Hi there, I'm just got my workstation with dual intel Xeon 3.8 GHz processor and 4 GB RAM on it. The computational speed (iteration per time step) is however not notably faster than my previous PC (Pentium IV 3GHz, 1 GB RAM). The only different is on CPU resources utilisation which is only 25% on the new machine whereas 90% on PC. So I wish to fully (100%) utilise my CPU resources but I don't have any idea to do it. I appreciate greatly any suggestion from you in dealing this problem. I'm currently working with RSM model on 500k grid cells. Should parallel computing or something alike would speed up my computation? Thanks in advance.

1. the intel prescott-based cpu has a feature of high frequence and low efficency. so single 3.8G Xeon won't be notably faster than 3.0G P4, especially northwood-based P4. 2. i don't know what your OS is, but it seems the parallel computing already worked.(from 90% to 25%). 3. have a try, employ parallel computing at a less complicated work.

Go parallel (this will nearly double the speed) and switch off the hyperthreading (this won't change the speed, but will make the CPU utilisation look better).

I've experienced remarkle speed up by switching hyperthreading off in the bios. Try that.

Hi Lobby, I have two Xeon processor in my workstation but it seems not any faster, for your information I'm working on Windows XP Pro. Anyway thanks everybody for the suggestion.

Hi Michael, i know that disabling hyperthreading will make the CPU utilisation look better but does that improve the iteration time?

Was it a real speedup (faster calculation) or did only the CPU utilisation went up? Fluent support always recommens to turn HT ON if the system runs stable.

RoM

It did in my case. I'm running Fluent on Windows 2000. I had a parallel job (2 pc's) where the slave machine was running with hyperthreading switched on. When I switched it off the time spend was reduced from 586 secs to 445 secs. Michael

That is why ROM, I'm looking for the way to make a faster iterations on my case actually and I'm not borther about the CPU utilisation, but I mention about CPU utilisation because I tought I might underuse my workstation resources. As I work on RSM model with 500k cells, the iteration time is quite slow even I'm running with my new workstation. So I will be happy to hear other people (in CFD) experience and suggestion. Thanks.

When I tested this, in Windows switching hyperthreading off made little difference to the speed. In Linux the picture was a bit more complex, but setting HT on and partitioning according to the number of logical CPU's (i.e. 4 on a dual CPU machine) really killed performance. I never got an advantage from HT, so it appears to be best to just leave it off. At least then the CPU utilisation makes more sense.