I am again curious about benchmarking for scientific number crunching on AMD and Intel, respectively. Most of these kinds of requests on this site are followed up by intense debates among a core of people who probably know a lot about AMD/Intel systems, but I don't feel like I'm getting the bottom line as far as number crunching is concerned.
SO... I'm looking to find the best system for my university lab. The main purpose of this machine is to crunch numbers through matlab, C or Fortran. As far as I understand I have the following options:
I don't know what mobo to get, but have been recommended the P4DCE+ i860 with the i860 chipset for the Intel system. As far as the AMD system is concerned; I don't know.
Price is an important factor, but performance in pure number crunching is the main purpose of this machine. I'm not interested in overclocking, as I don't want to jeopardize any warranty issues that it might interfere with.
Some issues that I'm interested in are: Hyperthreading, bus speeds, direct CPU to memory channels (?), etc.
Does anybody have links to benchmark tests that investigate this kind of performance? I'm not interested in graphics / game / mpeg-encoding and the usual benchmarking results. Anyone? AMD or Intel?
... I almost forgot: we're using Intel's Fortran and C-compilers for Linux... These won't work on an Athlon, will they? Intel claim a 40% speed increase over Lahey / PG just from using these Pentium optimized compilers...
-------------------
whats your budget?
-------------------
Well, we already have the HD, and the budget is about US$3000.
We need a 100mb/s ethernet card, a DVD (instead of CD), an 8MB videocard, no monitor (want headless system, but possibility to hook a monitor up for administration if necessary) and... well, the usual suspects.
I almost forgot: we're using Intel's Fortran and C-compilers for Linux... These won't work on an Athlon, will they?
They will. The Athlon/AthlonXP gets about a 30% boost when running code compiled with icc 5.x vs. gcc 2.95.x.
Here are some <A HREF="http://www.linuxhardware.org/article.pl?sid=02/03/14/1530218&mode=thread" target="_new">Linux-based test results</A> for AthlonXP vs. Pentium IV. There are some for AthlonMP vs. P4 Xeon on that site as well, but they're quite old--AthlonMP 1800+ vs. Xeon 1.7GHz.
Also some discussion of scientific engineering and MatLab in particular <A HREF="http://www.psychology.nottingham.ac.uk/staff/cr1/simd.html" target="_new">here</A>. Apparently MatLab does not support SSE2; since it's closed-source, there's not too much you can do about that. You can do parallelizing with MatLab to take advantage of multiple CPUs and possibly HyperThreading.
As far as HyperThreading goes, it seems to be a handle-with-care feature. In some cases it improves performance, and in some it actually causes a performance detriment.
As far as AthlonMP motherboards go, the Tyan Tiger 760MP is a very mature and stable motherboard. The Asus A7M266D has been released, but for a non-overclocked server, the only thing it really offers over the Tiger is USB 2.0, 66MHz+64-bit PCI, and possibly onboard IDE RAID.
<i>If a server crashes in a server farm and no one pings it, does it still cost four figures to fix?
I have now read numerous benchmarking tests on pure number crunching, and believe that an AMD system will do the job much faster, all test results are between 20-100% faster on the dual AMD vs. the dual P4 Xeon.
I have also checked for cost of these two systems and found the following (in CAD):
1. INTEL:
2*P4-2.0GHz Xeon = 2*720 = $1440
4*512MB ECC, Reg. RDRAM = 4*380 = $1520
MB Supermicro P4DCE+ = $690
add HD, keyboard, ethernet card, box, floppy, DVD and Video at ~$600 and get...
CAD $4250 + taxes.
2. AMD:
2*MP2000+ = 2*445 = $890
4*512MB ECC, Reg. DDR 266 = 4*270 = $1080
MB Tyan S2466N= $380
add HD, keyboard, ethernet card, box, floppy, DVD and Video at ~$600 and get...
CAD $2950 + taxes.
It seems that the Intel option is about 45% more expensive for less performance. AMD it is. Thanks, people, for input!
You are about to answer a thread that has been inactive for more than 6 months. If you still wish to proceed, please ensure that your posting is original and does not duplicate or overlap any prior responses to this thread.