• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

I am building a Supercomputer, and need your help!

W

WingZero94

Golden Member
I posted this over in the HT forum, but didn't know how many people browse it.


Ok ok, here goes. I am a noobie here, but thought you guys/gals could help out. We are building a supercomputer for my university. It will have 36 processors. They are doing parallel processing, which I don't totally understand.

Anyways this is what I need: Each computer has to cost apprx 600 bucks (36 of 'em). Each needs an 80GB HD, 512 RAM, and LAN. We have tried an Intel PIII 1000/815, AMD Athlon MP 1400/SISK7SEM, P4 1600/850, Celery 950/810. The to Pentium systems ran identical. The AMD was 10% slower and the Celery was 37% slower. What do you think we should use? It needs to be a Floating Point monster. TIA,

Paul
 
This is a sample I just built at Newegg.

ENLIGHT EN-72370AZ MEDIUM TOWER P4 /AMD ATX CASE - OEM with 340Watts Power Supply
ENLIGHT CPU COOLER FAN 60x60x20 MODEL# NH001E01 - OEM (was free with above case).
MAXTOR 80GB 7200RPM Model # 6L080J4 - OEM, DRIVE ONLY
KINGSTON KVR266X64C25/512 512MB DDR PC2100 RAM - OEM
EPOX EP-8K7A AMD761 AMD/ AMDXP DDR/SDRAM 200/266MHz FSB SOCKET A UltraDMA-100 MOTHERBOARD - RETAIL
Intel PRO/100 S Network Adapter
AMD 1600+/266 FSB Athlon XP PROCESSOR CPU - OEM

I think AMD's have great FPU performance and the above box before shipping was $544, so I don't believe shipping would be an issue. You should even be able to go up to the 1800+ for $16 more and still keep it under budget. I am pretty sure the above system would outperform any of the ones you listed............😀
 
That assumes they have budgeted for racks and cooling systems for them. Towers just need a place to stand...............🙂
 
Have you considered building 18 dual processor computer instead? Double up on the memory. Might be less headache. Might save room? I believe I have seen articles on the web with universities building linux based computer clusters. But then again, 36 computer are more distributed.

just a thought.

good luck!
 
I think because of the limitations with memory bandwith, single processor's with their own memory might be actually faster. The limitation is going to be the lan setup anyway, but if its something like a distributed project then each box will be crunching its own part and only having to serve results to the compiler unit and get more work. Are you writing your own multinode kernel? Clusters (Beowulf is one type) is what they are called right?
 
I am not too sure about the SIS chipset you are using with the AMD, but if you pair it with the VIA266A chipset with some DDR RAM, then the FPU performance should be better than that of the P4 for the same clock speed.

If you not gonna run the CPU in SMP modes, then I don't see the point of using the Athlon MP chips as they cost considerably more than the XP chips.

I don't know what number crunching app you will be running in the background...but for your information, this is what I can tell you about SMP compared to single processor:

Both are my machines:
Machine A
AMD T-Bird 1.4G overclock to 1.6G
AMD760 chipset
256MB DDR RAM
Win98SE

Machine B
Dual Celeron 300A both overclock to 495Mhz
ABIT BP6 dual Celeron motherboard running BX chipset
386MB SDRAM running at 110Mhz
Win2000

Running SETI@home (Fast Fourier Transforms calculations) on both machines, Machine A will take on average 27 hrs per job, Machine B takes only 12 hrs per job. So for this application, the computer running dual CPU where the motherboard does not have a memory achitecture specially written for SMP processing (based on BX chipset) would still run FPU ops over twice as fast as a single CPU of a generation newer and 3x the clock speed.

But I guess if your application requires a lot of memory bandwidth, the case will be different.
 
Are you talking about work units? Are those numbers from single work units in SETI? If they are my 1ghz Athlon @ 1.4 did them in 5 hours.
 
Are you talking about work units? Are those numbers from single work units in SETI? If they are my 1ghz Athlon @ 1.4 did them in 5 hours.

Woah, I have no idea now...
Out of the 16xxx hours worth of SETI time I have done, some 465 jobs were completed. That's an average of almost 35 hours. The fastest I have ever seen done was around 11 hours on my dual Celeron machine. It needs 55-60 hours to complete a work unit at my work computer - P3 550 128MB RAM.

I am using ver 3.03 SETI now. I think I should upgrade to 3.07 and see what happens. After reading the statisitcs page on SETI home page, it does seem your 5 hours for the Athlon 1.4 is the norm. I really think it's becoz I am running older version.
 
He has to be running the GUI. My Celery 466@525 does a WU in appx. 24 hours on low priority. My work machine (before I upgraded) was a P3/866 which did one every 8 or 9 hrs.
 
do NOT under any circumstance use a SIS chipset for what you are trying to do.
your memory bandwith is 1/5 that of any other non ddr chip set for use with an amd cpu.
and ddr easly doubles the band whith, of all other non ddr boards.
if you are tring to save money by having built in video and lan and what not then go with the nvidia 420 chipset. it will only cost about $20.00 more than the sis chipset and easly give you 5x the memory bandwith!!
also having used the sis chipset for both amd and intel i can asure you that you will just be throwing your money away.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top