And thinking in buying a portable to do my cfd out the university. I use fluent for 3D sails, and i limited to 1 million cells meshes whith the p4 2.6 500Mb Ram. But i´m not sure in which should i buy, the amd 64 3400 is maybe as expensive as p4 3.4. Could i use big enough meshes whith those computers? Which would be better for cfd? what things are important in the hardward configuration?.

thanks.

Get another portable with about the same configuration and run Fluent on them in parallel?

Or how about saving money for a dual or quad CPU machine and then leaving it on in your dorm room. Use the laptop for preprocessing and submit the job to the big boy via rcp, rsh through your campus network (which you would need for the 'floating' Fluent license anyway?).

Seriously, if you have a big case to run, a 1-CPU machine may not suffice even if the hardware config is the best available. 500MB is also the memory limit for a 1M cell, single precision model in STAR-CD. Shouldn't your uni have a cluster or 'big iron' mainframe you have access to?

I agree with cjtune that it is a waste of money to buy a portable to run heavy CFD calculations on.

What you should think about for a stationary system:

1. Check with the software manufacturer what kind of plattforms are supported (i.e. get the most out of your buy).

2. If both AMD and Intel are supported then check benchmarks at say www.tomshardware.com

3. As for memory I suggest the fastest the market can provide. That is some form of ddr 533 today. This all depend on your chipset ofcourse. And remember to buy alot of memory too since large simulations can drain memory resources and force some harddrive swapping (dramatically decreases performance).

4. If you have the money, go for a dual system (i.e. xeon or opteron in case of intel/amd).

5. If you are building yourself then remember to put some extra money into cooling.

Let me add to the discussion my recent experience on using notebooks for CFD.

I use daily a dual Xeon 3.06 GHz system (1MB L3Cache, 4GB RAM and NVidia Quadro graphics). Recently I benchmarked a high-end notebook with P4 3.2GHz (800 MHz FSB and Hyperthreading, 1GB RAM and a 128MB ATI 9600 Pro graphics).

On both machines I run my own general purpose CFD program (3D, arbitrary polyhedra, cell-centered Finite Volume method with hybrid MPI and OpenMP parallelization). To my surprise, on a single CPU the notebook was slightly faster (about 3%) mainly because of the faster bus (800 MHz vs. 533 MHz on Xeon). With shared memory parallelization (OpenMP), using two CPUs on Xeon versus two virtual (hyperthreading) CPUs (OpenMP) one the notebook, the dual Xeon system was faster by about 10%. Distributed memory parallelization (MPI) gave more speedup on the dual Xeon, but I will not give any numbers because MPI is more or less efficient depending on the amount of communication (and this is very problem specific).

The above sppedups I could confirm with two different problems, using 290 and 480MB of memory, respectively (the run times were abount 30 minutes for each case). The conclusion from the above benchmarks is that my application was strongly memory bound and it could run efficiently on a notebook. The post processing (Tecplot, MayaVi) was equally fast on both machines.

I'm far from suggesting notebooks for CFD. This depends very much on the program you want to use (I don't have any experience with Fluent). My message is that once you choose a high-end notebook (P4 3.2GHz+, 800 MHz FSB, hyperthreading and as much memory as it can is a must) you really need to try your application. You may be surprised by the results.

BTW, about AMD; Opteron is a great machine for CFD because it manages memory from the CPU (unfortunately you don't have mobile systems with Opteron). The things may change again, in favour of Intel, in the next months.

regards

DML

actually...

Opterons are now available in notebooks. One example is from Acer:

Acer Aspire 1511LMi Athlon64 3000+,512MB,60GB,DVD-RW,15"TFT.

I do NOT think it is a good idea to use NoteBook for CFD. I have done some benchmarks in both XEON and OPTERON, and OPTERON has better performance.

Thank you for your answer. Javier.