CFD Online Logo CFD Online URL
www.cfd-online.com
[Sponsors]
Home > Forums > Main CFD Forum

What do you think?

Register Blogs Members List Search Today's Posts Mark Forums Read

Reply
 
LinkBack Thread Tools Display Modes
Old   April 1, 2005, 09:50
Default What do you think?
  #1
Ford Prefect
Guest
 
Posts: n/a
Top hardware is a requirement for the CFD modeller, on that I think we all agree. But since top hardware ususlly comes with a top price, a bunch of us has to settle with less than the best. (www.top500.org)

So when we can't have the fastest computer on earth (or more) we redefine the problem. Some spend hours and hours going over algorithms to get a 5% boost in performance. 5% do make a difference so there's nothing wrong with that, but is it worth the time invested (this is something you only can be sure of when you have completed the optimization)??.

My simple solution to small budget and/or time is OverClocking.

-------------

Example *):

I want an AMD64 2.4GHz system to do some simple simulations on. Instead of buying an FX-53 or 3800+ I just invest in a simple 3200+. Needless to say the price of the AMD64 3200+ is way less than that of the 3800+.

I buy some RAM that reacts well to overclocking and proceed to increase the frequency of the system. A meager 400 MHz is easily obtained resulting in a 2.4GHz processor, equal to the much more expensive 3800+ processor **). I run some test program that stresses the components in the computer (e.g. prime95, memtest86, Sandra-burn-in) for say 48 hours straight and conclude that it is stable 24/7. I do the overclock in say 5-10 minutes and the performance boost is roughly 20%!

Drawbacks:

1.You might decrease the life-expectancy of the system. This is really Not a problem since it will become obsolete before this happens.

2. Basic knowledge is needed to prevent the system from serious harm. This is a standard drawback of all electronic equipment handling.

So why not try to overclock and feel the difference?

*) I'm not advertising AMD here, Intel works just as well.

**)There are more positive effects to this overclocking since the memory is overclocked as well, the computer is actually a lot faster than a regular 3800+ system with the same memory.

PS The example computer runs superPI 1million digits in 33 secs - total cost - less than 1000 euro.
  Reply With Quote

Old   April 1, 2005, 10:40
Default Re: What do you think?
  #2
MG
Guest
 
Posts: n/a
Although this above original post may look like a spam, but it raises a very fundamental question. Which way to go? I.e., what is the solution: ever increasing productivity of computers or smarter algorithms? Simple answer would be "why not both?". Unfortunately, how many smart algorithms have you seen in the past decade or so? It seems that all progress comes from the muscle rather than brains. The computational horsepower is just a part of the solution, as it keeps doubling every 18 (?) months. Meanwhile, the use of smarter algorithms has the potential of reducing the computer requirements by several orders of magnitude, not just percents.

It's not that I'm trying to convince anyone to look for possible alternatives, but conveying the message there might be some other ways of doing CFD (not just CFD -- computational mechanics).
  Reply With Quote

Old   April 1, 2005, 15:04
Default Re: What do you think?
  #3
Andrew
Guest
 
Posts: n/a
AMD64s beat equivalent P4s hands down when it comes to CFD work. I run LES simulations on an Athlon 64 3400+, and it can run stably for over 2 months at 100% load.

No-one in the CFD world needs as much computing power as those machines in the top 500 can offer, unless they want to do a DNS of an extremely high Re flow with a complex geometry. Modellers spend too long thinking about how to make their numerics extremely precise, when CFD should really be about producing good results as cheaply as possible. Most CFD researchers also forget that there is also some very interesting physics that occurs in fluid flows that simulations are uniquely capable of capturing, as it is often the case the experimental data is incapable of capturing the essential flow physics.
  Reply With Quote

Old   April 2, 2005, 11:07
Default Re: What do you think?
  #4
Ford Prefect
Guest
 
Posts: n/a
>It seems that all progress comes from the muscle rather >than brains.

I must disagree. Developing new hardware (e.g. Processor architecture) is no muscle buisness.

>Unfortunately, how many smart algorithms have you seen >in the past decade or so?

Exactly.

So the point in my post was to give one way of increasing "muscle" with the use of "brains" although this "brains" part is more the brains of a mechanic than a programmer.

  Reply With Quote

Old   April 2, 2005, 11:13
Default Re: What do you think?
  #5
Ford Prefect
Guest
 
Posts: n/a
>AMD64s beat equivalent P4s hands down when it comes to CFD work. I run LES simulations on an Athlon 64 3400+, and it can run stably for over 2 months at 100% load.

Well, not entirely true. Since the AMD platform usually is more expensive the intel platform is quite good when it comes to clusters. But in the single computer case I totaly agree.
  Reply With Quote

Old   April 3, 2005, 04:51
Default Re: What do you think?
  #6
dpshaka
Guest
 
Posts: n/a
over clocking is not a excellent way . i think parrall computation has a good perspective
  Reply With Quote

Old   April 3, 2005, 05:27
Default Re: What do you think?
  #7
Ford Prefect
Guest
 
Posts: n/a
Would you care to elaborate on that?
  Reply With Quote

Old   April 3, 2005, 20:46
Default Re: What do you think?
  #8
chen25
Guest
 
Posts: n/a
Interesting.
  Reply With Quote

Reply

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are On



All times are GMT -4. The time now is 06:19.