upgrade RAM vs CPU
in [Matlab]
|
From: joshua on 9 Aug 2010 09:38 i use matlab a lot, and i run a mac. the new mac pro's came out today, so i am debating where to allocate my resources (ie, RAM, SSD, CPU, cores, 64-bit matlab, etc.). in nearly all my scripts, the most time consuming operations are: .*, exp, ln, sqrt, etc. essentially, element-wise operations. the size of my arrays tends to be about 100x100 doubles. my plan is to go to an apple store, install matlab on a bunch of different mac pro's with different specs, and see which things seem to make a difference. but, perhaps somebody here knows what i could expect. i imagine since my operations are all element-wise, multicores would help. but, the arrays are not that large, so the overhead might make it not worth it. any insight on these matters would be much appreciated.
From: Andy on 9 Aug 2010 09:49 "joshua " <jv.work(a)jhu.edu> wrote in message <i3p0bs$s5r$1(a)fred.mathworks.com>... > i use matlab a lot, and i run a mac. the new mac pro's came out today, so i am debating where to allocate my resources (ie, RAM, SSD, CPU, cores, 64-bit matlab, etc.). in nearly all my scripts, the most time consuming operations are: .*, exp, ln, sqrt, etc. essentially, element-wise operations. the size of my arrays tends to be about 100x100 doubles. > > my plan is to go to an apple store, install matlab on a bunch of different mac pro's with different specs, and see which things seem to make a difference. but, perhaps somebody here knows what i could expect. > > i imagine since my operations are all element-wise, multicores would help. but, the arrays are not that large, so the overhead might make it not worth it. > > any insight on these matters would be much appreciated. 1. What are the specs of your current machine? 2. What is the budget for the new machine? 3. Are you absolutely set on a Mac? 4. Would you consider building the machine yourself?
From: Sean on 9 Aug 2010 09:56 "joshua " <jv.work(a)jhu.edu> wrote in message <i3p0bs$s5r$1(a)fred.mathworks.com>... > i use matlab a lot, and i run a mac. the new mac pro's came out today, so i am debating where to allocate my resources (ie, RAM, SSD, CPU, cores, 64-bit matlab, etc.). in nearly all my scripts, the most time consuming operations are: .*, exp, ln, sqrt, etc. essentially, element-wise operations. the size of my arrays tends to be about 100x100 doubles. > > my plan is to go to an apple store, install matlab on a bunch of different mac pro's with different specs, and see which things seem to make a difference. but, perhaps somebody here knows what i could expect. > > i imagine since my operations are all element-wise, multicores would help. but, the arrays are not that large, so the overhead might make it not worth it. > > any insight on these matters would be much appreciated. Well a 100x100 double is 80000 bytes which is nil so it doesn't sound like RAM is your limiting factor. I would think a faster processor would be your best bet. Also, I would HIGHLY doubt the Apple Store will let you load any executable file on their computers. You probably would need an institutional license to load MATLAB on these computers anyway and I HIGHLY doubt they want you adding a head count to their license for the 3 minutes you're going to use it. I'm basically saying don't waste your time at the Apple Store.
From: joshua on 9 Aug 2010 11:04 "Andy " <myfakeemailaddress(a)gmail.com> wrote in message <i3p10g$acq$1(a)fred.mathworks.com>... > "joshua " <jv.work(a)jhu.edu> wrote in message <i3p0bs$s5r$1(a)fred.mathworks.com>... > > i use matlab a lot, and i run a mac. the new mac pro's came out today, so i am debating where to allocate my resources (ie, RAM, SSD, CPU, cores, 64-bit matlab, etc.). in nearly all my scripts, the most time consuming operations are: .*, exp, ln, sqrt, etc. essentially, element-wise operations. the size of my arrays tends to be about 100x100 doubles. > > > > my plan is to go to an apple store, install matlab on a bunch of different mac pro's with different specs, and see which things seem to make a difference. but, perhaps somebody here knows what i could expect. > > > > i imagine since my operations are all element-wise, multicores would help. but, the arrays are not that large, so the overhead might make it not worth it. > > > > any insight on these matters would be much appreciated. > > 1. What are the specs of your current machine? macbook pro, early 2008, 2.4 GHz Intel Core 2 Duo, 4 GB 667 MHz DDR2 SDRAM osx 10.5.8, MATLAB Version 7.6.0.324 (R2008a) > 2. What is the budget for the new machine? not well defined. big differences in processing power are worth money. 3-4k would be nice for desktop, i've got displays and storage already. > 3. Are you absolutely set on a Mac? i guess not absolutely. i like mac for many reasons, i can't handle windows anymore. i guess i could handle some linux distro, if i could get the same speed for <<$. i haven't looked much at speed comps of macs vs. linux, although i know my mac runs matlab MUCH slower than my windows boxes. > 4. Would you consider building the machine yourself? not really, i have enough cash and not enough need for speed that i don't think i'd save enough to make it worth my while. i don't need the fastest thing in the world, i just want something pretty fast that will work ALL the time. i am happy to pay to not have to deal with troubleshooting hardware compatibility issues (i used to build my own machines, though not so well, i guess).
From: joshua on 9 Aug 2010 11:05
> Well a 100x100 double is 80000 bytes which is nil so it doesn't sound like RAM is your > limiting factor. I would think a faster processor would be your best bet. Also, I would > HIGHLY doubt the Apple Store will let you load any executable file on their computers. > You probably would need an institutional license to load MATLAB on these computers > anyway and I HIGHLY doubt they want you adding a head count to their license for the 3 > minutes you're going to use it. I'm basically saying don't waste your time at the Apple Store. i thought i could install a trial version. maybe i'll call them first.... |