improve the hspice speed with P4 HT

p4 ht optimization

I have done some experiments to see the speed of HSPICs in PC
HSPICE0403 running the testbench bigmos.sp,
I usingexternal timer to count (hspicetime report in LINUX is not corespond to real time)

condition speed(s)
-----using linux RH72 version of hspice0403
Athelon2000+ RH80(2.4.18) 28
p4 3.2g ht winxp+vmware4+RH80(2.4.18) 22
p4 3.2g ht winxp+vmware4+RH80(2.4.20SMP) 18
p4 3.2g ht RH Fedora1 SMP 19
------using windows version of hspice0403
p4 3.2g ht windows XP +hspice.exe 22
p4 3.2g ht windows XP+hspice_mt.exe(2 cpu option) 27
p4 3.2g ht windows XP+hspice_mt.exe(1 cpu option) 14
the conclusion is that the best option for hspice is runing under windows using hspice_mt with just 1 cpu option, we expect they can optimize the linux version in the future

Hi

I think it will worth if you test the following conditions too


1. You optimise the OS environmnet
BY:
- Disabling Antivirus, Firewall, Spyremoval Tools,
- Stopping non-important services
2. You test also in
- Windows 2000 Prof. with SP<4 and SP=4
- Windows 2003 64 bit version


tnx

Infact, the machine used to test the hspice is a newly installed winxp Home system( 1G 400Mhz ram) with just several useful thing and also the anti-virus is disabled
The point is that, it looks like for windows version of hspice, it must use the multi-processor enabled binary, but let the cpu itself do the multi-thread optimize(1 cpu option)
for linux, the SMP version of kernal will help to improve the speed, but not as good as windows do, they must reoptimized

i had known that the hspice under windows is not as accuratly as it under linux,because of the OS.

I'll be very interested in the benchmark of running HSPICE under the latest 2.6.x kernel.

I have a problem

what Tx size in this benchmark ??
if simulation time only 26sec ..
I think this circiuit is very small ..

if Tx size is large (whole chip simulation)
.. how about Linux/WinXP performance ?

tlihu said:

I'll be very interested in the benchmark of running HSPICE under the latest 2.6.x kernel.

Click to expand...

Hi

It is fine if you could test and run it under 64-bit linux cores (suse or fidora)

tnx

I agree.

Without optimization, linux and linux smp get almost the same speed (linux smp is a little faster).

i think it's not supported by hspice that muti-cpus