I need the fastest system available - please help

Chrissicom

n00b
Joined
Jul 7, 2005
Messages
37
We're having problems with short timeframes and we need a really really fast computer for a program called Hydro_AS-2D which is a program to calculate surface waterflow from SMS models. Our current computers Xeon 2.4 Ghz and regular P4 3 GHz take several days for bigger operations that is too long. Even an Athlon X2 3800+ isn't faster (actually slower than the P4). What we need is a machine with extreme cpu performance for floating point operations, hard disk, i/o, memory bandwidth all isn't the bottleneck, we just need cpu power.

It doesn't matter if the cpu costs 1000 USD or 2000 USD if we need two, I just need to know which CPU has the most power when it comes to simple mathematical operations and no need for i/o performance or memory bandwidth. The program is supposed to support Dual CPUs, but I haven't noticed that yet, since it doesn't use our Dual XEONs power (only one core used). But a dual system is still better since we can run the program twice then. Anyway please advice :)
 
If you're looking for the best, enterprise-class CPU (For workstation operation, not server option) the AMD Opteron Model 880 is the best you'll find. If has two cores on the single die, but if you want a single core solution, you'll want to go with a Model 854. The 880 will run you $2,500 or so. It'll sure blow the heck out of those Xeons you have running.

You aren't looking into the X2, or the 64's. Hell, even the 64 FX's aren't what you need. You aren't looking in the consumer market, anymore. If you want to get serious, you go enterprise-class. The 880 will be the best you can get, between Intel and AMD for mathmatical processing, and RAW CPU power.

101082.jpg


If your program allows it, does it provide you to use CPU scaling? If you rig a couple 880's together, you can scale them, and completely double, triple your work output.

If you need anything else, or input for a design layout, go ahead and PM me.

-turbineS-line
 
Unfortunately CPU scaling is no available option with this application. Is a single Opteron 8xx in any way faster than an Opteron 1xx or 2xx? An 8-processor system is a little more than I wanted, that's 10k USD just for the CPUs. I'd say 2000-3000 should be maximum.

Also I thought Opterons have excellent server performance and are very good with i/o performance and file copying but Intels were very much faster when it comes to just calculate some stuff with the need for clock speed. Do you have a comparable chart that shows how many cycles are processed per second by a Xeon or Opteron when calculating Pi ? Sounds like a simple thing and actually Hydro_AS-2D doesn't do anything very much different than multiplying numbers and such things.

Oh and the software supports Windows only, so the hardware must be Win 2003 compatible ;)
 
Actually, the benchmark picture I showed in my previous posts is just to show you where the Xeon line sits, compared to the AMD workstation line. I don't know if I can find any actual benchmarks on the Opteron, but one Opteron 880 will sure operate better than any Xeon you may be able to throw at the program you're using.

The scaling is only an option. The Opteron 880 is your perfect solution for what you need it for. All you need is one of them. Sorry if I wasn't clear in my last post, but you won't need 8 of them to do this job quickly, all it'll need it one 880 and you're set. :)

I'm sure, with the Opteron being a workstation class, it'll calculate Pi much quicker than any CPU I've ever looked at. But, there you go! Look around on google, and you may find something.

Hell, even call up AMD and see what they reccomend, but that's my suggestion.

turbineS-line
 
turbineS has a point. Just get the motherboard, RAM, and one Opteron 880. If it DOES end up taking advantage of the Dual-Core, you can always add a second Opteron 880 later.
 
If you just need one or maybe two CPUs there is no point in buying an 8xx series. The 8xx chips are identical to the 1xx and 2xx chips, they just are enabled for 4+ way SMP.

Also, that Benchmark is from AMDs site, take it with a grain of salt. Workstation/Enterprise chips aren't any faster than the home-grade chips, they just take motherboards that require more epensive RAM, have ECC code, and other features to make them more stable for mission critical applications. Unless you are running a server where a crash means thousands of dollars lost for every moment it is down, you don't need Xeon/Opteron.

If you already know that your application runs better on the P4s than on the A64s, I'd recommend you get a P4 840EE. Dual core plus hyperthreading means ultimate multitasking ability (so you can run several instances of your program at once).

The P4s are great number crunching chips, the EEs are better than even the X2s when it comes to multitasking. AMD kills Intel when it comes to bandwidth limited tasks optimized for single threads, like gaming. The P4 is still more than capable, and as you seem to have found, faster at your program anyway.
 
yeah I'd need to look at the program you are using a little more but I'd imagine if it was optimized for intel no matter what you do the best results will come from an intel based processor... b/c there is no way in the world a 2.4Ghz xeon should outperform a X2 3800... sorry but won't happen unless there is something weird with how the software uses the cpu performance... As far as benchmarks for straight cpu performance you can check out some benches I have on my website of similar systems: http://www.laynet-home.com/new_page_41.htm

I went from a dual xeon nocona core 3.0 Ghz system to a dual opteron 252 just b/c of the bench on the website that had a AMD FX55 destroying my intel system and almost beating the time for a single processor job by like 20 seconds. Just to note the AMD is running at 2.6Ghz compared to a 400Mhz increase that the intel has. BUT that was for rendering and not for the purpose of number crunching...

So long story short go intel if it works for your program but to be honest AMD most of the time is faster pound for pound... It sounds like you only run one instance of the program anyways so getting a dual core might even be a waist... I'd find a 3.8Ghz xeon or P4 and go with it b/c it sounds like that is right along the same line as what you are using and it works well but the 3.8Ghz should be a substantial improvement in speed being over 1Ghz faster and depending on the mobo it might upgrade seemlessly (I doubt it though)...
 
Very helpful input so far. I was also in the impression that there is no difference between the 8xx and 2xx models except of SMP capabilities. So it would indeed be a waste to buy 8xx series since we will definitely not run more than 4 instances on one computer. Right now the program does not support multi-threading but it should in the future and even now we can run 2 or more instances on multi-threading capable system so dual core or even dual core with HT is definitely the way to go.
About Xeon/Opteron stability, currently the program runs up to a week or more to have finished results and a crash after 5 days can actually mean a few thousand dollars losses. So stability is an important thing. In my experience with Intel CPUs though I noticed that the latest P4s are as stable as a Xeon system. Don't know about EE, never had one.

Here is some more info for you guys:
The developer of Hydro_AS-2D told me that it was programmed using Fortran and compiled using "The Intel Compiler" whatever that means. He also noted that the program is not CPU optimized which I get confused with since it's compiled with a compiler with Intel in the name :confused:

We've made a test run today with some small input on an Athlon X2 3800+ (our webserver usually) and a P4 3 Ghz (800 FSB version) workstation and neither CPU could outperform the other, they were equal in performance +-3%. Both Computers have 1 GB dual channel memory of the same brand (Corsair). So it really seems Intel is faster.

I know AMD has more performance with games and an Opteron is better for a file server than a Xeon in most cases, but that doesn't matter here. As I said I need pure calculating performance, it's just numbers...
maybe going for a P4 840 EE is the best option?

Just a question... The P4 840 EE is a Dual Core CPU with HT enabled right? So it's like a Dual Xeon with HT, logical-core wise?
 
Chrissicom said:
Very helpful input so far. I was also in the impression that there is no difference between the 8xx and 2xx models except of SMP capabilities. So it would indeed be a waste to buy 8xx series since we will definitely not run more than 4 instances on one computer. Right now the program does not support multi-threading but it should in the future and even now we can run 2 or more instances on multi-threading capable system so dual core or even dual core with HT is definitely the way to go.
About Xeon/Opteron stability, currently the program runs up to a week or more to have finished results and a crash after 5 days can actually mean a few thousand dollars losses. So stability is an important thing. In my experience with Intel CPUs though I noticed that the latest P4s are as stable as a Xeon system. Don't know about EE, never had one.

Here is some more info for you guys:
The developer of Hydro_AS-2D told me that it was programmed using Fortran and compiled using "The Intel Compiler" whatever that means. He also noted that the program is not CPU optimized which I get confused with since it's compiled with a compiler with Intel in the name :confused:

We've made a test run today with some small input on an Athlon X2 3800+ (our webserver usually) and a P4 3 Ghz (800 FSB version) workstation and neither CPU could outperform the other, they were equal in performance +-3%. Both Computers have 1 GB dual channel memory of the same brand (Corsair). So it really seems Intel is faster.

I know AMD has more performance with games and an Opteron is better for a file server than a Xeon in most cases, but that doesn't matter here. As I said I need pure calculating performance, it's just numbers...
maybe going for a P4 840 EE is the best option?

Just a question... The P4 840 EE is a Dual Core CPU with HT enabled right? So it's like a Dual Xeon with HT, logical-core wise?

It will show up as 4 CPUs - 2 phyical cores, 2 logical cores.

The P4EEs are supposedly very stable. Get a solid motherboard known for stability and not for overclocking (since I doubt you will be doing this, and you aren't going to get much more about of an EE chip anyway), maybe an Intel branded board, those are known for being pretty rock solid, good quality RAM, solid components all the way around (you will need a beefy PSU for the EE) and you shouldn't have stability issues, especially if the box is purpose built to run this one program.

I'm curious though, if is taking weeks for this simulation to run, does this program only work on Windows? I'm thinking Alpha/Power4-5/Sparc type systems might be noticeably faster crunching the numbers, if you have a means of running the software on something like that...
 
Yes the tool is Windows only and all the competing offices have the same problem as we have, they don't even have P4 3 Ghz they do it on slower machines in 2 weeks ;) :D

The developer is currently working on a multi-threaded version (that's why I want to go Dual Core and possibly HT) and also on a cluster version which can use the CPU power of several computers. I don't know how this should be achived though because there is only 1 step calculated at a time and step 2 requires step 1 to start so it will be tough to share workload between computers. When step 1 and 2 would be a seperate process then of course you could use two computers but since only a single step can be calculated at once I think clusters aren't going to work.

Thanks for all your input, when the system is built I let ya know the results :)
 
Yeah if the program does end up getting optimized in some form or another let us know cause that changes quite a bit... Especially if it gets 64-bit optimization or gets coded with a specific CPU brands instructions...

Could you distribute the processing work? IE: Have mutliple computers working together to crunch? Cause cost wise you might be able to get more bang-for-buck if you distributed the computing.

Just a though.
 
I'm not sure if you have already done it, but consider tweaking the computer. Turn off un-needed services, turn off any firewall and antivirus software (unplugged of course). If you are running XP, turn off eye candy. If your company uses corporate spyware (eg: "Altiris" or "SMS") uninstall it.

You can also tune process priority in Task Manager. Most processes run with a priority of "Normal" but you can increase priority for your number crunching software to "High" or "Realtime". Likewise, you can lower priority for some other processes.
 
Optimizing the OS or chaning the task priority has 0 effect on the application as we tested already. There is nothing else executed on the workstation anyway. Also running a live virus scanner during the calculation process has no effect. We've done it in a 3 days test, and both runs (with, w/o virus scanner) were within a timeframe of 5 minutes so I don't think it actually has any effect. A live scanner only scans files when they are accessed and Hydro_AS-2D is calculating most of the time and only writes a few KB every few hours into a simple Text file.
 
If you could try to explain what this program does in better detail...

How big are the files it deals with (not outputs, but inputs)
What type of data is the software trying to find or what is it doing with the data?

This can change a lot of what types of benifits you can durive from hardware and software changes.
 
I recommend:

2xAMD Opteron 64 model 254's (Troy Core) (2.8GHz w/1MB L2) (280's will work also.)
Tyan K8WE (S2895) nVidia nForce Professional
4x1024MB Corsair ECC Reg. DDR400 (NUMA enabled)

Mines solid and has always been perfectly stable. That setup would surely best the A64 X2 3800+. If for no other reason than shear clock speed advantage. Xeon's and Pentium 4's as well, but for architectural reasons there.

Taken from http://www.amd.com

http://www.amd.com/us-en/Processors/ProductInformation/0,,30_118_8796_8804,00.html
 
Sorry but 4 GB of Ram will be a waste of money, even 1 GB is more than enough for Hydro_AS-2D.

What it does in more detail (sorry if I don't use the correct english terms). We create terrain models with SMS (http://www.ems-i.com/SMS/SMS_Overview/sms_overview.html) of a river and the surroundings. Hydro_AS-2D is used to calculate the flood levels for a HQ10,30,100 etc. (a flood that happens ever 10,30 years etc.) using the SMS mesh model. You can find more details on our website at http://www.hydroconsult.net/leistungen_de.htm but it's all in German so most of you will not be able to read it. Also I can't explain these details pretty bad since I am only the system administrator responsible for the hardware and I am not into technical detail considering hydraulic engineering.

We have a file server in the office with a dual Xeon 2.4 GHz and 2 GB registered Ram which is a lot faster with file copying and network traffic than our normal workstations with 3 Ghz P4 and 1 GB of ram although all computers have the same SATA hard disks (in different sizes). Hydro_AS-2D is very slow on the Xeons though because as far as I could find out until now it only depends on the clock speed when you compare one Intel CPU with another. And the fact that an Athlon X2 3800+ can't beat a P4 3 Ghz also shows me that Intel might be the proper choice for a high-end machine dedicated to Hydro_AS-2D.
 
Chrissicom said:
Sorry but 4 GB of Ram will be a waste of money, even 1 GB is more than enough for Hydro_AS-2D.

What it does in more detail (sorry if I don't use the correct english terms). We create terrain models with SMS (http://www.ems-i.com/SMS/SMS_Overview/sms_overview.html) of a river and the surroundings. Hydro_AS-2D is used to calculate the flood levels for a HQ10,30,100 etc. (a flood that happens ever 10,30 years etc.) using the SMS mesh model. You can find more details on our website at http://www.hydroconsult.net/leistungen_de.htm but it's all in German so most of you will not be able to read it. Also I can't explain these details pretty bad since I am only the system administrator responsible for the hardware and I am not into technical detail considering hydraulic engineering.

We have a file server in the office with a dual Xeon 2.4 GHz and 2 GB registered Ram which is a lot faster with file copying and network traffic than our normal workstations with 3 Ghz P4 and 1 GB of ram although all computers have the same SATA hard disks (in different sizes). Hydro_AS-2D is very slow on the Xeons though because as far as I could find out until now it only depends on the clock speed when you compare one Intel CPU with another. And the fact that an Athlon X2 3800+ can't beat a P4 3 Ghz also shows me that Intel might be the proper choice for a high-end machine dedicated to Hydro_AS-2D.

Didn't say you needed it. I was recommending that setup for the CPU power. I'd decide between two higher clocked single core CPU's or two lower clocked dual dual core CPU's for four cores.
 
Hello Chrissicom, i m also thinking of using Hydro-AS 2D for my thesis but the problem is that i have not yet found any English tutorial or manual on Hydro_AS, Can you please help me in this regards.
 
Last edited:
Hello Chrissicom, i m also thinking of using Hydro-AS 2D for my thesis but the problem is that i have not yet found any English tutorial or manual on Hydro_AS, Can you please help me in this regards. <bibekkk@hotmail.com>

Dude, you just bumped up a 5 year old thread. In addition, Crissicom hasn't been on this forums since 2006.
 
Hello Chrissicom, i m also thinking of using Hydro-AS 2D for my thesis but the problem is that i have not yet found any English tutorial or manual on Hydro_AS, Can you please help me in this regards. <bibekkk@hotmail.com>

lol wut?? :confused: :p

thread_necromancer.jpg
 
Hahaha, I didn't realize the thread was 5 years old. I was baffled at their recommendations. I haven't heard anyone suggest a Pentium 4 in years.
 
haha, this thread got me also. I was sitting here thinking why has no one reccomended a 980x. lol
 
Haha I think I looked at the date by the 2nd or 3rd post...that Opteron chart looked awfully suspicious, lol.

Ahhh, 2005. Prior to the release of Conroe when the majority of sigs contained AMD rigs because Intel was playing catch-up. Seems like just yesterday!
 
Back
Top