Best configuration for CFX high performance
 

Hi, all.
I´m new in CFX and Ansys, and I´m not sure how to dispose the software for maximum performance. Should I install O/S and Ansys in separated hard disk?
and yet should I use another separate disk for the file, and another for the temporary memory access? If not all my disks have the same speed, wich one should I use for each task? The fastest for the O/S, for ansys, or for the files ?
Thank´s

The software can be installed on the same drive as the OS. But doing the simulations (and all the temp files which go along with it) on a separate drive is a good idea. This is not going to make much of a difference for CFX because it is usually not limited by hard drive speed, but this is very important for FEA (eg ANSYS Mechanical).

So for your fastest drive use that only for the simulation results. The slower drives can be used for the OS and ANSYS software.

Thank you very much ghorrocks.
I want to say, that someone told me to stop some windows services like search indexing,etc for best performance, and also to constantly defragment and even format the disk containing the temporary files.

This is rather obvious but I'll mention it anyway:

If you're working in a corporate environment with large networks, don't save simulation files on network drives. Save them locally (I.e. local C or D or hard disk drives). Once the simulation is complete transfer the files over to your network drive.

Ethernet connections are usually bottlenecks. Even if you're running gigabit ethernet.

I've found this speeds my simulations up quite noticeably.

You definitely want to stop all processes which gobble CPU. Have a look at the task manager and if any task is using up more than a small amount of CPU then turn it off. As for disk defrag - again, CFX is not usually limited by hard drive performance so this is unlikely to help much. It will make a difference to FEA models like ANSYS Mechanical.

In my experience gigabit ethernet is no bottleneck up to about 4 processes, and not much of a bottleneck at about 8 processes. But if you have two quad core machines running 4 processes connected by gigE then you will probably have a bottlenexk in the network connection - but 8 machines each running a single core is fine. Note this is just guidelines, individual cases could be different.

thank´s guys.
In my case I´ll work on a PC, single processor 4 real cores (8 threads) intel I7 3820 10 MB cache, 64 GB RAM, Windows 7 at 64 bit, and 2 SSD 120 GB each working under 2 SATA 6Gb/s ports, later on I may add a PCIe 3.0/2.0 x 16 with 4 SSD working under RAID 0 configuration dedicated for the working and temporary file memory. Would be a good choice to do that?. Until now I was working on a laptop with just 4 GB RAM an single hard disk. So I really hope I´ll notice a change in performance.
Another question: do you guys know where I can find a GOOD Forum for Ansys Mechanical? I did easily found on the web this CFD-online forum, and I´m happy for it because people like you help me a lot, but as for Ansys mechanical it´s another matter.

Thank´s a lot

I see you posted a very similar post in the Hardware section as well, so I'll copy and paste my response from there below

Adding to that, from what I've read here:
Do not de-fragment SSDs, that is for HHDs

Also Glenn, you say:
But if you have two quad core machines running 4 processes connected by gigE then you will probably have a bottleneck in the network connection - but 8 machines each running a single core is fine.

But I've read the exact opposite on this on some ANSYS documentation?

My other Post:
{{
That's a lot of questions. I'll give my opinion on this but am not all that experienced with everything so you could do some research or wait for other contributors to answer as well. Are you using CFX, Fluent, Mechanical?

Which motherboard you use is up to you and what you need. From this question it sounds like you will be building your own computer? Just make sure you get a motherboard that will stably support 64GB of RAM, and make sure the RAM says its qualified with that motherboard or visa versa. I like the ASUS p9x79 deluxe, and P9x7-WS myself, though the ASUS boards don't support TRIM with SSDs yet, more on that later.

If you are going to be working on models that require more than the amount of RAM you have and you have to use your hard drive as RAM, (known as paging or disk thrashing) that is going to make a huge hit on your performance. Hard drives are orders of magnitude slower than RAM, and it would be almost pointless to have a fast CPU since you will be completely bottle necked by the hard drive. It would be better to either get a dual socket motherboard with 128GB+ of RAM or distribute across multiple 64GB computers.

With that said, if you have to disk thrash, a single SSD will be much faster than any HDD raid configuration. There was an ANSYS mechanical benchmark that showed an SSD 8x faster than a 15k RPM hard drive while solving out of core.

I don't believe you will get much speedup or difference from using a separate SSD for a working drive, but I do it anyways since then I can swap out or change that working drive anytime I want without affecting the operating system or other installed programs. I'm not sure whether ANSYS recommends you install the program on the boot drive or another drive though.

SSDs also benefit from "TRIM", but It doesn't seem possible to use TRIM on all motherboards when using them in RAID. TRIM is a command that erases unused blocks of data written in an SSD. If the SSD has to write to a block that has data in it already it has to erase that data before it can write, which affects performance obviously.

What I plan (not sure this is the right thing to do, but it sounds good to me with what I understand) is have one SSD as the boot drive with ANSYS installed on it. Then have two SSDs in RAID 0 as a working drive with nothing else installed on it. That way I can use "secure erase" as a manual trim until it is supported in RAID mode, which will hopefully be soon. One trim is supported in RAID mode, then maybe I would install ANSYS on the working drive as well.

Others input would be appreciated since I'm not 100% sure this would be the best configuration.
}}

Yes, Erik, have I´ve been naughty posting almost same question in other sections? ;)
I really was desperate to get some advises, because I´m willing to buy my computer during this week. I need to solve some FSI problem, and my laptop its not powerfull enough.
So I really want to thanks for all your contributions. You guys are really kind taking your time to answer questions from a novice like me.

I can´t afford a dual processor or more. So if I did understood correctly from here and other forums, the best I can do is to put together O/S and Ansys in same SSD disk, let´s call it main disk, and use other SSD disk/s for temporary and scratching files, once solved, save my file to the main disk, and later on erase or format the SSD disk/s. And I belive, even without this "TRIM" option, would be right to use 2 or more SSD disks in a RAID 0 configuration. Of course if all my RAM, 64 GB in my case, will be overused.
One more thing, If I´m not wrong, I belive, that if not all disks are same speed, I should use the slower for the OS and Ansys, and the fasters for the files. Am I right?

Thank´s a lot

The bottleneck in this case is often the connection of the motherboard and the network card rather than the network itself. Either way, for a grunty quad core machine you will need a high-end network to keep up with it. If you only have one or two processses per node then a garden grade network card is fine.

CFX does not gain by having a SSD. ANSYS mechanical will, however. But don't waste your money by putting the OS and ANSYS on a SSD. Just use a garden grade HD and spend the money you save on a higher spec CPU. CFX really just needs CPU grunt.

Glenn is correct, you don't need to use an SSD as the boot drive, an HDD would work fine here and have much more storage capacity.

Also not all SSD's are created equal, and I believe we need high IOPS more than high sequential read/write speeds. I've never done any benchmarking on which ones would be best, but I'm going to use the Samsung 840 Pro as they seem good from what I've read. Any better recommendations would be welcome.

Yesterday I went to the computer store, very happy, after reading yours and other suggestions about best configuration hardware. With money on my pokets, I had the intention to buy my new computer. But guess what! The guy on the shop, told me , that if I use an SSD for scratching memory or temporay memory or virtual memory, my SSD will start to slow down and even fail after some months. He suggested to use SSD for the OS and Ansys, and a HDD for memory and working files.:confused::confused::confused:

Comments are very wellcome

Thank´s

Sounds like a heap of flaming codswallop to me. Time to see what some wise people say:
http://www.tomshardware.com/forum/26...disk-life-span
http://forums.macrumors.com/showthread.php?t=1343568

So, I read this as saying that of course heavy use will wear it out but it would wear a HDD out too. And it will last a few years by which time you will be buying a new computer anyway. So you form your own opinion, but I stand by my first sentence :)

And I will say again that using a SSD for the OS+ANSYS drive will do almost nothing to speed your simulations up, and using a SSD for the scratch drive will not speed CFX up much either. The only thing it will speed up is structural FEA like ANSYS Mechanical.

Hoooolly cow! :):):)

If I did understood correctly, heap of flaming codswallop means, in that case, that the guy at the computer store had no idea what he was talking about and/or was too conservative.
And yes, I agree, I´ll be changing my new computer in a few years.... (I have a 1 year old 500€ laptop that performs much better than my 6 years old 2 xeon workstation). Tech capabilities increase and change at the speed of light.
So despite what the computer store guy told me, I´ll be using the HDD for the OS and the SSD for the files and memory. :D
And again Glenn, you are right, for CFX only purpose, my laptop performs perfect and Superfast, but since I´m doing FSI, structural analisis perform HYPERslow in my 500€ laptop, and even crashes sometimes.

thank´s for your help

Yes, that would be a good way to put it. Sorry if my Australian attitude offends but I like to get straight to the point :)

Ha Ha Ha! :):) No ofense at all. Better for me so I can learn some english expresions besides learning in computer hardware in the same time.
You know, it was even difficult to find this expression meaning over the web.
Furthermore, I allways like to get STRAIGHT to the point.... for good or for bad

Hi,

I have a question which is close to the topic of this page.
Sometimes, when I am making the Pre-CFX file, because of lack of memory the file gets closed. If the HD supports RAM, why the file gets closed?
I expect the process becomes slower but it should not be stopped, I suppose.
Is there any special change which needs to be done in order to make sure that HD is supporting RAM when it is necessary?

BR,

Please show any error message it produces.