Matlab with MacBook Air

I'm considering purchasing a MacBook Air and just wanted to know if MatLab will run well on the Air?

Thanks, Jeff

Answer 1

[Walter Roberson]

With today's substantial updates to the Air series, they may well be worth considering.... except that you still wouldn't want to get just 2 GB of memory on the smaller one. They now have i5 processors and can be ordered with i7 processors. I have not yet taken a detailed look at them.

[Jeff]

I was sort of counting on a significant upgrade with the release of new airs. I agree that 4 GB ram is probably necessary - and it's only an additional $100. Matlab work on the air would probably only be done at night when there is nothing to do or on an airplane.

Thanks for your input.

[Jeff]

What do you know about the intel hd video?

Answer 2

[Andrew Newell]

Here are some benchmarks I obtained using bench. Note that my Mac Book Air (the last entry) has only 2GB memory.

1. Windows 7 (64-bit) 2.66 GHz Intel Core 2 Quad

4 threads, Windows 7 Enterprise, 2.66 GHz Intel Core 2 Quad, 4 GB memory, NVidia Quadro FX 1500, MATLAB 7.11.0.514 (R2010b)

2. Windows XP (32-bit) 2.66 GHz Intel Core 2

2 threads, Windows XP SP3, 2.66 GHz Intel Core 2 CPU 6700, 3.25 GB memory, NVidia Quadro FX 1500, MATLAB 7.11.0.514 (R2010b)

3. Windows Server 2003 (64-bit) 2.4 GHz AMD Opteron

8 threads, Windows Server 2003 Standard x64 Edition SP2, 2.4 GHz AMD Opteron, 32 GB memory, Rage XL PCI, MATLAB 7.11.0.514 (R2010b)

4. Windows 7 (64-bit) 2.39 GHz Intel Xeon

16 threads, Windows 7 Enterprise x64, 2.39 GHz Intel Xeon E5520, 12 GB memory, NVidia GeForce 9500 GT, MATLAB 7.11.0.514 (R2010b)

5. Windows 7 (32-bit) 1.6 GHz Intel Atom

1 thread, Windows 7, Intel Atom CPU N270 @1.60 GHz, 1.0 GB memory, Mobile Intel® 945 Express Chipset Family, MATLAB 7.11.0.514 (R2010b)

6. Mac 10.5 (64-bit) 2.66 GHz Intel Xeon

4 threads, Leopard 10.5.5, dual 2.66 GHz Intel Xeon, 4 GB memory, NVidia GeForce 7300GT, MATLAB 7.11.0.514 (R2010b)

7. Mac 10.6 (64-bit) 2.27 GHz Intel Xeon

16 threads, Snow Leopard 10.6.2, 2.37 GHz Intel Xeon E5520, 12 GB memory, NVidia GeForce 9500 GT, MATLAB 7.11.0.514 (R2010b)

8. Linux Debian (64-bit) 2.66 GHz Intel Core 2

2 threads, 2.6.26-2-amd64 kernel, 2.66 GHz Intel Core 2 CPU 6700, 4 GB memory, NVidia Quadro FX 1500, MATLAB 7.11.0.514 (R2010b)

9. Linux Ubuntu (64-bit) 2.27 GHz Intel Xeon

16 threads, 2.6.30-020630-generic kernel, 2.37 GHz Intel Xeon E5520, 12 GB memory, NVidia GeForce 9500 GT, MATLAB 7.11.0.514 (R2010b)

10. MacBookAir 10.6 (64-bit) 1.86 GHz Intel Core 2

4 threads, Snow Leopard 10.6.7, 1.86 GHz Intel Core 2, 2 GB memory, NVIDIA GeForce 9400M, MATLAB Version 7.11.0.584 (R2010b)

     LU        FFT       ODE       Sparse    2-D       3-D
1.   0.0577    0.1153    0.1641    0.2628    0.5081    0.7387
2.   0.0775    0.1132    0.1199    0.2364    0.3300    0.5703
3.   0.1042    0.1348    0.2572    0.7519    0.5510    0.4986
4.   0.0386    0.0546    0.1415    0.2079    0.4124    0.7271
5.   1.8844    0.7826    0.7091    1.5299    2.4052    1.9797
6.   0.0760    0.1770    0.1817    0.3503    0.5827    0.6359
7.   0.0465    0.0995    0.1892    0.3535    0.4546    0.6841
8.   0.0658    0.0945    0.1636    0.2567    0.2651    0.2044
9.   0.0515    0.0434    0.1727    0.2814    1.3914    1.6377
10.  0.0866    0.1767    0.2660    0.3801    1.0964    1.3405

Answer 3

[Walter Roberson]

http://www.apple.com/macbookair/compare.html

I'm afraid I don't see much point in having 1066 MHz DDR3 memory with an 800 MHz Front Side Bus.

2 GB really isn't enough memory to run any MATLAB version that is currently purchasable. On Mac, I don't think I'd be comfortable even with the maximum 4 GB you can equip the devices with.

1.4 GHz isn't very fast, even if you run both cores. The desktop system I'm using at the moment is 4 GHz single core Pentium 4, a two or three generation hand-me-down because no-one else thought it fast enough to bother with; if you are accustomed to running systems designed this century, you likely won't find 1.4 GHz an enjoyable experience for computation.

The graphics aren't anything special. The 320M integrated chipset has 32 shader cores and no dedicated memory, using 256 MB of the system memory (reducing the amount available for MATLAB). See here for some benchmarks

It is difficult to findy any official nvidea acknowledgement that the 320M exists in its CUDA information, but the evidence I found suggests that the 320M supports Cuda 1.2 capabilities (not enough for MATLAB to use, but enough for Accelereyes Jacket to use); that evidence also indicates 48 CUDA cores, not the 32 Apple officially claims.

If your question is whether MATLAB will run on the Airbook: then Yes, but I would push the 4 GB memory in ASAP.

"run well"... ummm... compared to what?

[Jeff]

Thanks for the details and information. Truly appreciated. I'm a Physicist and don't really like running slow programs. Looks like I'll need to look towards the MacBook Pro. I'm currently running it on a 3 - 4 year old IBM thinkpad. So I thought a newer Air would be able to handle matlab. Looks like it would, but not so well.

Thanks again.

[Walter Roberson]

I did some research earlier this year, _after_ the latest generation MacBook Pro; I found that it was no longer possible to buy any MacBook Pro with a graphics card that supported CUDA 2.0 or later. It _was_ possible to buy CUDA cards (one from Apple directly, one from a third party with Apple blessing) for the G5 Server. The 320M was pretty typical of the graphics capability of last year's MacBook generation; this year Apple went for completely different graphics that were not from NVIDIA at all, including one MacBook Pro that used graphics integrated in to the Intel CPU. If you look at one of the MacBook Pro or even iMAC line that has an Intel i5 or i7 in it, you probably will not be getting even CUDA 1.2 capability _or_ any ability to purchase a CUDA 2.0 compatible graphics card like used to be available.

Anyhow, CUDA is not necessarily important for you. I was looking to replace my old burnt-out Windows system and I was thinking of buying in to CUDA 2.0 capabilities so I could run SETI@Home and the like, and so that I could start studying up on the topic; what I determined was that I couldn't do it for less than $6000 in a Mac these days, but that for half that price I could get a very decent Linux system from a firm such as liquidnitrogenoverclocking.com . I'd *prefer* not to be driven back to PC hardware, but unfortunately it appears that Apple is getting more and more in to the "computer experience" and less and less in to high performance computing :(

[Jeff]

I decided to read a little about CUDA/MatLab since I really had no idea what it was. Anyway, what I gather is that this allows for some processing to be performed on the graphics card as well as the CPU. Is this basically the idea. Also, in general MatLab typically only uses the CPU and CUDA is utilized only with certain toolbaoxes, can't remember the box.

At any rate, you are correct. There are MacBook Pros that use integrated video and if I were to upgrade my purchase to a pro, I would not go for the IV. It looks like I'll be purchasing the MacBook Pro and not the Air. Kind of stinks as I was really likeing the Airs.

[Sean de Wolski]

Matlab on my Macbook pro is fine. Not all of the toolboxes are available with macintosh but unless you need DAQ or some other more obscure toolbox you should be fine.

This Macbook pro is three or four years old: 4GB, 2.5Ghz, 667MHz, DDR2 duo. I still frequently (at least weekly) dump the whole system by trying to create 9Gb matrices.

[Walter Roberson]

Yes, CUDA is the use of the graphics card (or similar architecture) to perform parallel computations. The CUDA instruction set is limited, but what it can do it can do pretty quickly. If you are interested in "high performance" computations then you probably want to investigate CUDA.

CUDA use does require special code, or at least code operating on a special class of objects; as far as I know it is not integrated in to any of the other toolboxes.

Is it possible for you to get access to the kind of Air that you would be interested in purchasing? With access and a MATLAB trial license, you could see for yourself whether performance was satisfactory for your purposes.

5 hours or 7 hours for the battery, though... wonder if that includes the situation where you are doing heavy computing on it?

[John Melonakos]

Walter is correct that Jacket supports these Macbook Air computers with the 320M GPU. Good luck!

Answer 4

[jl]

I am interested in MacAir 13" with Matlab as well. If anyone has done a Bench on these, please post it up.

As for the memory, 4G is marginal but having said that, the SSD disk system should compensate greatly with any swapping.

thanks.

Answer 5

[Andrew Newell]

Some follow-up thoughts: I think the most important comment that Walter made was "compared to what?" To which I would add, "For what purpose?"

It seems a bit silly to compare this laptop to industrial-grade desktops. Most people considering a Mac Book Air are not in the market for 4GHz hand-me-downs, let alone the current top models. The Mac Book Air should be compared to other laptops.

Similarly, there are a lot of uses of MATLAB that don't require blazing speeds or even 4GB memory. I use the Air when I'm traveling, and most of the time I don't miss the extra power. Balance that against the ease of carrying it around.

One major caveat: Very few hotels have good wireless support, and this caused me no end of trouble for a while. Fortunately, there is a cheap solution. If you are traveling, get an Ethernet-to-USB adapter!!!!

Answer 6

[jl]

Hi Andrew, Thanks very much for the results! They are better than I expected, so the just released Macbook Air should blitz it then.

Answer 7

[Jeff]

As a followup to this chain, I purchased the Air with 4 GB Ram this weekend, installed MS Office 2008 and MatLab R2011a. Everythings works well. In fact, the startup time for MatLab is significantly faster with my Air than it is with my desktop PC. The Bench script does indiciate that it is at the bottom, second from the bottom to be accurate, in terms of running times. However, for what I am doing I am seeing anything significant yet.

Thanks for everyone's input. I appreciate your help.

Jeff

[Walter Roberson]

Did you get the i5 or i7? Which front-side bus?

[Jeff]

I went cheap and bought the 1.7 GHz i5 processor with 4 GB RAM. My mistake was getting the 128 GB hard drive. However, I will be taking it back this afternoon for the 250 GB version.

Answer 8

[Jeff]

I chose to cast two votes. Each was helpful in their own way.