Memory vs Storage: What's the Difference? (RAM vs SSD)

In everyday life: two "GB" numbers on a spec sheet

Spec sheets for phones and computers list two numbers in GB, such as "memory 8 GB" and "storage 256 GB." Because the unit is the same, they look like amounts of the same thing. In fact, they are the sizes of two completely different parts. When a store asks "256 GB or 512 GB?", it usually means storage.

A GB (gigabyte) is a unit for amounts of data. One byte can hold one English letter. Giga (G) means a billion, so 1 GB is about a billion bytes. There are two ways to count it exactly, which we cover later.

Memory holds what you are using right now

A CPU is the part that takes a program's instructions one at a time and carries them out ("What Is a CPU?"). While the CPU works, its instructions and data sit in memory. Memory is also called "main memory" or RAM.

When you launch an app, its contents are loaded from storage, explained next, into memory. The CPU then works on the data in memory. The web pages open in your browser and the document you are editing also sit in memory. The more memory there is, the more apps and data it can hold at once.

Memory today almost always uses a kind of semiconductor chip called DRAM (pronounced "dee-ram"). DRAM can be read and written very quickly. In exchange, it loses its contents when the power goes off.

Losing contents when the power goes off is called being "volatile." This is why the unsaved part of a document is lost if the power suddenly cuts out. "Saving" means copying data from memory into storage.

Storage is where things are kept

Storage keeps photos, videos, apps and documents, even when the power is off. Keeping data without power is called being "nonvolatile."

Phone storage uses a semiconductor chip called NAND flash memory. Below, we simply call it "flash memory." USB flash drives and SD cards use the same kind of chip.

Most computers today also have storage built from flash memory. This is called an SSD (solid-state drive). Storage holds far more than memory, and it costs less for the same capacity. In exchange, it is much slower to read and write.

Why split them? Speed, capacity and cost

Imagine an ideal device that is fast, large, cheap and keeps data without power. With it, there would be no need to split memory and storage. But with today's technology, faster kinds cost more per byte, and larger, cheaper kinds are slower.

So computers stack several kinds of storage devices with different traits. Levels closer to the CPU are faster and smaller; levels farther away are slower and larger. This order is called the "memory hierarchy." The numbers on the right of the figure show roughly how long one read takes. A nanosecond (ns) is a billionth of a second, and a millisecond (ms) is a thousandth of a second.

At the top, registers are tiny storage spots inside the CPU that hold the numbers being calculated. There are only a few of them, but the CPU can use them without waiting.

Next comes the cache. It keeps copies of often-used parts of memory right next to the CPU. Programs tend to reuse the same data within a short time, so this trick works well.

At the bottom, an HDD (hard disk drive) records data magnetically on spinning disks. It must move its reading part and wait for the right spot to spin around, so it takes time. HDDs are still used for keeping data, because they offer large capacity cheaply.

How it works ①: DRAM is a tiny container of charge

Now let's look at how the chips actually hold data. Computer data is written as strings of 0s and 1s. One 0-or-1 is called a "bit," and 8 bits make a byte.

DRAM stores one bit with one tiny switch and one tiny container that holds electric charge. The switch is a transistor ("What Is a Transistor?"). The container is a part called a "capacitor." If the container holds charge, it means 1; if it is empty, it means 0. Robert Dennard of IBM came up with this design in 1966.

One set that stores one bit is called a "cell." A DRAM cell has only two parts, so it can be made very small and packed tightly on a chip. 8 GB of memory has about 69 billion cells.

The container, however, slowly leaks its charge. Left alone, a container holding a 1 can no longer be told apart from a 0. So DRAM regularly reads each cell and writes the same value back, before the contents fade. This is called refresh.

In PC memory, every cell is rewritten once every few tens of milliseconds. At normal temperatures, that is every 64 ms for the DDR4 standard and every 32 ms for the newer DDR5. The "D" in DRAM stands for "dynamic," meaning it needs this constant upkeep.

When the power goes off, refresh stops too. The leaked charge does not come back, so the contents soon disappear. This is why memory is volatile.

How it works ②: flash memory traps electrons in a pocket

A flash memory cell is a single, slightly unusual transistor. A transistor has an electrode that turns it on and off, called the gate. This transistor has a small "pocket" under its gate.

The pocket is wrapped in a film that does not conduct electricity. This film is called an "insulating film." When electrons (the tiny particles that carry electric current) are trapped in the pocket, the transistor becomes harder to turn on.

To read the cell, a fixed voltage is applied to the gate. If the transistor turns on, there are no electrons; if it stays off, there are. The insulating film keeps electrons from escaping even with the power off, so the data stays.

In the classic design, this pocket is called a floating gate (an electrode that floats, connected to nothing). Most high-capacity products today use a film that traps electrons and holds on to them, instead of a pocket. Samsung began mass-producing chips with cells stacked vertically in 2013, and they also use this method.

NAND flash memory was presented in 1987 by a Toshiba team led by Fujio Masuoka. It links cells in a chain, like beads on a string, to save space. This lets makers build large capacities cheaply.

To write, a strong voltage pushes electrons through the insulating film into the pocket. To erase, a strong voltage pulls them back out. Repeating this slowly damages the film, and it gets worse at holding electrons. This is why flash memory can only be rewritten a limited number of times.

So the controller chip inside an SSD or phone spreads writes around, so they do not pile up in the same place. This technique is called wear leveling.

What happens when memory runs short?

With too many apps open, memory can fill up. What happens next depends on the operating system (OS), the basic software that manages the whole computer. Here we describe typical cases, simplified.

On a Windows PC, data that has not been used for a while is moved out of memory into a special area in storage. This area is called the "page file." The freed memory can then go to the apps in use now. But when the moved data is needed again, it must be read back from storage, and things slow down meanwhile.

According to Android's developer documentation, Android does not use storage this way. Frequent writes would wear out the flash memory. (Some phone makers add their own features that do use storage.) Instead, Android frees memory by compressing data inside memory, or by closing apps left open in the background. Apple's iOS on the iPhone also closes apps when memory runs short.

An app that was closed this way is loaded again from storage the next time you open it. This is why an app you switch back to after a while may restart from its first screen, or take a moment to appear. The more spare memory there is, the less often apps need to be closed.

Why does "256 GB" show up as less?

Buy a 256 GB phone, and the settings screen shows less space than that. One reason is that the OS and preinstalled apps already take up space. Another is that there are two ways to count a "GB."

Storage makers count 1 GB as one billion bytes. In computing, though, it is often handier to count in steps that double each time. So some software counts 1 GB as 2 to the 30th power (1,073,741,824) bytes. Counted this way, 256 GB comes out as about 238 GB.

According to Seagate, Windows shows capacity using the second way of counting. According to Apple, today's iPhones and Macs count in billions of bytes. So the same storage can show a different number depending on which device you look at it on.

To tell the two apart, the IEC (International Electrotechnical Commission) decided in 1998 to call 2 to the 30th power bytes a "GiB" (gibibyte). In practice, though, both are still almost always written as "GB." For memory, the "8 GB" by custom uses the 2-to-the-30th count, following the definition of JEDEC, the industry group that sets semiconductor standards.

Common misconceptions

By the numbers

The first four rows are rough waits for a single read. Stretch the cache's 1 ns to 1 second, and memory takes about 100 seconds, an SSD about a day, and an HDD about four months. Just one step down, from memory to SSD, makes the wait roughly a thousand times longer.

The SSD figure is close to the read time of the flash chip itself (about 0.05 ms in a 2019 Toshiba Memory document). How much data moves per second is a separate number: the 990 PRO reads up to 7,450 MB per second in long runs. Even so, for the wait on each single read, it is no match for memory.

Learn more

• Compare in a table: DRAM vs. NAND flash, DRAM vs. SRAM (the memory used for caches), SSD vs. HDD, NAND vs. NOR flash, NVMe vs. SATA SSDs
• Terms for the next step: 3D NAND (cells stacked vertically), TLC (3 bits in one cell), LPDDR (DRAM for phones), virtual memory (how memory is managed)
• Try the numbers: the image, audio and video data size calculator (a rough guide to how much fits in storage)
• See it in figures: the memory hierarchy and the layouts of DRAM and NAND cells are in the list of figures
• Start from the foundation: "What Is a Semiconductor?"