How to run ANSYS CFX-v11.0 (with 4 tasks) in a Intel quad core using windows XP of 64 bits?

Which of these run modes do i have to choose?

a) Serial b) PVM Local Parallel c) PVM Distributed Parallel d) MPICH Local Paralell for Windows e) MPICH Distributed Parallel for Windows

Do I have to make a different configuration on setup of my motherboard?

In Portuguese: Como funciona o ANSYS CFX-v11.0 (com 4 licenças) em um processador com 4 núcleos, quad core da Intel, usando o Windows XP de 64 bits?

Qual destas modalidades de funcionamento eu tenho que escolher?

a) De série b) Paralela Local de PVM c) Paralela distribuída PVM d) Local MPICH paralelo para Windows e) Paralela distribuída MPICH para Windows

Eu tenho que fazer alguma configuração diferente na configuração (setup) da minha placa mãe?

Assuming CFX is set up correctly and you have appropriate licenses you should be able to use PVM or MPI. You should obviously choose Local. Distributed is for running on many different machines.

Thanks!

Use MPICH Local Paralell for Windows, PVM sucks.

Don't expect too much performance benefit when using more than 2 cores. The bus cannot feed all cores fast enough on these CPUs for scientific applications.

but i intend to use windows xp 64 bits with 4 GB of ram. Will I have the same problem?

My PC:

Processor: Intel Core 2 Quad Q6600 (Double Core of processing) ( http://produto.mercadolivre.com.br/M...8-mb-cache-_JM );

•Socket 775 LGA;

•8MB Cache L2;

•Clock real 2.4Ghz;

•Quad Core;

•FSB 1066Mhz Motherboard S775 ( http://produto.mercadolivre.com.br/M...2-duo-quad-_JM );

•Chipset Northbridge: Intel 975X - Southbridge: Intel ICH7R;

•3 x PCI - 2 x PCI Express 1x - 2 x PCI Express 16x;

•Support memories with ECC - Support Dual Channel DDR2 - Support up to 8GB of memory - 4 slots DDR2 SDRAM (240 pinos) - Support memories DDR2 533 / 667 / 800;

•Chipset: Marvell 88E8053 - Ports: 2 ports RJ-45 - Speed: Gigabit (10/100/1000Mb/s);

•Channels: 8 (7.1) - Codec: Realtek ALC882M (high definition) - Connections: Output S/PDIF optical coaxial;

•2 channels SATA 3Gb/s provided by chipset JMicron JMB363, with option of RAID 0 / 1 and JBOD (being a port e-SATA) Memory 4GB;

•4GB;

•DDR2;

•800Mhz;

•PC 6400.

Hi,

Yes, the limitation is in the hardware. Intel Quad core is a waste of money for CFD. AMD quad core is much better, but their processors are not as fast.

Glenn Horrocks

Sorry Glenn, we just started with new Xeon 5460 (quadcores) 3.16 GHZ on linux with dual processors boards. They beat the sh*t out of AMD. I ran the CFX benchmark and it works great. As long as we use maximum of 2 cores per CPU we get nearly 100% parallel efficiency. I'm glad we didn't go for the 5160 Xeon and use them as single cores. Using the 5460's as dual cores save a lot of money on hardware.

After Intel launches the new Nehalem CPU's (with memory controller on CPU like AMD), Intel is the choice for CFD applications (and no I don't have shares in Intel).

Hi,

Good to hear things have moved on. Intel is working hard on fixing the multiple core speed problem and as you say when Nehalem appears the problem should be fixed.

Are your machines dual CPU as well? That is 2x 5460 quad core or just a single quad core per box?

Regards, Glenn

Glenn,

the machines are indeed dual CPU. Compared to our old Xeon who had a desastrous parallel efficiency, this looks great so far. I must say I'm actually suprised your Q6600 has the same performance. The 5460 is new (45 nm), has a higher clock frequency, and is more than 10 times as expensive.

Well, so much for the economics. I haven't seen any CFX11 benchmarks on the latest AMD (Barcelona) actually.

Hi Bart,

I was also quite surprised how fast it is in serial. Do you think the small size of the benchmark problem is distorting results? In my experience the benchmark in CFX gives similar speed figures to other larger and more complicated simulation benchmarks so I think it should be OK.

Looking at www.spec.org it seems there are a lot of processors within the top 10% in performance on single processor. I haven't looked at the throughput metric closely but I but this is where the difference between the top CPUs and the also-rans are.

Regards, Glenn

At least we know now where the profit margin is in the CPU market. Also my $100 intel Core duo 6750 at home is doing quite good. I must say that there is a lot of progress for us in the parallel efficiency. The old Xeon's we had were just terrible. I was considering AMD but they are just not fast enough en not available in abundance. I understand by the way that CFX 12 is going to have more (and bigger) benchmark examples? I do believen that on a core by core comparison the present benchmark is ok. A bigger one will just be testing you interconnect.

How big of a benchmark do you want? There are no plans.

According to a moderator (CFX) on the Ansys CFX customer portal there will be new benchmark files in CFX 12. But no details. A nice benchmark would have say 200000 nodes. Not too big (no time/resources for lots of people to run it).