Hardware coice for high performance in Mathematica

[joao9_49]

Hello,

I am planning to buy a laptop and my main concern is to get one that allows the numerical and symbolic package Mathematica to run with high performance.

I've seen in previous discussions that SSD is important for computer speed but in a Mathematica forum, that was not considered to be essential. The reason invoked was that Mathematica does not need to write on disk very often while running the kernel.

Do you have any ideas on this that could share with me? What hardware specifications should I be paying attention to?

My best regards
Joao

 

[MauveCloud]

I expect Mathematica would benefit from multiple processor cores and large amounts of memory, but wouldn't be affected so much by the storage or video card (though it might be able to use gpgpu functionality by now, so you might research that).

 

[Fulgurant]

Hello,

I am planning to buy a laptop and my main concern is to get one that allows the numerical and symbolic package Mathematica to run with high performance.

I've seen in previous discussions that SSD is important for computer speed but in a Mathematica forum, that was not considered to be essential. The reason invoked was that Mathematica does not need to write on disk very often while running the kernel.

Do you have any ideas on this that could share with me? What hardware specifications should I be paying attention to?

My best regards
Joao

Not familiar with the program, but you're almost certainly looking for CPU muscle -- a Core i7 machine. High memory is always nice too.

Look for a Core i7 three-thousand series (anything with a 3 as the first digit, which means third-generation Core, or an Ivy Bridge processor), and 8+ GB RAM (preferably 16). The integrated graphics on the CPU should suffice for your purposes; and if you don't need a dedicated GPU, the laptop ought to be reasonably priced even with a top-end CPU. Everything else should take care of itself.

 

[Fulgurant]

I expect Mathematica would benefit from multiple processor cores and large amounts of memory, but wouldn't be affected so much by the storage or video card (though it might be able to use gpgpu functionality by now, so you might research that).

Good point. According to the Wiki article, Mathematica does support CUDA and OpenCL as of 2010. I'm not sure how cost effective it'd be to try to exploit that in a laptop though.

 

[getochkn]

It does support GPU acceleration

"To use Mathematica's built-in GPU computing capabilities, you'll need a graphics card that supports OpenCL or CUDA, such as many cards from NVIDIA, AMD, and others."


So a good video card would probably help a lot with complex programming.

http://reference.wolfram.com/mathematica/guide/GPUComputing.html

Some of the GPU enabled features.

 

[Fulgurant]

It does support GPU acceleration

"To use Mathematica's built-in GPU computing capabilities, you'll need a graphics card that supports OpenCL or CUDA, such as many cards from NVIDIA, AMD, and others."


So a good video card would probably help a lot with complex programming.

http://reference.wolfram.com/mathematica/guide/GPUComputing.html

Some of the GPU enabled features.

Right, so then the OP would be looking for a Radeon of some sort, because nVidia's (current) consumer-grade video cards are notoriously inaccurate (at compute task)? Dunno if that translates to the laptop space.

 

[jaguarskx]

As mentioned above, the best CPU to go for it a quad core Intel "Ivy Bridge" Core i7 CPU; they have the "QM" designation. Any mobile Intel CPUs with the "M" designation are simply dual core CPUs.

There are basically 4 different model groups of the quad core i7 QM series (I included the "XM" as well; Extreme Mobile). Each group basically has slightly increased clockspeed, amount of cache and prices:

i7-39xx XM = Very expensive (Over $1,000 just for the CPU itself), but best performing CPU with 8MB of cache. Needless to say any laptop with this CPU will easily cost you at least $2,200+.

i7-38xx QM = Just 200MHz slower than the i7-39xx and has a more reasonable price of around $550 for the CPU itself.

i7-37xx QM = Just 100MHz slower than the i7-38xx with only 6MB of cache, but price lower at around $400 for the CPU.

i7-36xx QM = These quad core CPUs are generally only sold to system builders and are the most affordable Core i7 CPUs. They are slightly slower (by 100MHz to 300MHz depending on the model) than the i7-37xx series, but they have 6MB of cache.

My advise is to simply buy a laptop with an i7-3615QM which is the most powerful of the "inexpensive" Intel mobile quad core CPUs. Of course, if money is not an object then look for a laptop with at least an i7-37xx QM CPU, but just realize that the more performance you want, the more you have to be willing to pay. For example:

The i7-3720QM is clocked at 2.6GHz (Turbo Boost to 3.4GHz / 3.5GHz / 3.6GHz) and has 6MB of cache. The i7-3920XM is clocked at 2.9GHz (Turbo Boost to 3.5GHz / 3.6GHz / 3.7GHz) and 8MB of cache. Is it worth spending around $700 more for the i7-3920XM? That depends on how much performance you are willing to spend money on. $700 can almost by two i7-3720QM CPUs.

 

[Fulgurant]

Excellent breakdown by jaguarskx.

It might be best if the OP finds a couple of models that fit his price range and we can judge them. Failing that, a price range from the OP would be very helpful. It's kinda hard to discuss this stuff meaningfully in the abstract.

EDIT: As a sidenote, I would urge the OP to think very carefully about how much he/she really needs a mobile platform. If performance is the primary goal, and the OP plans to do all of his serious work in a stationary location, then a desktop will perform better, more reliably, and cost significantly less. If the OP truly needs mobility then more power to him; some people do.

I just think it's worth emphasizing that most people really don't. Tying yourself to a notebook just because you kinda-sorta like the idea of computing at the local coffee shop isn't a good idea, IMO.

 

[joao9]

Hi, thanks very much for your comments. The nature of the issues involved are clearer now.

best regards

Joao