Volatile vs. non-volatile memory: When the power goes off, what survives?

Volatility, the article stresses, only describes data behavior at power loss. It does not by itself indicate speed, maximum capacity or the job a given memory performs.

The two main forms of volatile memory are DRAM and SRAM. DRAM is the working memory behind laptop or desktop specs of 16GB or 32GB of RAM. Desktop systems commonly use DDR5 DRAM, while phones and mobile devices typically use low-power LPDDR5X. Each DRAM cell holds a bit in a capacitor and transistor; the charge leaks continuously, so DRAM must refresh itself while running. When the system loses power, the contents disappear. DRAM is not always packaged as removable sticks: AMD's Ryzen AI Max chips, for example, can work with up to 128GB of unified memory.

SRAM is also volatile but does not need those refresh cycles as long as power is available. Because it is built from latching circuitry that uses several transistors per bit, it can deliver faster access than DRAM and is well-suited to processor caches. Engadget points to the Ryzen 9 9950X3D, reviewed by the outlet, as an unusually cache-heavy chip with 144MB of combined L2 and L3 cache — a figure still tiny next to the gigabytes of DRAM in main memory.

For non-volatile memory, the most familiar kind in consumer hardware is NAND flash, used in SSDs, tablets, USB drives and memory cards. An SSD contains flash memory for data retention and a controller that handles reading and writing, corrects errors and spreads writes over many cells so that no small group wears out far faster than the rest. The same design applies to SATA and NVMe SSDs.

NOR flash is another non-volatile type, typically used for code rather than large files. It can store the firmware that boots hardware and loads an operating system, and is particularly useful in embedded devices because a processor can often execute code directly from the NOR chip without copying it to RAM first.

The article notes that NAND and NOR show "non-volatile" is a property rather than a single class of component, just as DRAM and SRAM are both volatile but serve very different purposes.

Modern devices constantly move data between the two kinds of memory. When someone opens a photo-editing app, the software and the saved image start out on non-volatile storage; the system then loads needed code and data into DRAM, and the processor cache keeps frequently accessed information closer to the CPU. The temporary copies do not need to survive a normal shutdown, the article explains — only saved changes must eventually reach non-volatile storage.

Hibernation is a clear example. Windows writes the contents of volatile memory to a hibernation file on non-volatile storage before DRAM is powered down, then reads that file on resume to restore the session. Sleep mode is different: the session stays in RAM and is lost if power cuts out. This is why a laptop specification can list "16GB memory" alongside a "512GB SSD."