Report Builder was designed to support a theoretically unlimited file size, because it doesn't load any of the data records into memory (a.k.a. RAM). Instead it just scans through the records on disk every time it needs to compute something. These disk scans are probably the most expensive operation when computing results - especially if your data file is on a network drive or behind a VPN.
RAM is a temporary storage on your computer that is usually much faster than a hard drive. So, if your dataset will fit in RAM, you'll probably experience much faster computation times by putting it there instead of a hard drive. (Newer solid state hard drives (SSD) are much faster than the traditional spinning disk hard drives, so you might not experience as much of a performance gain using RAM vs. a solid state drive).
When you click "Load Dataset into RAM", Report Builder will read the records from your dataset and place them into this faster temporary memory (RAM). When you close a project, Report Builder will release this data from RAM and your operating system will recycle it when needed.
Be careful not to load too much data into RAM. You'll just have to experiment on what that amount is. But here's why: when Report Builder needs to use RAM, it requests RAM from the operating system. The operating system is designed to pretty much always say "yes, you can use that RAM" even when it is running thin or running out of RAM. If the operating system truly runs out of RAM, it will start using the hard drive as RAM and things will get REALLY slow. Your computer may become slower and slower and eventually reach a screeching halt. So please save your work before loading into RAM and use your best judgement. I didn't want to cap the amount of RAM people can use, because some people might appreciate being able to use a huge amount. A possible future improvement would be a setting which allows the user to specify a cap, but perhaps that will come after some experimentation on a good default number.