Does general stata use take advantage of GPU power?

[Matt Manierre]

I'm in the process of budgeting for a new Stata machine and I'm wondering if anyone knows if I should bother with getting a high end GPU. I don't use mapping software very often, so the only motivation I have for investing in this would be if Stata's performance would be boosted by it. Anyone know if Stata actually benefits from this when doing SEM, HLM, or similar models?

Other than that, I'm assuming the best things to invest in for Stata performance are 1) CPU; especially if using MP 2) 8gbs+ of ram 3) an SSD for active projects.

[Friedrich Huebler]

It is not clear which "mapping software" you are referring to but try the test described in this Statalist message: http://www.stata.com/statalist/archi.../msg00362.html

[Ariel Karlinsky]

stata loads the active project to RAM regardless of size and speed of Hard Drive/SSD. Also, seeing as how RAM is very very cheap I would put much more than 8G. 16 is a "nice" minimum and in some cases 32 is probably not overkill.

[Matt Manierre]

Friedrich- I was primarily thinking of something like ArcMap, which are very GPU intensive. I know stata has some mapping features but I don't intend to use them in the near future.

Ariel- Thanks for that advice, I'll make RAM the top priority over an SSD.

[wbuchanan]

Which OS are you using? The SSD will definitely help if you find yourself doing lots of I/O operations (e.g., creating a bunch of different files that get stored on the hard drive). The more RAM you can get the better since it will also allow the rest of the computer to function normally without having to swap if you are doing something memory intensive (e.g., you can watch your favorite cat videos on YouTube while fitting models to a large volume of data without as much performance degradation). The processor speed and number of processors is next. That all said, while Stata doesn't directly leverage GPU processing at this time - to my knowledge - I imagine it could provide some performance gains when rendering graphs and/or would make it possible to use the Java/C APIs to write code that leverages the GPUs.

[Matt Manierre]

Which OS are you using? The SSD will definitely help if you find yourself doing lots of I/O operations (e.g., creating a bunch of different files that get stored on the hard drive). The more RAM you can get the better since it will also allow the rest of the computer to function normally without having to swap if you are doing something memory intensive (e.g., you can watch your favorite cat videos on YouTube while fitting models to a large volume of data without as much performance degradation). The processor speed and number of processors is next. That all said, while Stata doesn't directly leverage GPU processing at this time - to my knowledge - I imagine it could provide some performance gains when rendering graphs and/or would make it possible to use the Java/C APIs to write code that leverages the GPUs.

I'd be using 64 bit windows 10. By the sound of it I'll be beefing up the RAM quite a bit- cat videos during bootstrapping is a time honored part of my research process. Thanks for the input!

[wbuchanan]

In my experience with Windows more generally it tends to be fairly miserly with releasing memory and seems like it wants to cache any/everything possible. In that case, if you can get something with 64/128GB RAM you'll definitely make sure to be able to handle fairly large datasets and should rarely be at a lack for RAM (unless of course you are editing the cat videos while bootstrapping). A while ago I added an SSD to an older MacBook pro that I was using regularly for work and noticed a decent performance boost, but definitely invest in lots of fast RAM and a speedy CPU. If you're able to budget for Stata MP, I'd also recommend that as well. In the few times I've used a single core version of Stata I can't imagine going back.

[Matt Manierre]

Thank you wbuchanan. I will contemplate adding stata MP to my wish list as well, though it is obviously pretty costly compared to SE. Then again, if I get a i7 with 12 threads I might as well take advantage of the whole thing.