There are already to many complements on this issue,but i have no ideas about the solver using Cores or Ram for a perticualr problem of Ansys,CFX-11.

I have Intel Xeon E5320 Quadcore,1.86 GHz,1.86 GHz, 15.9 GB of Ram,Using MS windows XP prof X64. It shows 4 cores in the performance(CPU usage history) while running.Sometimes uses 100% while we run 2 or more processes at a time.

can anybody tell me how it shares the RAM for its four cores or do we need any setup to use max ram for all cores. Actually We r thiking to add another quadcore,will that boost the solver speed by double? Or should we switch over to two Dual Cores?

Your current set-up is hopelessly inefficient and under performing.

Your mistakes are:

-Wasting an enormous amount of money on RAM you don't need. You don't need more than 1-2 GB per core. CFX is cpu limited not ram limited.

-Your chip clock speed is far too low.

-You're using an Intel quad core which scales poorly over 2 cores. At best you can effectively utilise 3 of the 4 cores.

Given CFXs high licensing cost it's imperative that you maximise the performance of each core. For the money spent on that box you could have built another system with 2-3X more performance.

The best bang for your buck would be stringing several desktop systems together e.g.

-3 GHZ dual core Intel chips. Intel desktop chips overclock so well and stably that you can just get e6600 2.4 or e6700 2.66 GHZ variants and perform a trivial FSB overclock.

-Desktop motherboard that supports ECC ram e.g. Intel BX2.

-1-2 GB RAM per core.

-GigE interconnect

If you want to stick to server hardware then use:

-Two 3 GHZ E5160 cpus

-Standard dual socket mobo

-FBDimm ram 1-2 GB per core

You can chain as many of these boxes together as you need with gigE.

If money is no object continue with the e53xx quad core chips but understand that you will only be able to effectively use 3 of the 4 cores.

AMD has a native quad core solution coming in late 2007/early 2008 ... Barcelona. However, current indications are it will perform little better than Intels upcoming Penryn class chips.

In 2008 Intel will launch Nahalem which will really offer a substantial performance boost and do proper multisocket scaling e.g. 4+ sockets.

Joe is quite right, I almost make the same mistake as you. You are much better off running in distributed parallel using dual core machines. Previous postings on the subject give about a 1.9 increase in performance going from 1 to 2 cores and a 1.4 increase going from 2 to 4 cores.

Create an cluster of the fastest dual core machines you can, with about 3-4 gig ram each machine running XP32 or XP64.

Stu

Agreed, using dual core machines in a cluster configuration gives excellent performance per £/$ invested in the hardware.

As per our experience, We see all four cores getting used equivalently when multiple runs or a single run with multiple processes specified.Many Of the times using more than 13 GB of Ram.

Also we got to know that 64 bit(xp)OS uses upto 4 GB of Ram per core.Thats why we went to 16 GB.

But we have not benchmarked anything. Still not decided to go with clusters,but in future if we want to enhance the performance,whether clustering will be a better option or an another quadcore will be effecitve as well as economis?