CFD Online Discussion Forums

CFD Online Discussion Forums (
-   CFX (
-   -   INTEL vs. AMD (

Michael Bo Hansen June 8, 2001 04:14

There is a lot of talking going on of which of the processors from AMD and Intel is the fastest one. Even though that clock frequencies from Intel machines is higher than the ones from AMD, computers based on AMD outruns Intel in a typical office package test. Is there anyone who has any experience in which CPU to choose when one wants to upgrade the hardware?

Thanks Michael Bo

John C. Chien June 8, 2001 04:43

(1). Determine your application first. (2). If your application can be handled by several computers, then find one which is cheaper. (3). If you want to be the fastest computer of your application in the office, then the choice is obvious. (which might be more expensive sometimes) (4). My computer has been upgraded many times, from 486 to pentium, then to AMD. But still, the choice depends on the particular application. The next upgrade will happen soon. It will be AMD if I keep the desktop, or it will be Pentium III if I move on to the laptop.

Li Xinfeng June 9, 2001 07:34

I use dual xeon 733. It's depend on your money in your package.

Jan Rusås June 9, 2001 09:23

Try to search in the main forum, about this topic. Until recently it was believed that the amd was the fastest one, for several reasons.

Only if the code is optimized for the P4 it would be faster than the amd. To my knowledge there is not any CFD codes that is optimized for P4. The same goes for the xeon, I recall from the main forum that the xeon only was 1 or 3 % faster than the pentium, not worth the (much) extra money. A test (main forum) though stated that the P4 was faster than the amd in a test with Fluent.

I have acces to two machines, one with pentium III and one with amd (Athlon) and with the test case (in this forum), I found that the Athlon with same CPU speed is faster than the P3 when using CFX. As a small note, my Athlon sucks compared with my Inlet based machine, when I use build, probably due to the graphic card in my Athlon.

I think the problem with some of the test, is that they only focus on the CPU, many other parameters is also important.( Ram type, Bus speed, disk cache , disk controller, I do not know much of this)

The test in this forum only used very little memory, disk acces, so in a test with a larger problem, with huge memory usage, some of the conclusions might be reversed.

Recently I have changed my pentium based machine, so I was in the search for a new. I selected a pentium machine !! (The reason for this, was that I wanted a Compaq machine and they only delivers pentium for their dual cpu machines, so in this case it was easy to select)

I you have the money why not investigate the possibility of an Alpha processor based machine from compaq, I do beleive they are much faster than Amd and Intel based machines.

Regards Jan

John C. Chien June 9, 2001 13:59

(1). There is an article in the July issue of PC World, page-164. The test shows that Athlon is faster than P4, for business applications. (2). For CFD, it is probably due to the inefficiency of the commercial code, rather than the hardware. (3). At least in the business world, Athlon is faster.(although the clock speed is lower)

Jan Rusås June 9, 2001 16:15

I do also think that the Athlon is much faster, but it would still be nice to see more test results from a real CFD problem, a large problem - several CPU hours and alot of memory usage etc. You may correct me if I am wrong, I do remember from the main forum that Jonas concluded in a new test that the P4 was faster with the Fluent code, surprising. Jan

Li Xinfeng June 9, 2001 22:16

I do also think that the Athlon is much faster, but for the situation of Intel, computer company only offer the intel base computer or pay more attention to intel cpu computer especial advanced computer. if I have enough money, I will buy a compaq alpha workstation.

John C. Chien June 9, 2001 22:22

(1). Athlon1.333GHZ is faster than P41.7GHZ for business applications. (2). For small memory access,(200kB?),Athlon1.333GHZ is again faster than P41.7GHZ, based on LINPACK test. (3). But because SDRAM is slower than (DDR SDRAM), (RDRAM), in main memory access, which is required for large memory access, P41.7GHZ (using RDRAM OR DDR SDRAM) is faster than Athlon1.333GHZ (using SDRAM). (4). But, in any program, you are not just solving a large matrix. So, there is a potential that for large CFD applications, P41.7GHZ can run faster than Athlon1.333GHZ. (5). But if you write a program which does not access the main memory in very large chunks, then, I think, Athlon1.333GHZ is faster than P41.7GHZ. (6). In other words, the performance of these computers are non-linear in terms of the memory usage. Athlon is always faster when memory access is small. (7). So, the performance will depend on the actual cfd program written, algorithm and memory size access. (I guess, RDRAM AND DDR SDRAM are more expensive than SDRAM) (8). This can push P4 into special application class,(such as large CFD application which requires large memory access) and it is definitely not a good trend. Because CFD applications can't support the CPU development.

Jonas Larsson June 10, 2001 06:27

Yes, that is true, and I was equally supriced to see how well the P4 runs Fluent. The benchmarks were of realistic sizes (up to several hundred thousand cells). There has been some more discussions about this on the Fluent forum recently - check out the thread entitled "What to buy ?". I don't know if CFX will run as well though.

Charles Crosby June 19, 2001 16:54

I would like to add a datapoint if I may. I'm in the process of getting a cluster together for tackling large problems, and we were able to borrow an AMD 1.2 GHz Athlon with DDR RAM and AMD760 chipset and a 1.4 GHz Intel P4. We benchmarked several grid sizes, from unrealistically small to "real-world", using two CFD codes, CFD-Fastran and a custom-written Euler code. Basically, the P4 was between 5% and 10% faster on both codes, although we were able to slightly overclock the AMD with Gigabyte's Speedtune utility (to about 1266 MHz) without causing it to overheat or go unstable. At this speed it matched the P4's performance almost exactly. However, the P4, with RDRAM, is significantly (50% to 70%) more expensive. There is little doubt that the Athlon's floating point unit is faster, but memory bandwidth does count, and the P4 is the clear winner in that department. FWIW, the chipset choice for the Athlon is very important. We also briefly tested another 1.2 GHz Athlon, but this used the Ali Magik chipset, and it was significantly slower than the equivalent machine with an AMD760 chipset.

All times are GMT -4. The time now is 02:55.