harfoush ba...@msn.com ___ Bf-committers mailing list Bf-committers@blender.org http://lists.blender.org/mailman/listinfo/bf-committers
[Bf-committers] split cycles cuda kernels from BF_NUMJOBS
hi as currently added support cuda binaries for scons, specifying BF_NUMJOBS=4 will start 4 cuda kernels compile processes and that consumes more than 4GB RAM to do so i think there should be a separate build option for these. i use scons and msvcsp1 on windows 7 x64 i5 4 cores, 4GB DDR3 Regards Yousef Harfoush ba...@msn.com ___ Bf-committers mailing list Bf-committers@blender.org http://lists.blender.org/mailman/listinfo/bf-committers