What is the best platform to run ABAQUS/CAE?

[megamanS]

My school's Sun SPARCs just won't cut it and I'm sure there's something better than a P4. Any ideas?

 

[gsiener]

We ran on W2000 machines with eXceed at my school. This was on P3 HP workstations. I'm sure if you wanted you could run on a 2x opteron and you'd see amazing results (considering ram is a huge factor and opterons can address 16GB (or more?) ram).

Also, not sure if you use Pro/E at all, but the newest version can distribute processes, make a server farm/cluster very useful.

 

[alexruiz]

Originally posted by: gsiener
We ran on W2000 machines with eXceed at my school. This was on P3 HP workstations. I'm sure if you wanted you could run on a 2x opteron and you'd see amazing results (considering ram is a huge factor and opterons can address 16GB (or more?) ram).

Also, not sure if you use Pro/E at all, but the newest version can distribute processes, make a server farm/cluster very useful.

I second the Opteron suggestion. Or Athlon 64 if budget is constrained.... I don't know how well it would run in an Athlon XP, but I assume it would run great too.

 

[HokieESM]

Originally posted by: megamanS
My school's Sun SPARCs just won't cut it and I'm sure there's something better than a P4. Any ideas?

How much money are you talking? It also depends GREATLY on what models you're running. I'm in grad school doing computational mechanics (specifically penetration/metal forming simulations)... and I've seen a LOT of mixed results.

For big elastic problems or any method that actually STORES the stiffness matrix, RAM quantity is the big thing. Believe it or not, until you solve THAT issue, most everything is immaterial. Once that is taken care of, its usually just getting a fast floating point processor. Unless you're doing shockingly huge models, any currently produced processor is relatively fast. The P4 works well (I have one here that's great). I've used AthlonXPs and seen good results (especially considering the money on the mid to low end). I'm sure the Athlon64 would work well. For all out performance, you should go to an Itanium. MOST finite-element codes don't benefit greatly from multiple processors.... so you have to get the best single processor available. Now if you're doing CFD, you can parallelize until your heart is content... so go to a 2xCPU (or more!) system.

For methods that don't store the stiffness matrix, it gets complicated. The code I've written is VERY memory bandwidth dependent. The P4 stomps all over the AthlonXP. The Athlon64 should run well. I know the PowerPC chip in the G5 runs relatively well. Cache also becomes very important... especially if you can store the solution vector in L2. BUT, I've seen methods that are entirely CPU bound, too..... so you get the best FPU processor you can find. The Athlon64 would be great.

Sadly, computational mechanics isn't really cut and dried (of course, nothing really is....). Get some friends to benchmark the code you plan on using--that's the only REAL way to tell.

If you're looking for sane prices, either a P4C or an Athlon64 would serve you fine. The "jump" to a 2x system can be worthwhile for a CFD modeler... but not for finite element methods. The jump to an Itanium class PC... well, make sure someone else is paying for it. If you've got a LOT of work to do, I'm still of the thought that 2 P4s and a KVM switch is better than the Itanium for this kind of work.

Anyhow... good luck.

 

[megamanS]

Well, a P4 with eXceed is the best I've got to work with now. The Sun SPARCs are just really really old, apparently. Thanks for the advice.
BTW, how does Mechanical Desktop compare for modeling?

 

[megamanS]

My main issue was that it would take almost 10mins just to load over the network on most any machine. But the Suns couldn't redraw worth crap.