[Bf-cycles] Bf-cycles Digest, Vol 75, Issue 4 - Size of tiles for hybrid system CPU+CUDA/OpenCL/Phi (Jaros Milan)
Marc Binsted
marc at binsted.plus.com
Mon Jul 24 13:43:15 CEST 2017
Hi Milan,
I've been quiet for a long time with regards this subject. However,
Jaros, I noticed your post this morning and I recall a post from Matheus
Mortatti a few months ago with regards distributed OpenMP and it
intrigued me. I am also working on similar projects to both Matheus and
his colleagues and also with your work regarding a distributed client
for a very different purpose.
I am working in the UK, developing accelerated simulation and offline
multi-renderer solution for film and entertainment with the advantage of
an HPC environment. A multi GPU prototype is complete and we have a
large production system under construction. I have resolved some
interesting issues over memory management, in-core and out-of-core for
various GPU architectures, utilising RDMA over infiniband and ethernet.
There are many optimisations including automatic tile size and process
optimisation for efficient use of computing resources. This is where my
work crosses into Brecht's and Sergey's work with regards Cycles, BVH
sorting and kernel sizes and split kernel efficiencies.
Kind regards,
Marc
Marc Binsted
marc at binsted.plus.com
marc.binsted at theory.digital
On 24/07/2017 11:00, bf-cycles-request at blender.org wrote:
> Send Bf-cycles mailing list submissions to
> bf-cycles at blender.org
>
> To subscribe or unsubscribe via the World Wide Web, visit
> https://lists.blender.org/mailman/listinfo/bf-cycles
> or, via email, send a message with subject or body 'help' to
> bf-cycles-request at blender.org
>
> You can reach the person managing the list at
> bf-cycles-owner at blender.org
>
> When replying, please edit your Subject line so it is more specific
> than "Re: Contents of Bf-cycles digest..."
>
>
> Today's Topics:
>
> 1. Req: Access to new branch for distributed rendering using MPI
> and OpenMP (new Blender client not only for Cycles) (Jaros Milan)
> 2. Req: Access to new branch for rendering on Intel Xeon Phi
> (Jaros Milan)
> 3. Re: Req: Access to new branch for distributed rendering using
> MPI and OpenMP (new Blender client not only for Cycles)
> (Sergey Sharybin)
> 4. Re: Req: Access to new branch for rendering on Intel Xeon Phi
> (Sergey Sharybin)
> 5. Size of tiles for hybrid system CPU+CUDA/OpenCL/Phi (Jaros Milan)
> 6. Re: Size of tiles for hybrid system CPU+CUDA/OpenCL/Phi
> (Sergey Sharybin)
> 7. Re: Size of tiles for hybrid system CPU+CUDA/OpenCL/Phi
> (Brecht Van Lommel)
>
>
> ----------------------------------------------------------------------
>
> Message: 1
> Date: Sun, 23 Jul 2017 13:07:02 +0000
> From: Jaros Milan <milan.jaros at vsb.cz>
> Subject: [Bf-cycles] Req: Access to new branch for distributed
> rendering using MPI and OpenMP (new Blender client not only for
> Cycles)
> To: "bf-cycles at blender.org" <bf-cycles at blender.org>
> Message-ID: <d252029a-4837-4f98-9744-4ab98c802778 at email.android.com>
> Content-Type: text/plain; charset="utf-8"
>
> Hi,
>
> I would like to create a new Blender client not only for distributed rendering ( offline + interactive ).
>
> I used my version of client for rendering some scenes of Agent327.
>
> You can find more information here: http://blender.it4i.cz/research/distributed-rendering/
>
> It could use OpenMPI or Intel MPI ( for using more PCs with Intel Xeon Phis/GPU ).
>
> Best regards
>
> Milan Jaro?
> Research Assistant
>
> IT4Innovations national supercomputing center
> V?B ? Technical University of Ostrava
>
> 17. listopadu 15/2172 | 708 33 Ostrava-Poruba | Czech Republic
>
>
>
> e-mail: milan.jaros at vsb.cz | web: industry.it4i.cz | phone: +420 597 329 583 | map: goo.gl/maps/BpF0z
>
> -------------- next part --------------
> An HTML attachment was scrubbed...
> URL: http://lists.blender.org/pipermail/bf-cycles/attachments/20170723/b49cc85a/attachment-0001.htm
>
> ------------------------------
>
> Message: 2
> Date: Sun, 23 Jul 2017 13:32:49 +0000
> From: Jaros Milan <milan.jaros at vsb.cz>
> Subject: [Bf-cycles] Req: Access to new branch for rendering on Intel
> Xeon Phi
> To: "bf-cycles at blender.org" <bf-cycles at blender.org>
> Message-ID: <1e12a4c6-bc62-483f-b30f-4f343612bd7b at email.android.com>
> Content-Type: text/plain; charset="utf-8"
>
> Hi,
>
> I would like to create a new computed devices for Intel Xeon Phi (KNC, KNL).
>
> I used my version of new computed devices for rendering of almost all scenes of Agent327 (http://blender.it4i.cz/rendering/agent-327-operation-barbershop/).
>
> You can find more information here: http://blender.it4i.cz/research/rendering-on-intel-xeon-phi/
>
> Now I use Intel compiler but I would like to change it to ISPC open sorce compiler.
>
> I would like to open question about size of tiles for hybrid system CPU+CUDA/Phi ( CUDA, OpenCL, OpenMP require bigger tile than CPU ). Sergey has the some idea about this.
>
> Thanks.
>
> Best regards
>
> Milan Jaro?
> Research Assistant
>
> IT4Innovations national supercomputing center
> V?B ? Technical University of Ostrava
>
> 17. listopadu 15/2172 | 708 33 Ostrava-Poruba | Czech Republic
>
>
>
> e-mail: milan.jaros at vsb.cz | web: industry.it4i.cz | phone: +420 597 329 583 | map: goo.gl/maps/BpF0z
>
> -------------- next part --------------
> An HTML attachment was scrubbed...
> URL: http://lists.blender.org/pipermail/bf-cycles/attachments/20170723/6cd97773/attachment-0001.htm
>
> ------------------------------
>
> Message: 3
> Date: Sun, 23 Jul 2017 15:54:47 +0200
> From: Sergey Sharybin <sergey.vfx at gmail.com>
> Subject: Re: [Bf-cycles] Req: Access to new branch for distributed
> rendering using MPI and OpenMP (new Blender client not only for
> Cycles)
> To: Discussion list to assist Cycles render engine developers
> <bf-cycles at blender.org>
> Message-ID:
> <CAErtv27FnnWbYe5SDrp334aVkvOPzEBqPMfActvH0znbq3_A0g at mail.gmail.com>
> Content-Type: text/plain; charset="utf-8"
>
> Hi,
>
> While i can see how this is benefitial for technical/farm usage, there are
> some concerns related on usefulness for Blender users:
>
> - How difficult it is to set up such an OpenMP farm ?
> - How much portable it is (Windows, OSX, Linux, BSD) ?
> - What are the supported devices (my guess here is that there will be no
> GPU support, as in CUDA and OpenCL) ?
>
> There are also some technical questions:
>
> - Why is OpenMP better for final render? I would assume that could cause
> quite huge networking overhead.
> - Why do we need some distributed device support which is rather limited to
> what OpenMP is offering instead of improving existing device_network.cc
> (that oen would at least support all GPUs, will allow to do some data cache
> etc..) ?
>
> Those are the main concerns of having this device exposed to Blender users.
> On another hand, if all the changes are living in device_openmp.cc and
> kernel/kernels/openmp with minimal code overlap, i done feel against having
> devices which are used by tehcnical guys and not exposed to Blender users.
>
> That's y current questions and thoughts here anyway.
>
>
> On Sun, Jul 23, 2017 at 3:07 PM, Jaros Milan <milan.jaros at vsb.cz> wrote:
>
>> Hi,
>>
>> I would like to create a new Blender client not only for distributed
>> rendering ( offline + interactive ).
>>
>> I used my version of client for rendering some scenes of Agent327.
>>
>> You can find more information here: http://blender.it4i.cz/
>> research/distributed-rendering/
>>
>> It could use OpenMPI or Intel MPI ( for using more PCs with Intel Xeon
>> Phis/GPU ).
>>
>> Best regards
>>
>> Milan Jaro?
>> Research Assistant
>>
>> IT4Innovations national supercomputing center
>> V?B ? Technical University of Ostrava
>>
>> 17. listopadu 15/2172 | 708 33 Ostrava-Poruba | Czech Republic
>>
>>
>>
>> e-mail: milan.jaros at vsb.cz | web: industry.it4i.cz | phone:
>> +420 597 329 583 <+420%20597%20329%20583> | map: goo.gl/maps/BpF0z
>>
>>
>> _______________________________________________
>> Bf-cycles mailing list
>> Bf-cycles at blender.org
>> https://lists.blender.org/mailman/listinfo/bf-cycles
>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.blender.org/pipermail/bf-cycles/attachments/20170724/37fd4b24/attachment.htm
More information about the Bf-cycles
mailing list