|
[Sponsors] |
January 28, 2009, 10:31 |
Hardware extension advice needed
|
#1 |
Guest
Posts: n/a
|
Hi all, Among others) I have following problem. In close time I am going to rebuilt my hardware. Currently my computer is Core2Quad q6600 with 8GB of ram. OS is WindowsXP64 and 32bit ansys. I am going to extend the system (not the software) and wanted to ask more experienced users for advice on the field. 1)I want to buy single chassis workstation. The cluster solutions are planned in the future but now I dont want to build a multi-computer cluster. My goal is to improve the speed as much as possible in single machine. 2)I was planning to buy 2xQuad Xeon Series 54xx. The price for that system is still problematic. Searching the forum I noticed posts written about 2 years ago, that CFX scales poorly above three cores of single processor. Since that what is a most suitable way: investing in dual Quad-Cores or two as-fast-as-possible dual Core Xeons which are much cheaper nowadays. Can anyone say if it is true about the bang for cache advantage of dual cores now? Can anyone present test results of some test case on modern xeon E54xx quad cores versus fasters dual-core xeons. 3)I have heared that new i7 platform has no memory bottle neck but has problems with CFX. 4)Since I have 32bit ansys the memory for single solver is up to 4Gb? Is it true? If so, is this a bottleneck in dual core machines (maximum number of 2 processors time 2 cores time 4Gb = 16 Gb)?
I would be very grateful for serious responses since the funds for the machine are limited and I want to spend it reasonably. Luk |
|
January 28, 2009, 19:06 |
Re: Hardware extension advice needed
|
#2 |
Guest
Posts: n/a
|
Hi,
It is not just CFX which suffers poor scalability at quad core, it is all applications which use quad core. It is due to the Intel CPU design, not the fault of the software. For extensive benchmark results look at www.spec.org. The most relevant results for CFX is the CPU floating point benchmarks. You can get an idea of both serial and parallel performance from spec. The new i7 CPUs are a bucket load faster than previous intel CPUs. I know of no problems with running CFX on an i7 machine, it should go like a train. If you want to have 4-8 cores on a single machine you should consider AMD machines. They scale to multiple cores better than Intel chips, but the new i7 chips fix most of that. You have a 64 bit system running a 64 bit OS so why have you installed 32 bit CFX? You should install the 64 bit version. Regards, Glenn Horrocks |
|
January 29, 2009, 04:29 |
Re: Hardware extension advice needed
|
#3 |
Guest
Posts: n/a
|
Hi Glenn, I hoped Your answer. Going from the bottom. - My reason for using 32 bit ansys is price. Down here, the 64bit ansys is very expensive. Anyway I do not think its a problem as long as we are not going to I7 Core. I have heard that this is VERY fast machine in ansys but there are problems with that stuff running multi-core. I am simply afraid to invest in I7 because I am not sure if I must buy for example cfx v12 for multi-core runs. Did You see the I7 running multi-core in cfx 10? If so, then I7 is my choice. - There are some results of runs of cfx benchmark in ansys communities but are difficult to compare. For example - my actual system is q6600 and it seems to go within 35 sec of CPU time. There are also answers of 2xE5460 going 18 seconds but under linux, pvm and I dont know it is total or processor time. - I was not considering AMD since I have heard that single core is slow comparing to Intel.
Thanks, Luk |
|
January 29, 2009, 18:47 |
Re: Hardware extension advice needed
|
#4 |
Guest
Posts: n/a
|
Hi,
When you buy ANSYS you get both 32 and 64 bit versions. You do not need to pay additional for 64 bit. Talk to your support rep. I have not seen a CFX benchmark on the i7 but the spec benchmarks are very good and the next machine I will be buying will be the workstation/server version of the i7. If you have any questions about the benchmarking on the CFX Community site put a post up there and I am sure you will get a helpful answer. If you are looking at 8 core performance then AMD machines become competitive. Definitely give them a look. Glenn Horrocks |
|
January 30, 2009, 05:44 |
Re: Hardware extension advice needed
|
#5 |
Guest
Posts: n/a
|
Hi, 1) For me it is not a problem to have 32bit ansys at all. The only thing I am thinking about is possibility of running the cfx on i7 core. Fred reported on this forum about the fact that he was unable to run cfx on i7 in way different than serial. Anyway in serial run the processor beat the 5335 going multi-core. 2) Looking through the SPEC floating point tests I noticed 2 things: a) the i7 is fastest. b) the single dual core e8600 Intel has about the same speed as 2xE5450 cores!!! For me it is strange.
Luk |
|
February 1, 2009, 17:53 |
Re: Hardware extension advice needed
|
#6 |
Guest
Posts: n/a
|
Hi,
I find it hard to believe the i7 can only run serial. I bet Fred had a problem with his setup and there is no problem with multi-processor on i7. Glenn Horrocks |
|
February 2, 2009, 04:38 |
Re: Hardware extension advice needed
|
#7 |
Guest
Posts: n/a
|
If You are sure of that, than it can be said that I7 is probably the fastest (and for sure relatively to price the best) solution for CFX.
Luk |
|
February 2, 2009, 19:05 |
Re: Hardware extension advice needed
|
#8 |
Guest
Posts: n/a
|
Hi,
There is no question in my mind that the i7 is the best value for money processor at the moment. My next computer upgrade will be an i7, or the workstation/server equivalent. Glenn Horrocks |
|
|
|
Similar Threads | ||||
Thread | Thread Starter | Forum | Replies | Last Post |
[snappyHexMesh] Advice needed | lindsaywestraadt | OpenFOAM Meshing & Mesh Conversion | 2 | February 18, 2010 03:20 |
[OpenFOAM] Xwindows crash with paraview save | srinath | ParaView | 1 | October 15, 2008 10:37 |
SUPG for 2-D Navier-Stokes advice needed.. | nicole | Main CFD Forum | 2 | November 23, 2002 01:06 |
advice needed! | Filip | Main CFD Forum | 2 | April 26, 2002 16:27 |
advice needed in solving linear system | yf yap | Main CFD Forum | 5 | February 5, 2001 01:21 |