Best desktop PC / configuration / charastectics for Stata

[Francois Durant]

Hi,

I spend a lot of time working with Stata, handling sometimes medium to large datasets (4-100Go). I am about to buy a new desktop PC. I would like to know what are the important characteristics to optimise Stata performances. Is there anyone that could:
1) recommend their top 3 choices for desktop computer (already built, ready to buy on Amazon, Best Buy etc)
2) the characteristics I should Care about

thanks a lot in advance!

ps: I am located in the US (in case you want to give me specific store recommendations)

[Leonardo Guizzetti]

There is probably no "best" choice, just a "good enough that you can afford." What that means depends on how much time and money you can invest to source the hardware and software.

You don't mention what version of Stata that you currently use, or which analyses currently cost you the most amount of processing time, so I can provide some generally helpful ideas to keep in mind, in roughly the order of importance I would consider in my own work. If anyone has different thoughts, I'd like to read them.

1) Stata - upgrade to MP with at least 2 or 4 cores. Stata is generally faster for most operations with more cores, and regression modeling in general can be sped up (nearly linearly) with more cores. There's no sense in having cutting edge hardware if your license cannot use it. Incidentally, the version of MP will determine the minimum number of cores that your CPU will need, more on that below.

2) Memory - Your datasets are on the larger side, and Stata works a lot in memory. With the addition of frames in Stata 16, this is ever more true. A modern Windows laptop needs 8 GB minimum to multitask comfortably, but I would ensure I had at least 16 GB to give Stata extra room. If you have reason to think you need more memory, you can increase this to 32 or 64 GB on any modern desktop, but that might be overkill for your needs.

3) CPU - my own rule of thumb is that I like to have the number of cores that my Stata license uses + 2 as a minimum, because I want to be able to multitask while Stata is occupied. An Intel Core i7 or AMD equivalent will be helpful. A common configuration is to have 8 cores (= 4 physical cores each with 2 logical cores), so I would start here and look for the fastest clock speed I could afford. At the end of the data, the CPU does the calculation so faster is better, but not necessarily all important for you needs.

4) Hard drive - depending on how often you read/write your datasets from disk, and whether Stata needs to cache its results while computing, hard drive latency can be a huge bottleneck. Here I would get a fast solid state drive with enough space to hold your datasets that you will actively work with in a session. In a desktop computer setting, I like the idea of having a solid state disk for Windows and data operations for the speed, but having a larger capacity magnetic media hard drive for storage and archiving. You may consider at the start of your session copying the necessary datasets to a working folder on your solid state disk to work with, save the results, then clear the working folder.

The rest is just gravy. Oh, and maybe a nice, wide monitor (or two) for working with multiple windows open at once.