Re: which hardware for numerical analysis < USD 8000 ?
From: David Wilkinson (david_at_wilkinson6337.freeserve.co.uk)
Date: 02/13/05
- Next message: Roman Werpachowski: "Re: which hardware for numerical analysis < USD 8000 ?"
- Previous message: Hiu Chung Law: "Smallest positive eigenvalue of a singular matrix"
- In reply to: bv: "Re: which hardware for numerical analysis < USD 8000 ?"
- Next in thread: Roman Werpachowski: "Re: which hardware for numerical analysis < USD 8000 ?"
- Reply: Roman Werpachowski: "Re: which hardware for numerical analysis < USD 8000 ?"
- Reply: A.L.: "Re: which hardware for numerical analysis < USD 8000 ?"
- Messages sorted by: [ date ] [ thread ]
Date: Sun, 13 Feb 2005 08:27:24 +0000
bv wrote:
> Pavel Pokorny wrote:
>
>>can you tell me your suggestion for a workstation
>>for less than some $ 8000 appropriate for numerical
>>solution of differential equation (no really huge systems)?
>
>
> The days when WS performance was towering over the PC are long gone.
> It's more of a question how soon will these overpriced machines
> disappear altogether. For many, solving DE systems even on, long
> scrapped, 386s was more than enough - nowadays any mainstream PC, for <
> $1000, will suit your needs just fine.
>
You may be right for DEs but not for PDEs. Trying to solve 3-dimensional
PDEs for Computational Fluid Dynamics calculations shows just how
painfully slow and limited the average PC really is. My 1.33GHz AMD
CPU'd machine is a bit past it now but current PC chips are only up to
just over 3 GHz. This is about 2.3 times as fast, which sounds good
until you realise that computing time goes up proportional to number of
points cubed. If I increase the number of calculating points, or
singularity panels in my case, by a factor (2.3)^(1/3) = 1.32 that uses
up all the speed increase. Since these panels are a surface array this
means factoring the number in each of the two directions by sqrt(1.32) =
1.15. So, I could have 15% more points in each direction, hardly worth
the trouble and expense!
However I see there is a new "Cell" chip which is about 20 times as fast
which might at last put some real computing power into the PC, if there
are any left over from games machines!
- Next message: Roman Werpachowski: "Re: which hardware for numerical analysis < USD 8000 ?"
- Previous message: Hiu Chung Law: "Smallest positive eigenvalue of a singular matrix"
- In reply to: bv: "Re: which hardware for numerical analysis < USD 8000 ?"
- Next in thread: Roman Werpachowski: "Re: which hardware for numerical analysis < USD 8000 ?"
- Reply: Roman Werpachowski: "Re: which hardware for numerical analysis < USD 8000 ?"
- Reply: A.L.: "Re: which hardware for numerical analysis < USD 8000 ?"
- Messages sorted by: [ date ] [ thread ]
Relevant Pages
|