A good computer for theoretical physics

[urbana]

I am an undergraduate physics major preparing to enter my senior year. My college requires all physics students to complete a significant and original piece of research; mine will be an examination of the antiferromagnetic Potts model using the multiple histogram method. Much of this and the work I plan to do in the future heavily involves programming and simulation, so this summer I was planning to buy a laptop computer for portability.

What computer makes/models are best for this purpose? What sorts of things should I look for in a computer?

 

[Vid]

Fastest processor you can afford. If know how to program in threads to take advantage of multicore processors then that would be your best bet. If not, get the fastest single core you can. Also, if you plan on dealing with a large amount of data you'll want a decent amount of ram which shouldn't be a problem since laptops with over 2gb of memory are fairly common. Also, a lot of laptops come with 5400rpm harddrives which could bottleneck when reading/writing to/from ram so you'll probably want 7200rpm HDD. That shouldn't be a big deal as those are becoming more common aswell; it's just something you'll want to pay attention to. Other than that, the graphics card won't make a difference unless you plan on using vista. Although, if you're using this as a serious research laptop, you'll probably want to run XP or linux. Or if you've got a lot of money to burn and you're into that kind of thing, you can get a mac.

A good place to start shopping around:

http://www.newegg.com/Store/SubCategory.aspx?SubCategory=32&name=Laptops-Notebooks

Also, many manufacturers let you customize laptops on their websites so that might be something to look into aswell.

 

[j93]

Cant you SSH into a more powerful computer?
I would think its one of those situations where you could use pretty much any computer with linux/windows on small sets to test out analysis and then run on a school computer. I don't see why you would try to run huge sets on your personal computer especially a laptop, I am pretty sure others agree with me or I would know more graduate students who have done so.

 

[urbana]

SSH'ing to a larger, linux-based campus computer is exactly what I had planned on doing, but downloading an open source feature such as cygwin to try out smaller programs is more what I wanted a notebook for -- that's what my academic advisor does and it seems to work quite well for him. I was looking at some of the lower-end ThinkPad notebooks from Lenovo (which seem to work better than some other companies' high-end notebooks) and they appear to have the faster processors and larger memories. They also offer a professional downgrade to XP and official support for Linux, which I certainly welcome.

Thank you both for your feedback.

 

[j93]

For your academic purposes you could use a laptop with the power of a 2003 Dell Inspiron 1000(ie not much required). It sounds to me you don't need to worry about the specifics of your laptop for academic purposes.
I would just buy whatever laptop you want based on your other needs , touchscreen for notes/diagrams maybe, or faster processing for computer games, bigger harddrive for entertainment. Dual-Booting Linux shouldn't be complicated to set up on your own. Partition space for it and Linux Setup should take care of rest.

 

[Vid]

I would still recommend a 7200rpm drive over a 5400. I think Lenovo uses mostly 5400, but I'm not certain.

 

[ice109]

I would still recommend a 7200rpm drive over a 5400. I think Lenovo uses mostly 5400, but I'm not certain.

for what? all the video editing he'll be doing? i don't know of many times when a physicist will be seeking bigger files than can be cached in processor, let alone memory.

 

[robphy]

Do you have any sense of the requirements of your computational task?
How memory-intensive, cpu-intensive, disk-intensive, etc... is it?

How much are you willing to spend?
And how long do you intend to use it before buying a new one to replace it?
How portable do want it to be? Screensize? Weight? Battery life?

I would certainly look for a fast dual-core processor with a large cache...
and I would get 4GB of RAM (from Crucial, if necessary). Although XP can't use all 4GB, you can set up a ramdrive using the portion it can't see.

I've been happy with Dell and Fujitsus.
I'd also look into TabletPCs... which I use for doing derivations and calculations, as well as for lecturing.

 

[diazona]

I would still recommend a 7200rpm drive over a 5400. I think Lenovo uses mostly 5400, but I'm not certain.

Actually I would say that's one of the things that matters least. Scientific computing is mostly done with data stored in RAM; in fact in many cases, nothing other than the initial data and final results (and the code of course) ever makes it on to the disk. Of course, to some extent it depends on the specific computation...

 

[Vid]

It's not just about the actual computation, but the difference between a laptop with a 7200 and 5400 drive for general use is quite significant. The difference from just boot times up is very significant. In general, it just makes windows more responsive and programs open faster. I'm not even talking about differences within benchmark utilities. It's a very noticeable improvement.

 

[j93]

Actually I would say that's one of the things that matters least. Scientific computing is mostly done with data stored in RAM; in fact in many cases, nothing other than the initial data and final results (and the code of course) ever makes it on to the disk. Of course, to some extent it depends on the specific computation...

I agree with diazona. It seems
needs for physics (subset of) needs for general use
Therefore your looking don't need to worry at all about what you need for physics.

The question really what laptop do you want for everyday use?

 

[AUMathTutor]

Lots of memory is a must. Get as much as your computer can handle.

If you don't know how to program in threads, your money would be better spend getting as fast a single-core processor as money can buy.

How much data will you have in main memory at a time? If the quantity is large, a faster disk (7200 RPM vs 5400 RPM) will increase performance appreciably *regardless of how much you write to the disk*. If your memory needs are not too large, and you do your I/O in I/O-only chunks and not in bits and pieces throughout the CPU-bound code, disk speed won't be as big an issue.

Probably the best advice I can give you is this: keep the computer clean of junk. You don't need 12 IDEs, 11 virtual computers, 10 office suites, 9 calculators, 8 CAD, 7 video games, 6 peripheral managers, 5 freeware system utilities, 4 web browsers, 3 command shells, 2 antivirus, and a google toolbar in a pear tree. This is your work computer. Keep everything organized and only put on the machine what you will need to do your job. Having 15 processes running in the background and 75% of the disk full up of junk isn't going to make your work any easier.

 

[robphy]

Lots of memory is a must. Get as much as your computer can handle.

If you don't know how to program in threads, your money would be better spend getting as fast a single-core processor as money can buy.

It seems practically every current-model laptop (not including netbooks, etc...) is at least dual-core. You might save a few dollars with an older model with a single-core and slower and smaller-L2-cache processor... but probably not much.

One advantage of a dual-core processor is the ability to do other things while the main process is running... like using an editor or other software to study any output logged to files. You could also start another instant of the process... e.g. a run with a different set of input parameters.

 

[nbo10]

Question. Can you even buy single-core processor computer these days?

 

[AUMathTutor]

Well, I guess what I meant to say was this: that if you're not going to be using threads, go for fewer cores and more clock speed per core than more cores.

 

[martin_blckrs]

It really depends on what kind of computations you will be performing. If you are dealing with large data sets (e.g. big lattices with many atoms in solid state), you will need a lot of RAM. Note however that in order to use more than 3GB RAM you'll need a 64-bit operating system.

Another thing is the speed of processors and the number of cores. Using more than one core requires you to know how to write programs using multiple cores, which might take some time to learn.

But remember that some computations can require extensive computing power, which you might not be able to obtain with a laptop, so before buying a $5000 notebook, ask your advisor how demanding the computations actually are and also look for computing resources at your university/ local computing center.

 

[Topher925]

For your academic purposes you could use a laptop with the power of a 2003 Dell Inspiron 1000(ie not much required). It sounds to me you don't need to worry about the specifics of your laptop for academic purposes.

I agree (except for the Dell part, get a IBM T42). To me it doesn't sound like you need anything fancy. And when you do get to the stuff that requires a lot of power, laptops are next to worthless anyway. Just get something cheap, reliable, light, and with good battery life.

Personally, I highly recommend the Samsung X460 and X360 series. I have the X460 and it is the nicest machine I have ever witnessed.

 

[AUMathTutor]

In reality, I think this endeavor is a little silly. You should really not be running any serious scientific projects on a laptop... small, quick ones, alright. For developing and debugging code on small test sets, alright. But when you do actual data runs, you have no business running that on a laptop, nay, even a custom desktop. Your school undoubtedly has computing resources far in excess of anything you could muster for $20,000.

I'd think any computer would do if you remote-connect into the school's network and do the development and execution remotely.

 

[chroot]

Honestly? The days when you needed to worry about details like front-side bus speed, memory throughput and L2 cache size are over. Personal computers have become very commoditized in the last decade. Processors and memory systems for personal computers generally far exceed users' needs.

I would suggest that you worry more about the human-machine interface than the processor. If the laptop is lightweight, has a big, bright, clear display, a responsive, easy-to-use touchpad, well-placed ports, and so on, you'll probably enjoy using it, even if it's 10% slower at matrix inversion than some other clunky beast.

If you really need the highest possible performance, you should look at benchmark figures for various models. (But, first, consider which benchmark test is most similar to the kind of number-crunching you're going to do.) It's generally impossible to guess which computer is faster at which task by looking only at a few numbers like processor speed or hard-drive rpm. There are just too many variables. That's why benchmarks were created.

- Warren

 

[Wellesley]

Question. Can you even buy single-core processor computer these days?

You'd be surprised at what you can find used. I have a Pentium III 1 Ghz, running XP...not my main computer, but it surfs the web easily enough.

 

[robphy]

Honestly? The days when you needed to worry about details like front-side bus speed, memory throughput and L2 cache size are over. Personal computers have become very commoditized in the last decade. Processors and memory systems for personal computers generally far exceed users' needs.

I would agree with your statement... for the typical [novice] user.
However, when customizing features [like processor and ram, etc...],
one might want to weigh the options of choosing an upgrade vs. the additional cost of that upgrade. (The best bang for the buck is usually the second-to-the-top choice.)

When I got my recent laptop, I wanted a lightweight portable workstation (for what I wanted it to do).
For some disk-intensive tasks, I'd get my older desktop to do it... even though my laptop is much speedier than my desktop.

 

[j93]

I agree (except for the Dell part, get a IBM T42).

It was an exagerration that you could use any current laptop and even bottom of the line laptops of a few years ago to fit his academic needs

In reality, I think this endeavor is a little silly. You should really not be running any serious scientific projects on a laptop... small, quick ones, alright. For developing and debugging code on small test sets, alright. But when you do actual data runs, you have no business running that on a laptop, nay, even a custom desktop. Your school undoubtedly has computing resources far in excess of anything you could muster for $20,000.

I'd think any computer would do if you remote-connect into the school's network and do the development and execution remotely.

This is what I have been saying all along. Buy based on your own non-academic needs.