In a recent review article (I forget) I learned that the AMD CPUs are faster even than the Pentium IV's for most applications. Similarly, I've read that the much advertised Rambus RDRAM performs very poorly compared to the SDRAM at (even) slower MHz! At the same time RDRAM is very expensive!

If there are people with experience/information that is contrary to the above, I'd really appreciate their input. However, if the above is correct then who are some of the more reliable/respectable PC "manufacturers", which use AMD-based CPUs?

I'd appreciate any input to this end.

Adrin Gharakhani

1) www.pcworld.com www.pcmag.com & www.tomshardare.com all have confirmed that the AMD K7/Athlon CPUs running at the same speeds at the Pentium 4 CPUs perform as well or better in the majority of benchmarks out there today, as well as the majority of applications.

2) If you buy a Pentium 4, god have mercy on you as you should well know the reasons why Intel is currently advertising the Pentium 3 on TV, not the P4.

a) They goofed on the P3 and made too many. Also, when it was first released, both CPU and motherboard chipsets from Intel had bugs that required them to recall them.

b) When they released the P4, guess what? The exact same thing!! Thus, they can't even make enough bug-free P4s to supply everyone, have a glut of P3s, and are sadly losing market share to AMD due to the non-optimization of the P4 for current applications.

(Yes, even a P3 running at the same speed as the P4 is faster in most benchmarks)

3) For good places: Gateway is one such vendor that uses AMD K7's.

However, you can get a well-built system for much cheaper if you know where to go.

eg. in California, there's a free publication monthly by www.microtimes.com.

all of the local vendors listed there should be able to get you a complete 1Ghz K7 AMD Athlon system with 17" monitor, CD-ROM, KB, Mouse, Modem, Sound, Spkrs, box, etc. -- the whole shebang for $599 or less.

YES, $599 for a 1Ghz system complete - If you spend more, it's either for a brand-name and their 24hour support and service (better make sure you're gonna be up a 3am or you don't need 24hour support, just 9-5 support that the www.microtimes.com listed vendors will provide for a 1year.)

For the best performance of these 500+Mhz PCs, better drop in a RAID IDE card and hook up 2-4 identical HDs for 2-4x the performance from your HD subsystem.

eg. you can currently buy the $99 Fasttrak RAID/66 raid card from www.shopper.com. benchmarks on this card can be easily found through a brief search of www.yahoo.com.

(of course, if you've got a 100ohm resistor and 5 minutes of time, you can solider it in and flash upgrade a Fasttrak Ultra/66 card into a RAID/66 card, but only spend a small $20 for the Ultra/66 card instead. Again, a quick search of Yahoo! will bring up the relevant do-it-yourself upgrade instructions to save you $$ on this card upgrade - buy another HD with the $80 difference).

At work, a 1Ghz AMD system runs rings around all else with either a 70GB 7200rpm IBM HD or a RAID subsystem. Both will provide superior performance on both CPU and HD intensive tasks, and sure plays the latest trailers from Charlie's Angels w/o a skip!

d =)

I have both a PIII 700 with rambus and a AMD athlon at 1000mghz.

I bought the amd athlon after the Intel PIII and I though It will be much faster than the PIII (based on reviews).

But for large 3D CFD calculations I was very surprised to discover that both delivered the same speed.

The speed of the microprocessor Isn't the only thing. For large calculations big chunks of data from RAM must be fed to the microprocessor as quick as possible, since the data can't always be brought from the level 1 and 2 cache. Therefore the speed and bandwith beween the cache memory and the microprocessor and from RAM and the cache play and important role. I know, the Intel PIII is faster for both (thransfer between the RAM and the memory cache and from the cache and the CPU).

For small calculations, where the data can stay in the memory cache (I beleive this is the case for some benchmarks) and the micropressor is always fed in, the Athlon can deliver the best results.

My results agree with some spec benchmarks:

http://www.specbench.org/osg/cpu2000...00.html#SPECfp

(look at dell precision workstation 733,P and AMD at 1000mgz)

(1). The CFD field is so small among the PC users, that CPU and PC are not designed specifically for CFD users at all. (2). Traditionally, super-computers are designed to handle the number crunching of CFD problems. And the workstations are used only to do pre- and post-processing of graphics intensive tasks. (3). I have to say that using PC to do large scale CFD job is a poor man's approach. This includes the use of workstation, which is not much faster than the current PC at all. (4). So, benchmarking of CFD programs on PC will always give conflicting results. One also has to ask "what CFD program", "which CFD problem" and "what mesh size". (5). The good news is: PC will provide a dynamic enviroment for those experts who really knows CFD and knows how to write a code to take advantage of the PC enviroment. (if you are running a commercial CFD code on workstations or PC to solve complex problem with chemical reactions, make sure that you are not taking away all of the computing resources from the rest of the company.) (6). There is a huge mismatch between the degree of difficulty of 3-D CFD problems and the available computing hardware platform such as PC and workstation. This has a great impact on the solution reliability and accuracy obtained.

I can second Sebastien's comment. We bought a dual P-III/800 and a 1GHz AMD K7 about 6 months ago. My idea was that the AMD would make a good development machine, and the P-III would be better for production runs. The code we use is not parallelised, so the dual processor machine's only advantage is the ability to perform two simultaneous runs. Even though we run the 1000 MHz AMD at 1100 MHz (without problems or extra cooling), it is not really any faster for CFD than the P-III, despite the higher clockspeed and faster floating point processor. It seems logical that memory bandwidth is the bottleneck. The AMD certainly completely humiliates the P-III for small benchmarks where memory bandwidth is irrelevant, but that means little for CFD work. The three things we need now are 1) DDR-SDRAM 2) dual AMD motherboards and 3) 64-bit AMD processor. DDR-SDRAM is already here, and it would be interesting to see if it allows the AMD to open the gap on the P3 for CFD applications. Dual and 64-bit are on the way as well ...

Interesting information, thanks for sharing it! I have one small comment - if you look on the web-page with spec results that Sebastien gave you'll find specs for an Athlon 1.2 GHz CPU running in a Gigabyte GA-7DX motherboard with PC2100 DDR SDRAM - This configuration beats all PIII based machines by a margin, although the margin isn't as big as you would expect.

You should also remember that all PIII results are with PIII machines with PC800 RDRAM memory. This memory is about 4 times more expensive than the PC133 memory used in most Athlon machines. Hence, the Athlon machines will be much cheaper if you need a lot of memory.

Thanks all for the invaluable information based on first-hand experience.

Jonas, as you already mentioned RDRAM memory improves the performance but apparently for a very tiny amount. Indeed, the price/performance will not justify buying rdram. Check out the very interesting article at http://www4.tomshardware.com/mainboa...529/index.html and/or search for rambus at http://www.tomshardware.com

Thanks again for the info

Adrin Gharakhani

some additional comments:

Intel: The P4 needs a special compiler. Does this compiler exist ? Is still a cfd-code compiled and optimized for the P4 ? benchmarks ? On the other hand: Intels roadmap is 2GHz this year. 2Ghz with optimized code....

Why a RAID-System ? I do my calculations with ram ..

What about dual-systems for the P4 ? I heard it is not possible...but can you use GBit-network ?

How important is the L2-cache in cfd ? If it is: xeon (dual) 1GHz with 2MB cache...if it is not: celerons with rdram ?

I'm not a hardware-expert but I was astonished about the P4-benchmarks: The P4 is slower than the Athlon, but most testers say, the P4 has a great new architecture....

Jurek

The P4 needs a special P4-optimizing compiler using the new SSE2 instructions to come anywhere close to the spec numbers you find in the web-page given by Sebastien. With a standard PIII optimized code the performance of the P4 at 1.5 GHz will be close to the PIII at 1 GHz.

Today only Intel has P4-optimizing compilers. I asked Fluent a few weeks ago if they intended to supply a P4 optimized version soon and their answer was that they did not have any immedate plans for it but if there were signficant customer demands they would... I think that the conclusion is that optimized P4 codes will take some time. They were even still waiting to get a P4 machine.

About the L2 cache - it doesn't have a significant impact. I've benchmarked a PIII against a Xeon and the difference in speed at the same clock-freq. is negligable running both our in-house codes (compiled with Absoft fortran for Linux) and Fluent for Linux.

Forget the celerons... too slow memory bus.

(1). You can send e-mail to him about the hardware side of questions. I don't think he is reading cfd-online, because I asked him to answer this AMD question specifically for me. (2). He has the most popular Toshiba libretto website: adorable toshiba libretto , http://www.silverace.com/libretto/ . You can visit his website and e-mail him there also.