Complex Math, Accurate & Fast :: PC or Mac?

[FlamDrag]

I know I'm in a biased group here, but I need to ask anyway.

If memory serves, Mac OS or the G5 processor does not do math as well as Pentium class processors. Is this true?

I also need advice for a hardware and software solution (PC or Mac) that will handle very complex math calculations with obscene quickness and accuracy.

Thanks,

FD

[Dr. Dastardly]

What?! Macs are the best at complex mathematics.
Look here for a great program dealing with complex math or just take a look at some of their super computers.
Heres a thread with the latest.

[crazzyeddie]

IIRC, the G5 does faster floating point operations than the P4. The G5 also has two FPUs per CPU, right?

[Abstract]

I don't know which ones will do maths more accurately, but I thought this was dependent on the software used, along with the method used by the software to take integrals, for example. Its very difficult to say, but things like Matlab is available for Macs now through X11.

I always thought floating point arithmetic became less accurate as the number got bigger (think 10^11 or something), and was much more (relatively) accurate when dealing with small numbers such as decimals. I didn't think accuracy depended on whether it was Mac or PC, IBM or Intel or AMD, although I could be wrong.

Floating point is faster on a G5, though.

[FlamDrag]

Seriously, I know that it SHOULDN'T matter if it's Mac or PC etc etc. but I have fuzzy memories of one processor or another actually generating incorrect answers on down the line. So, I was just trying to shave the fuzzy memory a bit.

I didn't realize mathematica was as diverse as it is.

So if I'm doing math and I want it done now, is there any particular part of the hardware that I need to upgrade to the hilt?

[Timelessblur]

for the hi end stuff like Math program and stuff OSX comes up a bit short. there are just few option advible but this is many because this program are more gear to the engineering comunity which is mainly PC side.

I have never been a fan of mathlab. I personly like mathcad a lot more and I have not heard good thing about mathematica compared to the other ones which I think are planly just better (Mathcad Mathlab and mapple are all better)

[RacerX]

Originally Posted by FlamDrag

but I have fuzzy memories of one processor or another actually generating incorrect answers on down the line.

Your memory isn't be too fuzzy, you are most likely remembering the Floating Point error that arose in some early generations of Pentium processors. Intel still carries that badge of honor in circles where floating point is key. A whole generation of PCs had to have there floating point units disabled to make sure that the errors didn't pop up in calculations using apps like Excel.

Both POWER and PowerPC processors have always been more floating point intensive than anything based on x86. Pentium class processors are just not designed for that type of stuff. PC users usually beat users of other platforms up with benchmarks, but in pure floating point results, Pentiums just aren't up to speed.

In the workstation world, your floating point ability is what you are measured by. It was one of the reasons why fewer G5 processors were needed to out score a Pentium 4 processor based system in the Top 500 Super Computer lists, the benchmark there is floating point.

This is also why some Photoshop filters seem disproportionately faster on Macs than on PCs compared to other benchmarks. The filters are using floating point calculations. Anything that uses integer calculations tend to give an edge to Pentium based processors as that is what Intel has pushed in their products.

Also, you may note that when Apple displays benchmarks, they tend to emphasize tasks which are floating point intensive.

But, PC still scroll Word documents faster than any platform on the planet.

[Bruce Lee, PhD]

Originally Posted by FlamDrag

I know I'm in a biased group here, but I need to ask anyway.

If memory serves, Mac OS or the G5 processor does not do math as well as Pentium class processors. Is this true?

I also need advice for a hardware and software solution (PC or Mac) that will handle very complex math calculations with obscene quickness and accuracy.

Thanks,

FD

What software are you running? While altivec-aware apps will run quite quickly on g4/g5 processors, I'm pretty sure matlab and other popular computational packages are not altivec-aware. I believe it's because on the g4, the altivec processor only handled floats and not doubles. Perhaps someone else knows if this is still true on the g5. What do you mean by complex? By the time it hits the CPU, no computation is really complex. In any case, it seems to me the processor really isn't your concern here; the software matters quite a lot more.

[Chaszmyr]

Mac or PC really won't matter much, but Macs are just cooler

[outerspaceapple]

Originally Posted by Chaszmyr

Mac or PC really won't matter much, but Macs are just cooler

you can sure say that agian

[budugu]

I am no expert on this. but i used to work on a dual xeon (3.0 Ghz) machine, now i am using a dual 1.8 Ghz G5 at work. We run huge simulations using matlab and what not to solve a lot differential equations etc for simulating small sub systems in neuroscience. Even though i have to agree that rendering is much faster on my new mac (creating visual 3D models). If you want to use matlab there is nothing that beats the Xeons practically. frankly i never got to work on Opteron so i have no idea about them. I should also give in the fact that Xeon workstations came with a lot more ram than the G5s (3.5 GB vs 1.5 GB).

PS: i remember long long ago that matlab used to run extremely well on my Sun 5 Sparc station.

[Bu88a]

I'm a physics graduate student, so i'm using my powerbook for math all the time. between mathematica and latex i've got pretty much all the flexibility i need (although i also have an older dell laptop with debian on it that i keep around for odd stuff)...

it really depends on what specifically you need it for. but i think the mathematica community is probably the more diverse, and largest (as far as mathematics software goes). I have a 1 GHz 12" with 768MB and mathematica runs like a dream...

now if only i could save up for a dual g5....

[FuzzyBallz]

Opteron server.

[Mav451]

oh god, i just spent the past 4 days making these Matlab programs for ENCE201.

[earthtoandy]

Originally Posted by FlamDrag

Seriously, I know that it SHOULDN'T matter if it's Mac or PC etc etc. but I have fuzzy memories of one processor or another actually generating incorrect answers on down the line. So, I was just trying to shave the fuzzy memory a bit.

I didn't realize mathematica was as diverse as it is.

So if I'm doing math and I want it done now, is there any particular part of the hardware that I need to upgrade to the hilt?

hmm sounds fishy. sounds like issues with the algorithim rather than the chip. because a chip wouldnt get very far if it couldnt do math since thats what it does

[Erik R.]

Originally Posted by Timelessblur

for the hi end stuff like Math program and stuff OSX comes up a bit short. there are just few option advible but this is many because this program are more gear to the engineering comunity which is mainly PC side.

I have never been a fan of mathlab. I personly like mathcad a lot more and I have not heard good thing about mathematica compared to the other ones which I think are planly just better (Mathcad Mathlab and mapple are all better)

So, what you seem to be saying is that only a few options for math is available on the mac, and that Matlab is no good.
Also Mathematica is no good, but Matlab is better and so is Maple.

As it happens we use Mathematica (VERY good app) AND Matlab on the Macintosh platform on my university. Mathcad is not good for a number of the things we do. it is simply not good enough. Maple is available for Mac, we have it in an older version here.

Maybe you could share with us what kind of computations you use mathcad for?

[Loge]

There was supposed to be a G5 optimized version of Mathematica in the works; not sure where this is up to.

http://www.wolfram.com/news/g5development.html

[minton]

These are the overall benchmark results of running a test notebook in Mathematica 5.0 on various hardware: (higher score = faster)

AMD64 3200+, SUSE Linux 9.0 for AMD64 : 3.55376

Athlon 64 3200+, Revision SH7-C0, 2.25 Ghz, 1MB L2, Windows : 3.5179

Opteron 244 (1,8 GHz), 2GB, SuSE Linux 9 Pro for AMD64 : 3.40254

AMD Athlon 3200+, 2GB RAM, WinXP Pro : 3.23186

Athlon 64 3200+, Revision SH7-C0, 2.0 Ghz, 1MB L2, Windows : 3.1481

AMD Athlon 64 3200+, 512 MB, Windows : 3.02386

Opteron 244 (1,8 GHz), 2GB, Windows XP Pro SP1 : 2.83638

Opteron 244 (1,8 GHz), 2GB, SuSE Linux 9 Pro for i386 : 2.73023

Pentium 4 XT ("Extreme Edition"), 3.2 GHz, 2GB, Win XP Pro : 2.6867

xp-2700, 2.17 GHz, 333 FSB, 1 GB, win-xp-pro : 2.54757

Athlon 2800+, 512 KB cache, 333 MHz FSB, Win XP Pro : 2.50588

Pentium 4, 3.2 GHz, 2 GB, Win XP Pro : 2.48287

Dell Precision 650, 4X3.06GHz Xeon, 512KB L2, 4GB, Win XP Pro V5.1 : 2.43878

P4, 3.2 GHz, 2GB RAM, 512kb, XP SP1 : 2.41661

IBM ThinkPad T41, Pentium M 1.7 GHz, 2.0 GB RAM, Win XP Pro : 2.39752

SGI Altix 3300, 1.3 GHz, 19 GB, Linux : 2.22437

P4-B, 3Ghz, 4GB, W2K3 : 2.1325

Mac G5, 2GHz dual, 1.5GB RAM, OS X 10.3 : 2.08193

The Thinkpad Pentium M, even with bottlenecks which are inherent in laptops, scores higher than the dual Powermac G5. The test notebook includes calculations like

Timing[PrimeQ[2^9689 - 1]][[1]]
Timing[N[Pi, 800000]][[1]]

See more at http://smc.vnet.net/timings50.html

[Catfish_Man]

Some issues:
1) Mathematica does not use dual processors, unless you're running two computations at once. If you're doing that, make sure to double any benchmark scores you see for dual machines (unless they make sure to use them effectively).

2) Macs can do 32 bit math extremely quickly if properly written, because of Altivec, but it doesn't support 64 bit math.

3) If the task and software allow it, a cluster of cheap PCs will almost certainly win in performance/$

[Edot]

Originally Posted by Catfish_Man

Some issues:
3) If the task and software allow it, a cluster of cheap PCs will almost certainly win in performance/$

Not according to Virginia Tech and the new Colsa!

[Timelessblur]

rember 2 CPU at most only give you a 25% improvement. 2 cpu does not eququal 2 times the speed never has and never will.

By the looks of it the more complated OS eat more and more power off the CPUs. Also remeber that a lot of calucations can not be split up since it take an answer before it to get the next step. Normally for calucations RAW cpu speed is among the most imporant thing for doing it fast. Now you may be able to run mulitple varibles though the same eqiuation at the same time but 1 process pretty much handles the number crunching for each run.

[FlamDrag]

I'm sorry that I'm being so vague with my questions; I'm in the very early stages of an idea and trying to get a handle on the computing power I'll need.

I won't need to generate 3D models, but I will need to generate the answers to about 40 problems all at once. In short, what I need is "burst" power. A computer that will sprint through problems, not run a marathon. So a small cluster might be the way to go.

On the software end, I may end up writing my own code.

Linux & Unix solutions are not out of the question either.

Cost is not a deal beaker.

[Bruce Lee, PhD]

Originally Posted by FlamDrag

I won't need to generate 3D models, but I will need to generate the answers to about 40 problems all at once. In short, what I need is "burst" power. A computer that will sprint through problems, not run a marathon. So a small cluster might be the way to go.

On the software end, I may end up writing my own code.

if writing code for a cluster sounds like fun and your problem parallelizes naturally, a cluster can be a good way to go. it can also be a real PITA to write the code and configure everything, so that's something to consider if this is just for fun.

why don't you write a little C program which represents the sorts of computations you'll be doing and post it here? if you write clean cross-platform code, which shouldn't be hard for this sort of thing, i bet a bunch of folks would be happy to compile it and benchmark it for you. if you know how to write altivec code, or write for one of the altivec-enabled libraries included with os x, you could write an altivec version as well.

[jmfr123]

Why the hell are you guys comparing a Pentium 4, it just sucks for pure Mathematics. The winner here could be the Ahtlon Family of processors from AMD, or EVEN their Opteron line. Specially that now, It all ready has Linux 64-Bit and soon to be released Windows XP 64-Bit. I have seen many tests where Apple's G5 and AMD's Athlon/Opteron both win 50/50. Depending on what optimizations you use on the Processors. Oh, and we know some of the top of line Athlon machines cost less than $2,200.

[cwtnospam]

Originally Posted by jmfr123

Why the hell are you guys comparing a Pentium 4, it just sucks for pure Mathematics. The winner here could be the Ahtlon Family of processors from AMD, or EVEN their Opteron line. Specially that now, It all ready has Linux 64-Bit and soon to be released Windows XP 64-Bit. I have seen many tests where Apple's G5 and AMD's Athlon/Opteron both win 50/50. Depending on what optimizations you use on the Processors. Oh, and we know some of the top of line Athlon machines cost less than $2,200.

You're forgetting to add the cost of downtime due to hacker and virus attacks, not to mention dealing with tech support just to get the thing working on your network, printing, etc. That will make your cheap Windoze based PC much more expensive than a Mac.