[Opensim-dev] Feature Proposal: BulletSim GPU Acceleration via OpenCL
SignpostMarv Martin
opensim at signpostmarv.name
Tue Sep 10 18:23:36 UTC 2013
I would hazard a guess that DSG would benefit from OpenCL-enabled
physics, as physics chunks with higher requirements could be sent over
to a server with appropriate hardware, while the less taxing chunks
could be sent to "standard" servers.
~ Marv.
On 10/09/2013 19:02, Sean McNamara wrote:
> On 9/10/13, R.Gunther <rigun at rigutech.nl> wrote:
>> A quick reply, i see one problem. most of the servers (inlcudeing mine)
>> dont have realy openCL support.
>> So not sure what benneifit you want to get from openCL if the GPU dont
>> support it.
> This may be true, but it is changing. Old CPUs (Core 2 era, Nehalem,
> etc.) didn't have an on-board GPU, but if you use a modern CPU,
> chances are fairly good it'll have a GPU. The number of GPUs in
> datacenters is going up very rapidly, especially since the Ivy Bridge
> generation of E3 Xeons introduced on-board GPUs on several models.
>
> Also, many datacenters are switching over to desktop-grade hardware to
> reduce costs. The perception that "Xeon is the only way if you are
> serious about your data" is evaporating, partly because server-grade
> hardware is imperfect and still crash occasionally, and partly because
> the quality of desktop-grade hardware is actually very high. The
> advantages that Xeon used to have, like huge L3 cache, are becoming
> less relevant, especially in comparison to the added cost of purchase
> (and thus the cost of hosting plans / rentals involving these chips).
>
> I agree that Linux based servers with only an Intel IGPU will have a
> hard time for a little while until the Intel GPU OpenCL support is
> fully stabilized and optimized, but Intel and other community members
> over at the mesa project are working on it, and you can see the
> enormous activity in this space just by reading the commit logs over
> at the Mesa project: http://cgit.freedesktop.org/mesa/mesa
>
> I've updated the wiki documenting some of the advantages and
> disadvantages that we have discussed.
>
> Thanks,
>
> Sean
>
>> On 2013-09-10 19:30, Sean McNamara wrote:
>>> Hi, all.
>>>
>>> I think my proposal mostly speaks for itself, so I won't belabor the
>>> point in this email. Suffice it to say, if you are interested in
>>> performance of OpenSimulator, and in particular physical objects
>>> performance, please check out my feature proposal here and provide
>>> feedback, either by editing the page directly or replying to this
>>> email.
>>>
>>> http://opensimulator.org/wiki/Feature_Proposals/BulletSim_OpenCL
>>>
>>> Thanks!
>>>
>>> Sean McNamara
>>> _______________________________________________
>>> Opensim-dev mailing list
>>> Opensim-dev at lists.berlios.de
>>> https://lists.berlios.de/mailman/listinfo/opensim-dev
>>>
>> _______________________________________________
>> Opensim-dev mailing list
>> Opensim-dev at lists.berlios.de
>> https://lists.berlios.de/mailman/listinfo/opensim-dev
>>
> _______________________________________________
> Opensim-dev mailing list
> Opensim-dev at lists.berlios.de
> https://lists.berlios.de/mailman/listinfo/opensim-dev
>
More information about the Opensim-dev
mailing list