How much memory does your computer need to run Stata for Windows?

The amount of memory Stata needs to run efficiently is completely dependent on the size of the datasets you need to load. You can actually make Stata run more slowly by allocating too much memory to it.

If the total memory allocated to all programs (including the OS itself) on your computer is greater than the total RAM you have, some of the memory will have to be represented on disk as virtual memory. Anytime virtual memory (the part of memory represented on disk) has to be accessed, the computer will slow to a crawl.

Some users think if giving Stata 40 MB is good, giving Stata 60 MB or even 80 MB must be better. This is not true. Once enough memory has been allocated to Stata to allow it to load the current dataset, with enough extra room for whatever temporary variables, programs, macros, etc., might be needed during analysis, no speed improvements will be seen by giving even more memory to Stata. In fact, you may hurt Stata’s performance by forcing the OS to use virtual memory if you allocate too much memory to Stata.

To understand why, think of memory as a rectangle on a hypothetical computer. On this computer, imagine Stata is the only application running and the OS uses 0 memory. Give this computer 32 MB of real RAM, and allocate 40 MB to Stata:

        +----------------+
        |                | <-- 32 MB real RAM
        |                |
        |                |
        |                |
        |                |
        |                |
        |                |
        |                |
        |                |
        +----------------+
        |                | <--  8 MB virtual memory
        |                |
        +----------------+
   40 MB total allocated to Stata

Let us imagine that the dataset with which you want to work is only 20 MB in size. You might think that, because the dataset is smaller than the amount of real RAM on the computer, you do not have to worry about virtual memory or swapping. Not so! The dataset is stored as a smaller rectangle within the larger one. Each observation of the dataset can be looked at as “row” in memory. Depending on the shape of the rectangle of data, some of the observations of the data may extend into the virtual memory region:

        +----------------+
        |########        | <-- 32 MB real RAM
        |########        |
        |########        |
        |########        |
        |########        |
        |########        |
        |########        |
        |########        |
        |########        |
        +----------------+
        |########        | <--  8 MB virtual memory
        |########        |
        +----------------+
   40 MB total allocated to Stata

Anytime those observations are accessed, the OS will have to use the virtual memory, swapping it into real RAM and slowing computation to a crawl.

Because the dataset is only 20 MB in size, we do not need to allocate 40 MB to Stata. We could probably allocate 25 MB or even be generous and give 30 MB to Stata. Either would avoid swapping and should provide enough room for Stata to store temporary variables, macros, programs, etc.

        +----------------+
        |#############   | <-- 32 MB real RAM
        |#############   |
        |#############   |
        |#############   |
        |#############   |
        |#############   |
        |#############   |
        |#############   | <-- ### represents 20 MB of data within 30 MB
        |                |     allocated to Stata
        |                |
        +----------------+

Rather than spending time trying to calculate the exact amount of memory to give to Stata, it is usually easier to just experiment a little. If you are working with 20 MB datasets, give Stata 25 MB. If you get a “no room to ...” error message, you know you need to give Stata a little more; try 30 MB. Do not give Stata 60 or 80 MB—this is overkill and can only lead to possible use of virtual memory and slow performance.