Optimal Hardware Configuration for ANSYS 1

[Shockdesigner331]

Hello all, I'm hoping that someone can answer a few questions for me. I am trying to spec out a DESKTOP computer to run ANSYS optimally (aka DREAM SYSTEM). I'm new to ANSYS, so please forgive any misconceptions that I may have.

64 bit or 32 bit? (I'm pretty sure I should go with 64.)

Operating system? (I'm mainly a windows user. xp 64 ed?)

Processor? (Core 2 duo? not sure if that is 64 bit though)

Memory? (I'm pretty sure that 3gb is the limit for xp, maybe I should just wait for vista and go higher?... is vista going to be 64 bit?)

Video Card?

Hard Drives? settings? (I want at least 10k rpm SATA or SCSI drives)

Any other suggestions?

I've tried searching the posts for this material, but after the 4th screen or so I decided that it might be best to get new, up to date info.

Thanks.

 

[cbrn]

Hi,
have you already looked at the ANSYS Inc. document where the recommended specs are listed? I don't remember exactly where it was, but I seem to remember that it was pretty easy to find.
In addition:
- 64 bits vs 32: of course useful if you must handle very large problems. Careful: don't buy a hi-end 64 bits platform just to analyze a 3D tube with 790000 solid elements... Machine power must save you AFTER you have done all YOUR possible to set up an efficient model
- XP 32 bits handles "officially" up to 3GB, but as ANSYS has its own memory manager, it is still unclear whether you are able to use all 4GB or not under XP32
- multi-processors: consider it only if you already reached the top of the tops with a single processor: not all ANSYS' code is multi-proc, so it's better to have one 2x-powerful processor rather than two 1x-powerful processors
- mass storage: SCSI sub-system is a must

Regards

 

[doatis]

Good question

-Since there is no price difference for 32/64-bit OS (Windows XP), I would go with 64-bit for a dream system

-I would go with a dual-core processor. This will allow you to run other applications while ANSYS is solving. I've had good luck with AMD Athlon X2, but there is some good buzz about the Core 2 Duos (run cooler). The price between the two (Intel vs. AMD) is fairly negligible. I wouldn't go with more than 2 processors, since the speed gains don't scale linearly

-This is where it gets expensive. If you get a 32-bit Windows OS, get 2GB of RAM (there's no need for any more, since you can't use it). If you get 64-bit, I've seen systems in excess of 12 GB ($$$)

-Actually, ANSYS v10 runs ~30% slower on SCSI when compared against SATA. Fastest appears to be SAS drives. Using two drives at RAID 0 (striped) is also a good idea.

There is actually a presentation done by PADT (a channel partner in Tempe, AZ) that goes over some stuff:

ftp.padtinc.com/public/seminars/2006_09_27_ANSYS_HPC.zip

Hope this helps,
Doug

 

[Shockdesigner331]

Thanks for all the info. I'll check out that presentation.

I have checked out the Ansys document with recommended system requirements. I was actually quite dissapointed with them (elementary at best). I would think that a rather large company such as ANSYS would run benchmarks and report their findings online. There is also no mention as to what chipsets their software is optimized for. If I'm going to spec out a computer and components from competing chipsets that are similar priced I might as well opt for the one that preforms better.

I have read that the SATA drives running in RAID 0 have beat SCSI drives in different compiling applications, but again, no mention of speed on the ANSYS website. Do they even test out different hardware platforms?

 

[cbrn]

Hi,
as per personal experience, 3GB under Win XP 32 bits DO ARE used completely by ANSYS 10, provided that /3GB switch is set in Boot.ini. Still no certainty with 4GB, since new workstations at work are equipped in that way but too recently provided (I still have no statistics).

Also, as regards SCSI vs SATA, at work we now have SATA disks with 5 msec seek time and 10000 rpm: EXTREMELY fast, but when using multitask heavily, SATA seems to remain a disaster if compared to the latest SCSI architectures (still present in some workstations which will soon be dismissed for uniformity's sake... sigh...) I don't have numerical benchmark though, so on this point Doatis' considerations surely are worth be considered.

I don't want to make advertisment, mine is only a comparison term: the workstation I use is a HP xw4400 and it can solve a multi-step analysis with elasto-plastic non-linearities as well as large displacements turned on (full Newton-Raphson algorythm), 275500 DOFs, 21 steps, in 24 mn. A linear elastic single-step analysis on the exact same model is solved in 20 sec or so. Sufficiently fast for what I do, but this might proove largely insufficient for others... Consider that the largest model I ran on this machine had 550000 elements and about 990000 nodes (so about 2.97 MDofs).

Regards

 

[Mech151]

Check out the dell 490. You can get 16gb ram, 2x 19" monitors, dual hard drives in raid 0, and 4 cores for under 5K. A better solution would be the 'server' class workstation like cbrn is discussing. Also, Sun has a line of computers with AMD processors. They are rumored to have better performance than intel chips. Three comments on improving performance:
1. Raid 0 configurations can create highly fragmented files, so defrag often.
2. For large memory systems, turn the page file off.
3. Cleare the windows/prefetch subdirectory.
(I reboot using a .bat file that takes care of 1& 3):

del /q C:\WINDOWS\Prefetch\*.*
defrag c:\
gpupdate /force /boot