Sign-in / Sign-up
Your question

Quad core machine for numeric computations

Tags:
  • CPUs
  • Quad Core
  • Product
Last response: in CPUs
August 28, 2009 1:40:07 PM

Hi All,

I know, that it seems to be the forever question again: AMD vs Intel. basically yes,
but I can specify my needs quite exactly, and would be interested in your opinions.

So, I want to build a relatively cheap Quad core machine to carry out heavy numerical calculations, mainly in the MATLAB environment, and to run some own C++ code for the same purpose.

What I need is a stable machine, which will often run for days on full load, so I do _not_ want to overclock.
Furthermore I will need big amount of RAM, at least 8 Gb, that's what I am going to buy in the first round.

Additionaly the system should be small in dimensions, as I dont have much place, and relatively silent. I dont plan to extend the machine by buying some more HDDs, optical drives, whatever, so a uATX form would fit my needs.

Basicaly the frame of the configuration would be like:
*Antec P180 mini case, with Antec 550W PSU
*some Kingston 8Gb DDR2 800 modules (i dont want the current fastest one, as i am on tight budget), most probably in 2x4Gb kit, for further upgrade possibility up to 16 Gb.

So here comes the question of AMD vs Intel:
what I found to be similar price:
*M4A78-EM board (HD3200 AMD 780G/SB700) µATX + Phenom II X4 945 (4x3.00 GHz)
or
*MB P5Q-VM G45 µATX + Intel Core 2 Quad Q8200 (4x2.33 GHz)

*Scythe MINI NINJA CPU cooler

I dont care about the graphic stuff, any integrated "crap" will do the job, I dont play games.

The CPU clock rate I think is not the most important issue, important is the Quad Core, as the computations can take good advantage of the multi-core systems.

So what do you think ?

More about : quad core machine numeric computations

a b à CPUs
August 28, 2009 2:23:33 PM

Why the hell does he need a
synthetic benchmark?
Related resources
August 28, 2009 3:49:36 PM

Cryslayer80 said:
Why the hell does he need a
synthetic benchmark?


That's the closest thing I could find on Tom's to what he asked for based on what I know. Do you have a different one to suggest? If you can find a non-synthetic math based on it would be very helpful for him.
a b à CPUs
August 28, 2009 4:17:14 PM

Well, I use my i7 for MATLAB all the time, and it is less well threaded than you might think. On pure, large matrix operations, it is extremely well threaded. However, on almost any other task, it seems to be more governed by single thread speed and memory speed than anything else. Because of that, I'd recommend an i7 as the first choice, and a Phenom II as a second choice. Both have integrated memory controllers, which helps with performing many of the MATLAB operations. A Core 2 Quad, while not a bad CPU, will not perform as well as an i7 or Phenom II in this case.

Also, if you get the Phenom II, get an AM3 CPU and motherboard, and use DDR3 memory. The speed boost will actually help in matlab, and DDR3 is not as expensive as it used to be. This would be a good choice: http://www.newegg.com/Product/Product.aspx?Item=N82E168...
a b à CPUs
August 28, 2009 4:23:18 PM

cjl said:
Well, I use my i7 for MATLAB all the time, and it is less well threaded than you might think. On pure, large matrix operations, it is extremely well threaded. However, on almost any other task, it seems to be more governed by single thread speed and memory speed than anything else. Because of that, I'd recommend an i7 as the first choice, and a Phenom II as a second choice. Both have integrated memory controllers, which helps with performing many of the MATLAB operations. A Core 2 Quad, while not a bad CPU, will not perform as well as an i7 or Phenom II in this case.

Also, if you get the Phenom II, get an AM3 CPU and motherboard, and use DDR3 memory. The speed boost will actually help in matlab, and DDR3 is not as expensive as it used to be. This would be a good choice: http://www.newegg.com/Product/Product.aspx?Item=N82E168...


+1. Nicely put, cjl. Totally agree. While the I7 might have a slight advantage over because of the triple channel memory, it and PII will be very close and both are great values. Core2, however, cannot get my recommendation for serious multitasking and calculation. It's architecture is just too outdated.
a b à CPUs
August 28, 2009 5:41:44 PM

Cryslayer80 said:
+1. Nicely put, cjl. Totally agree. While the I7 might have a slight advantage over because of the triple channel memory, it and PII will be very close and both are great values. Core2, however, cannot get my recommendation for serious multitasking and calculation. It's architecture is just too outdated.

Wow, you really didn't read a word I wrote...

The Core 2 Quad is a wonderful choice for multitasking and certain calculations. MATLAB however seems to love memory bandwidth. I'm not talking about a slight difference either. Some programs are amazingly strongly affected.

One program I wrote runs nearly twice as fast on my i7 compared to my friend's Core 2 Quad (3.7GHz) system, and runs more than 1.5x the speed of a different friend's Phenom II (3.8GHz) system. Yes, the PhII beats the Core 2 Quad quite handily (even though they are roughly equal on most other tasks), but the i7 still flattens the PhII by nearly 50% for this specific task. Now, the amount of benefit an i7 will see does depend heavily on the exact kind of code, but my point was that specifically matlab loves memory bandwidth, and therefore a core 2 quad would probably not fit the OPs needs.

Don't try to twist my words into claiming that an excellent line of CPUs (the Core 2 Quad series) is terrible. I wouldn't recommend them now, but that's more due to the fact that they are on a basically dead socket, and i5 is coming out fairly soon. It has nothing to do with their performance.
a b à CPUs
August 28, 2009 5:56:44 PM

Yeah anything just not flattened it is getting boring.
a b à CPUs
August 28, 2009 6:14:03 PM

This sort of multitasking? I was unaware that running a single program (Matlab) was considered multitasking. Thank you for enlightening me.

As for you not liking flattened? OK, how about destroyed? Annihilated? Hopelessly outperformed? You claim it doesn't happen, but in some MATLAB code, there really is a 50% performance difference between the i7 and the Phenom II. That's pretty significant. The Phenom is still a good choice though, if you can't afford the i7. It will easily outperform a core 2 quad in matlab.
August 29, 2009 9:17:46 AM

cjl said:
Well, I use my i7 for MATLAB all the time, and it is less well threaded than you might think. On pure, large matrix operations, it is extremely well threaded. However, on almost any other task, it seems to be more governed by single thread speed and memory speed than anything else. Because of that, I'd recommend an i7 as the first choice, and a Phenom II as a second choice. Both have integrated memory controllers, which helps with performing many of the MATLAB operations. A Core 2 Quad, while not a bad CPU, will not perform as well as an i7 or Phenom II in this case.

Also, if you get the Phenom II, get an AM3 CPU and motherboard, and use DDR3 memory. The speed boost will actually help in matlab, and DDR3 is not as expensive as it used to be. This would be a good choice: http://www.newegg.com/Product/Product.aspx?Item=N82E168...


Guys, thanks for answer. Actually it is quite clear for me too, that i7 is one of the best performing cpu for this sort of tasks, but honestly it is well beyond the limit of my budget. And that's the same reason, why I wouldn't buy DDR3 RAM. The above two listed configurations are somewhere around my limit, and I can't and don't want to afford more.

It is true that MATLAB is well threaded in most situations, mainly for linear algebra stuff, matrix-vector operations. It is also sure, that loops are not optimized at all: if the code only contains loops without parallelized matrix-vector operations, it will use only one core.

But what is behind the parallel performance of MATLAB, is mainly the optimized library for both the intel processors and the amd processors. I know that the Intel Math Kernel Library is well-made and very well optimized, but I don't know anything about the AMD Core Math Library. Actually, it also depends on which MATLAB version one uses, but It is true that there are no reliable tests about the MATLAB performance of the newest AMD processors.

I read tests, where AMD Phenom X4 9850 performed worse in MATLAB, then a Phenom X3 8750. I also read opinions, that the AMD Core Math Library should be disabled in MATLAB, and one should switch to the Intel MKL, regardless the precessor brand, as MKL performs better.

http://ixbtlabs.com/articles3/cpu/phenom-x4-matlab-p1.h...

But this all seems to be messy,
and I am full of urban legends and fairy tales, that the AMD phenom 2 X4 _supposed_ to perform 2x..3x...4x...100x faster then Core2 Quad in MATLAB. Does anyone have real comparisons ?

In my opinion, the performance of a cpu in an application is on one hand determined by the cpu it self, and on the other hand (which is not less important) by the optimized or non-optimized code, that it runs. And this is what I don't know, how good the support is for AMD processors in such applications, how good thez run for example with applications compiled with the Intel MKL, or Intel Compiler.



August 29, 2009 4:45:27 PM

Also google on "GPU supercomputer MATLAB"

I don't do GPU supercomputing and I don't use MATLAB, but if you are putting a system together, you should look at it.