[Sidefx-houdini-list] Houdini, sims & workstation specs

Marten Blumen martyb at gmail.com
Wed Apr 1 14:13:03 EDT 2015


Thanks Jordan.  Ron, sounds right, all motherboards have ram attached to
their CPUs. Afaik Numa comes into play when the 'other' Cpu needs to shift
large amounts of ram to process on certain operations, i.e. such as the
pressure projection solve on high-res sims.  The larger the sim, the more
this should occur. In Houdini I think it's controlled by the Intel's
Threading Building Blocks. Overall, it's not something too much to care
about, but I do :)

Btw it gets very interesting to dig into distributed computing; if you're
curious, check out Lamport timestamps & Vector clocks.. it's kind of
amazing that distributed computing works :)

http://research.microsoft.com/en-us/um/people/lamport/pubs/time-clocks.pdf
http://en.wikipedia.org/wiki/Vector_clock
https://www.threadingbuildingblocks.org

For checking OpenCL ram allocation,  Houdini's About dialog is accurate.

On 2 April 2015 at 03:12, Jordan Halsey <jordanhalsey at gmail.com> wrote:

> Marten,
>
> Will do.
>
> There is a tool in the Nvidia distro's called nvidia-smi
> <https://developer.nvidia.com/nvidia-system-management-interface>.exe that
> will display gpu stats while in process.
>
>
> *Jordan Halsey*
> *maya | houdini | nuke | ae*
> *www.jordanhalsey.com <http://www.jordanhalsey.com>*
> *323-273-9995*
>
> On Wed, Apr 1, 2015 at 1:17 AM, Ron Schab <ron.schab at gmail.com> wrote:
>
> > Hey Marten,
> >
> > thanks for the heads up re DualCPU's.  However if I am not mistaken the
> > Z800's are setup that each CPU has it's own set of memory banks allocated
> > to it ... aka 96gb max per CPU, which isn't that bad still.
> >
> > " The HP Z800 supports a maximum amount of dual in-line memory modules up
> > to 96GB in a single processor configuration and 192GB in a dual processor
> > configuration. " (
> >
> http://www.servermonkey.com/parts-and-upgrades/memory/hp/z800-memory.html)
> >
> >
> > I've also got an 8GB nvidia quadro in my laptop, which I been meaning to
> > test openCL with ... just haven't gotten around to it ... I could run a
> > quick test, but not sure how to check the gpu ram usage on windows.  How
> > would I reliable check how much gRam is used vs it "swapping"?
> >
> > cheers
> > Ron
> >
> >
> >
> > On Wed, Apr 1, 2015 at 5:53 AM, Marten Blumen <martyb at gmail.com> wrote:
> >
> > > Perfect! In 'About Houdini' there is an OpenCL section, look for:
> Device
> > > Address Bits & Global Memory.
> > >
> > > Thanks!
> > >
> > >
> > >
> > > On 1 April 2015 at 16:45, Jordan Halsey <jordanhalsey at gmail.com>
> wrote:
> > >
> > > > I have the titan coming in this week or early next , as well Nvidia
> > > dropped
> > > > off a m6000 the other day which I am looking forward to installing
> and
> > > > testing.
> > > >
> > > > What do you need to do?
> > > >
> > > >
> > > >
> > > >
> > > > *Jordan Halsey*
> > > > *maya | houdini | nuke | ae*
> > > > *www.jordanhalsey.com <http://www.jordanhalsey.com>*
> > > > *323-273-9995*
> > > >
> > > > On Tue, Mar 31, 2015 at 8:20 PM, Marten Blumen <martyb at gmail.com>
> > wrote:
> > > >
> > > > > We need someone to test the Titan X 12Gig; the Nvidia drivers for
> > > OpenCL
> > > > > are 32bit afaik, which means OpenCL is limited to 4GB. AMD drivers
> > > which
> > > > > are 64bit(?) allows 4GB+ for OpenCL
> > > > >
> > > > > On 1 April 2015 at 16:08, Marten Blumen <martyb at gmail.com> wrote:
> > > > >
> > > > > > Numa is non-uniform memory access - each cpu has a bank of memory
> > > > > > physically located next to it, for speed, and Numa is when the
> > other
> > > > cpu
> > > > > > needs to access that bank of Ram. That connection is the Quick
> Path
> > > > > > Interconnect and is not as fast as the connection from CPU to
> it's
> > > > local
> > > > > > ram.
> > > > > >
> > > > > > http://en.wikipedia.org/wiki/Non-uniform_memory_access
> > > > > >
> > > > > > On 1 April 2015 at 15:57, Jordan Halsey <jordanhalsey at gmail.com>
> > > > wrote:
> > > > > >
> > > > > >> Marten,  what is Numa?
> > > > > >>
> > > > > >> Zoran, I have the same machine, but added in a k5200 as well as
> > the
> > > > > k4000
> > > > > >> and am extremely happy with the machine. I have been using
> > redshift
> > > > > render
> > > > > >> which can use both cards and is wicked fast.
> > > > > >>
> > > > > >> The new Titan 12 gig came out which is probably what I would buy
> > at
> > > > this
> > > > > >> point.
> > > > > >>
> > > > > >>
> > > > > >> *Jordan Halsey*
> > > > > >> *maya | houdini | nuke | ae*
> > > > > >> *www.jordanhalsey.com <http://www.jordanhalsey.com>*
> > > > > >> *323-273-9995*
> > > > > >>
> > > > > >> On Tue, Mar 31, 2015 at 2:54 PM, Marten Blumen <
> martyb at gmail.com>
> > > > > wrote:
> > > > > >>
> > > > > >> > All sounds good but beware of Numa with more than one CPU. The
> > qpi
> > > > > hyper
> > > > > >> > transport between multiple CPUs can be a bottle neck compared
> > to a
> > > > > >> single
> > > > > >> > processor, in theory.
> > > > > >> >
> > > > > >> > On 1 April 2015 at 09:44, Zoran Arizanovic <
> > > > > zoran.arizanovic at gmail.com>
> > > > > >> > wrote:
> > > > > >> >
> > > > > >> > > agree completely
> > > > > >> > > i have hpz800 with two xeons , ssd, 72gb ram
> > > > > >> > > the only mistake /problem with it I've put quadro 4000 in
> > > > > >> > > and it overheats a lot
> > > > > >> > >
> > > > > >> > >
> > > > > >> > >
> > > > > >> > > Zoran Arizanovic
> > > > > >> > > LinkedIn <
> > > http://www.linkedin.com/pub/zoran-arizanovic/4/177/295>
> > > > > >> > >
> > > > > >> > >
> > > > > >> > > On Tue, Mar 31, 2015 at 8:40 PM, Ron Schab <
> > ron.schab at gmail.com
> > > >
> > > > > >> wrote:
> > > > > >> > >
> > > > > >> > > > agreed with above.
> > > > > >> > > >
> > > > > >> > > > If you are on a budget I can recommend the HP Z800's. I
> got
> > a
> > > > > barely
> > > > > >> > used
> > > > > >> > > > dual hexacore for 700Euros (12 months warranty included).
> > It
> > > is
> > > > > not
> > > > > >> > the
> > > > > >> > > > newest nor fastest HW u can get these days, but they are
> > still
> > > > > >> really
> > > > > >> > > good
> > > > > >> > > > and take up to 192gb of RAM ... just a thought :)
> > > > > >> > > >
> > > > > >> > > > cheers
> > > > > >> > > > Ron
> > > > > >> > > >
> > > > > >> > > >
> > > > > >> > > > On Tue, Mar 31, 2015 at 8:35 PM, Peter Bowmar <
> > > > pbowmar at gmail.com>
> > > > > >> > wrote:
> > > > > >> > > >
> > > > > >> > > > > Oh, GPU is actually useful for Grain (Sand) simulations,
> > so
> > > if
> > > > > you
> > > > > >> > > think
> > > > > >> > > > > you might do a lot of those, then a better graphics card
> > > will
> > > > > >> > > potentially
> > > > > >> > > > > be useful there. Not as useful for Fluid or Pyro sims
> > > though,
> > > > > they
> > > > > >> > > never
> > > > > >> > > > > fit in the card's RAM except at preview resolutions.
> > > > > >> > > > >
> > > > > >> > > > > On 31 March 2015 at 11:34, Peter Bowmar <
> > pbowmar at gmail.com>
> > > > > >> wrote:
> > > > > >> > > > >
> > > > > >> > > > > > Yup, RAM > CPU > Diskspeed > graphics card. Spend the
> > > least
> > > > > you
> > > > > >> can
> > > > > >> > > on
> > > > > >> > > > > the
> > > > > >> > > > > > graphics card and the most you can on RAM first (64gb
> > the
> > > > > >> minimum)
> > > > > >> > > then
> > > > > >> > > > > > whatever is left on CPU :)
> > > > > >> > > > > >
> > > > > >> > > > >
> > > > > >> > > > >
> > > > > >> > > > >
> > > > > >> > > > > --
> > > > > >> > > > > Cheers,
> > > > > >> > > > >
> > > > > >> > > > > Peter B
> > > > > >> > > > >
> > > > > >> > > > > --Dinner is ruined!
> > > > > >> > > > > _______________________________________________
> > > > > >> > > > > Sidefx-houdini-list mailing list
> > > > > >> > > > > Sidefx-houdini-list at sidefx.com
> > > > > >> > > > >
> > > > > https://lists.sidefx.com:443/mailman/listinfo/sidefx-houdini-list
> > > > > >> > > > >
> > > > > >> > > > _______________________________________________
> > > > > >> > > > Sidefx-houdini-list mailing list
> > > > > >> > > > Sidefx-houdini-list at sidefx.com
> > > > > >> > > >
> > > > https://lists.sidefx.com:443/mailman/listinfo/sidefx-houdini-list
> > > > > >> > > >
> > > > > >> > > _______________________________________________
> > > > > >> > > Sidefx-houdini-list mailing list
> > > > > >> > > Sidefx-houdini-list at sidefx.com
> > > > > >> > >
> > > https://lists.sidefx.com:443/mailman/listinfo/sidefx-houdini-list
> > > > > >> > >
> > > > > >> > _______________________________________________
> > > > > >> > Sidefx-houdini-list mailing list
> > > > > >> > Sidefx-houdini-list at sidefx.com
> > > > > >> >
> > https://lists.sidefx.com:443/mailman/listinfo/sidefx-houdini-list
> > > > > >> >
> > > > > >> _______________________________________________
> > > > > >> Sidefx-houdini-list mailing list
> > > > > >> Sidefx-houdini-list at sidefx.com
> > > > > >>
> https://lists.sidefx.com:443/mailman/listinfo/sidefx-houdini-list
> > > > > >>
> > > > > >
> > > > > >
> > > > > _______________________________________________
> > > > > Sidefx-houdini-list mailing list
> > > > > Sidefx-houdini-list at sidefx.com
> > > > > https://lists.sidefx.com:443/mailman/listinfo/sidefx-houdini-list
> > > > >
> > > > _______________________________________________
> > > > Sidefx-houdini-list mailing list
> > > > Sidefx-houdini-list at sidefx.com
> > > > https://lists.sidefx.com:443/mailman/listinfo/sidefx-houdini-list
> > > >
> > > _______________________________________________
> > > Sidefx-houdini-list mailing list
> > > Sidefx-houdini-list at sidefx.com
> > > https://lists.sidefx.com:443/mailman/listinfo/sidefx-houdini-list
> > >
> > _______________________________________________
> > Sidefx-houdini-list mailing list
> > Sidefx-houdini-list at sidefx.com
> > https://lists.sidefx.com:443/mailman/listinfo/sidefx-houdini-list
> >
> _______________________________________________
> Sidefx-houdini-list mailing list
> Sidefx-houdini-list at sidefx.com
> https://lists.sidefx.com:443/mailman/listinfo/sidefx-houdini-list
>



More information about the Sidefx-houdini-list mailing list