Ansys Performance in rp5470 4 CPU

[Terry Haime]

I am comparing performance of ANSYS application (exactly same input data) between a PC with 2.4 GhZ XP Vs HPUX 11.11 rp5470. The memory in the PC is 2Gb and HP server is 12GB. Even though I can allocated more memory in HPUX, the elapsed time almost 75% more than that of the server. (in fact the server is using EMC/FC disks for temporary file management). Any comments from who is using HPUX server for Ansys FEA a simulations.

[Steven E. Protter]

I think the rp5470 assuming its recent clock speed will blow away the pc.

Here is why:

1) RISC, less instructions in cpu, more bang for less clock speed.
2) Memory. Thats what makes the test very unfair. HP-UX knows how to use the extra RAM.


Complicating factors:

Disk: Thats the big bottleneck, and an advantage to pc or pa-risc can change the whole outcome.

SEP

[A. Clay Stephenson]

Your statement "the elapsed time almost 75% more than that of the server" is probably reversed so that the HP-UX box is slower than the PC. The last time I ran Ansys was over 5 years ago so my experience is dated.You fail to mention how many processors are in the HP-UX box and what are the processor speeds. Ansys is so CPU intensive that as long as reasonable amounts of memory are available, whoever has the faster / more efficient CPU's wins. It would be helpful to post some Glance output; I assume that Ansys is multi-threaded so that all the processors are being used. The other aspect that I suspect is working against HP-UX is that it is 64-bit code; one the one hand that's a good thing because it allows much more address space for bigger models (or finer granularity) but it's bad because 64-bit code executes slower than 32-bit code.

For my last data points, HP-UX workstations were beating PC's but again my data are old.
I can remember some Ansys runs lasting for two weeks.

[A. Clay Stephenson]

Ooops, I see you did indicate the number of CPU's (4) in your title.

[A. Clay Stephenson]

The one other thing that would really clobber Ansys performance is if someone "improved" the timeslice value by reducing it below 10 (default); this is actually one of the few cases that might benefit from a somewhat larger timeslice.