It depends on how memory-bound the calculation is. If it requires a lot of memory access, increasing the number of CPUs beyond a the level where the memory bus is saturated is not going to make a lot of difference. This system, I'm assuming, has a single memory for all CPUs. That kind of a system requires the CPU caches to be kept full all the time. As soon as you hit a page fault (or the OS switches processes) all eight CPUs will have to dump their local caches at which point the performance goes down the toilet. Most super-computers have a large local memory for each CPU to avoid precisely this problem.
There are ways of partitioning this particle fluid problem so that it scales better over a number of CPUs but it's certainly not linear and I'm sure the RealFlow guys have gone there already.
- ½
> -----Original Message-----
> From: owner-xsi(at)Softimage.COM
> [mailto:owner-xsi(at)Softimage.COM]On Behalf
> Of Eric Lampi
> Sent: Tuesday, 25 April, 2006 17:08
> To: XSI(at)Softimage.COM
> Subject: Re: network distributed particle simulations?
>
>
> Yeah but doesn't it make a huge difference that the machine
> has 8 CPUs as opposed to your standard 2? I just never
> worked on a PC with more than 2 cpus, so I am not sure if
> there's other issues involved.
>
> I wonder if you would eventually be be able to "daisy-chain"
> multiple physics engine cards in one box.
>
> E
>
> Freelance 3-D Animator, F/X Artist
>
> ----- Original Message ----
> From: kim aldis <kim(at)cg-soup.com>
> To: XSI(at)Softimage.COM
> Sent: Tuesday, April 25, 2006 4:44:05 PM
> Subject: RE: network distributed particle simulations?
>
>
>
> > -----Original Message-----
> > From: owner-xsi(at)Softimage.COM
> > [mailto:owner-xsi(at)Softimage.COM] On Behalf Of Eric Lampi
> > Sent: 25-April-2006 20:24
> > To: XSI(at)Softimage.COM
> > Subject: Re: network distributed particle simulations?
> >>
> > I bet Real Flow would actually be managable with something
> like this.
> >
>
> I'm not so sure. They've been working with multiple
> processors and I know
> they thought about networked rendering but fluid systems need
> to know a lot
> about the particles around them. The issue of network
> bandwidth isn't a
> small one and it doesn't go away with clever software.
>
> ---
> Unsubscribe? Mail Majordomo(at)Softimage.COM with the following
> text in body:
> unsubscribe xsi
>
>
>
> ---
> Unsubscribe? Mail Majordomo(at)Softimage.COM with the following
> text in body:
> unsubscribe xsi
>
---
Unsubscribe? Mail Majordomo(at)Softimage.COM with the following text in body:
unsubscribe xsi
