CFD Online Discussion Forums

CFD Online Discussion Forums (
-   Main CFD Forum (
-   -   Itanium and CFD (

Jens Chr. Bennetsen June 13, 2001 09:12

Itanium and CFD

Since there has been quite a lot of discussion about the performance of Penitum (4) vs AMD processor. The last information seems to be that P4 is doing very well due to the fast memory access for commercial CFD codes.

BUT what about the brand new Itanium ???

Although it is not similar to CFD code, the LINPACK*1000 shows an impressive performance on the Itanium (>2300 Mflops) which is almost twice to the Alpha EV68 chip at 833 Mhz. The P4 and Athlon are doing less 350 Mflops for LINPACK*1000.

I know it has only been released since 4. june, but have anyone had it for testing ???

Kind Regards

Jens Chr.

John C. Chien June 13, 2001 13:27

Re: Itanium and CFD
(1). Do you think that the CFD result is a function of Mflops of a CPU? (2). Based on my theory, 99% of the time, we are getting the wrong results. Therefore, a faster CPU will generate more garbages at a faster rate. (3). Faster than a human can digest. So, when using faster CPU in doing CFD, the methods must be already very good. (4). It is better to slow down, and find a better method, than to invest in a faster CPU and spend most of the time in the garbage dump? Example: you set the parameters to generate a mesh, then the automatic mesh generator fail. So, you change a parameter, but again, it fail. I had to do this at least 30 to 50 times to get a good one. So, a faster CPU will definitely drive me to crazy. (5). Luckly, I am writing my codes now.

Scott Whitney June 13, 2001 14:11

Re: Itanium and CFD
There is a bit of discussion on this topic on the Fluent forum.

My quick thoughts: Itanium is great at floating point calculations, but a poor performer on all other aspects. Itanium was supposed to be out years ago, it was delayed so much that its successor will be released quite soon (maybe around the end of the year?). I find it a poor investment (easily $4000/processor) when the successor should have much improved performance. Itanium is Intels first attempt at 64 bit processing (and thus there may be hidden problems), second generation machines usually have most of the problems worked out.

If it took Mr. Chien 50 times to get a good mesh at 5 minutes per attempt (assuming they were complex 3D meshes) that would take just over 4 hours. With a fast computer (assuming 1 minute per attempt) the same 50 trials would take just under an hour. Thus at $50/hour for a trained engineer, a faster computer would save Mr. Chien over $150. A typical new PC processor costs around $150... I'll let you make your own conclusions. I do agree that faster does not necessarily mean better CFD.

John C. Chien June 13, 2001 14:31

Re: Itanium and CFD
(1). Well, part of it is a joke. (2). It takes oxygen for the brain to think. To think faster means more blood flow and higher blood pressure. This is a problem in US. (3). But I think, faster PC means more computer graphics which I like. The brain can process graphics more efficiently than using eyes to read numbers. (4).Unfortunately, current cfd program can not read computer graphic results yet.(to improve the convergence of solution) (5). CPU alone is not going to help the CFD (a good example is CRAY super-computer), but good computer graphics might help human to sort out the cfd problems.

Jens Chr. Bennetsen June 14, 2001 06:07

Re: Itanium and CFD
I am sorry that you think this way and that you have to let your sarcastic remark carry your away. Perhaps your problems are not important to solve and you do not like to test many difference things.

You like to keep your own code and solve all the problems in the world. Fin by me.

But lets face the fact, that some people here like to be able to solve flow problem in a fast and quick way in order to ask new questions. One way is to use faster and faster computer.

I think it a pitty that you so often get the wrong solutions, but maybe you have to put your question in a new way ! CFD is only a tool, it is not the only solution to the problem or question asked. Also since it is a kind of mathematical modelling it will always only be as good as the approximations you use.

Since this site is for discussion and exchanging ideas about CFD, I will suggest that you should go to a different site for your misplace remarks and stick to the agenda of this site.



John Van Workum June 14, 2001 09:24

Re: Itanium and CFD
The Itanium will be a fast chip, no doubt about it. And price will be high (until AMD releases there 64-bit chip or Alpha lowers prices).

But, the real benifit is when the software (in this case CFD tools) is written for the processor and the compilers to create the software are written for the processor. I haven't heard of any compilers for the Itanium and I'm sure the CFD vendors will be slow to optimize their codes.

It will be awhile before the industry gets their moneys worth out of the Itanium.

John Van Workum TTI

John C. Chien June 14, 2001 11:27

Re: Itanium and CFD
(1). Just remember that Intel is fighting for its survival or as the number one in PC CPU supplyer. The CFD application is too small and it can't make any difference for Intel. (2). Commercial CFD codes vendors are trying to make money, they don't invest money just because someone here is interested in Intel's new chip. (3). It is a 64-bit CPU, and you need a operating system and a compiler first. Without that,it is useless.

Joern Beilke June 14, 2001 15:17

Re: Itanium and CFD
SGI is working on the compilers (open source). The GNU compilers are also available.

D.M. Lipinski June 15, 2001 07:52

Re: Itanium and CFD
If you look for the compilers (C/C++, FORTRAN) for Ithanium CPU, just look at: Besides excellent performance, the compilers support parallel execution using OpenMP directives.



Philip Jones June 18, 2001 09:39

Re: Itanium and CFD

At Computational Dynamics, we have just got a couple of these boxes in and early benchmarks are pretty impressive. If interested, let me know and I'll post a full set of results once we have a more full picture.

John C. Chien June 18, 2001 11:38

Re: Itanium and CFD
(1). Please do. (2). Don't wait too long.

Charles Crosby June 18, 2001 14:13

Re: Itanium and CFD
Sigh.... No, faster does not necessarily mean better, but it is quite clear to me that those who belittle the importance of speed are working in research environments and not in product development. There is a huge amount of insight and understanding that can be gained by getting data at several angles of attack, for example, EVEN WHEN THOSE DATA POINTS ARE NOT NECESSARILY GRID INDEPENDENT OR FULLY QUALIFIED, as long as the CFD practitioner realises this. The product that counts is the one sold to the customer, not the CFD result, which is a small step in a long development process. To put it differently, sometimes it is necessary to use CFD as an approximate tool, much like a more sophisticated version of a simple pencil and paper calculation. There is, in fact, a continuous range of calculation methods that one can use when appropriate, ranging from a "back of the envelope" calculation, through what I would call "approximate CFD", all the way up to a fully qualified, mesh-independent, experimentally verified, fully converged, time-accurate CFD-solution. The trick is to know what you're dealing with.

OK, I'll get off the soapbox now. Sorry, can't help you with info about the Itanium, but I have done extensive benchmarking (with two different codes) to compare a 1.4 GHz Pentium 4 to a 1.2 GHz DDR-equipped AMD Athlon, and the result is that the P4 was approximately 5~10% faster with both codes, but about 70% more expensive.

John C. Chien June 18, 2001 16:06

Re: Itanium and CFD
(1). The Intel people should read your last comment. (2). In this way, they can avoid spending a lot of research money, just by cutting 70% off the price. (3). There is no easy way to win in the post-industrial revolution world. (4). The problem with CFD is: so far it is not a tool or a product,(yet). So, the need to do CFD is not essential. (but there is a potential in the future.)

Jens Bennetsen June 19, 2001 02:07

Re: Itanium and CFD

Well, okay, I am doing research, but a number of my friends also work with very large models and this is within advisory companies.

Also the main problem of using the P4 over AMD processor are the cost. And it may be Intel's real problem when entering the market for powerhouses that do a large amount of work in many offices.

As regards to Philop Jones comment. I am delighted to hear this and I am looking forward to the benchmarks, since we uses Star along with different reserach codes.

I known other CFD companies are working on the port, CFX has do the port, but not released it, Fluent has also done the port. But to look back the performance vs. cost are the main issues here.



Jurek June 20, 2001 02:10

Re: Itanium and CFD
Are the operating systems stable on the new Itanium ? I read, it is still beta-testing ( at the customer..)

Jens Bennetsen June 20, 2001 02:41

Re: Itanium and CFD

Well the Windows XP are still in beta, but Redhat 7.1 for the Itanium are a full 64 bit system, also the HP-UX are fully ported to this platform.

But it lacks general software ! It the same old thing when a new platform are introduced.



All times are GMT -4. The time now is 17:23.