Although many computer components range from a little complicated to esoteric, RAM is perhaps one of the best-known. From memes about dedicated RAM to looking at system requirements for video games, lots of people (or maybe even most people) have heard the term a few times. But as with all computer hardware, there's quite a bit to what RAM is and what it does.
RAM: A small but fast kind of memory for all kinds of processors
Random-access memory (RAM) is a very broad category of memory data storage that prioritizes performance over capacity, but not to an extreme. There are two main types of RAM: Dynamic RAM (DRAM) and Static RAM (SRAM). Colloquially, people generally mean DRAM when they say RAM, as SRAM is more commonly called cache. I'll be focusing on DRAM here, but we do have a separate explainer on SRAM/cache.
For a computer, you just need two basic components: a processor and storage. But early on in the history of computing, engineers realized storage mediums with lots of capacity were super slow and didn't allow the processor to run at full bore. RAM was created as an intermediary between permanent storage (like hard drives and solid-state drives) and the processor, with a focus on quick speed and lower latency. The basic idea is that you put lots of data you're going to need in the near future on RAM, swapping data in and out as needed over time.
The primary features that distinguish different kits of RAM are clock speed, bandwidth, capacity, and latency, but generally, RAM can only specialize in one or two of these things. Double data rate (or DDR) types of memory tend to have very high clock speeds and high capacity, with the latest DDR5 standard starting at 4,800MHz with up to 512GB on a single stick, whereas DDR4 starts at 2,133MHz and only has up to 128GB per stick. However, each generation of DDR increases the latency, and other kinds of DRAM may have lower clock speeds but more bandwidth.
Of course, RAM has to give something up for all its speed. Firstly, it's much more expensive per amount of data than permanent storage mediums. Today, 32GB of DDR5 RAM costs about $100, whereas you can get a top-end 1TB SSD for $100. Additionally, RAM needs continuous power in order to retain its data; otherwise, it just stops existing.
RAM isn't just divided into DRAM and SRAM though, as both categories have their own subcategories. The main type of DRAM we care about is SDRAM, or synchronous DRAM. This subcategory includes all the kinds of RAM with DDR in the name, such as DDR5 and DDR4 memory for CPUs and computer systems in general, and GDDR6 memory for GPUs. both of which are maintained by JEDEC. VRAM is also a subcategory of RAM, but it's not a technical category like SDRAM, and instead just refers to RAM that's made for GPUs. High bandwidth memory, or HBM, is a kind of VRAM like GDDR6 but isn't SDRAM.
The memory hierarchy and how DRAM is different from cache or SRAM
There are lots of different kinds of memory in a computer, and the memory hierarchy can be best summarized with a pyramid (as seen in the image above). The CPU with its cache is at the top, DRAM is in the middle as main memory, and more permanent storage devices like HDDs, SSDs, and network connections are at the bottom. It works pretty simply: Data in permanent storage can be copied over to DRAM, and then the cache in the CPU can grab data from the DRAM for immediate use. The higher you go up the pyramid, the faster and smaller things get.
You might be wondering why cache or SRAM hasn't just outright replaced DRAM since it's so much faster and right on the CPU. Well, the general rule of thumb with memory is that the faster it is, the more expensive it is to make, not to mention that size constraints are also a problem. Cache is very expensive to produce, especially because it's often on the same silicon as the processor cores, and lots of companies want their processors to be on the most expensive, cutting-edge node possible. Cache takes up lots of space and that makes production expensive for chips with lots of cache.
There's also one major technical difference between DRAM and SRAM, and it has to do with what "dynamic" and "static" mean in their respective acronyms. Keeping the power on in DRAM actually isn't enough to retain the data; anything stored in DRAM has to be constantly refreshed, so it's not lost. SRAM, however, can retain data as long as it's powered on, but this advantage is also part of the reason why it's so expensive to produce.
Things to consider when buying RAM for PCs and other devices
Most people really only need to worry about DDR DRAM, the kind you buy for desktops, laptops, and other devices. Now that I've covered all the technical aspects of RAM, it's about time to answer some commonly asked questions about buying and upgrading RAM to use in PCs, especially gaming PCs.
"How much RAM do I need?" is one of the most commonly asked computer questions of all time, and there has never been a single correct answer. Part of the problem is that memory requirements for software increase over time, so what was good even five years ago might not be sufficient anymore. Additionally, what you do with your computer will determine how much RAM you need. Generally speaking, 8GB is the minimum for running a Windows 10 or 11 PC smoothly, and 16GB will give you wiggle room for gaming and other intensive workloads. 32GB and beyond is for enthusiasts and users who work with creative software.
Then there are performance-related aspects like frequency and latency. Higher frequencies and lower latencies are both good for performance, but it's hard to improve both simultaneously. Ideally, you want to balance out these two things and consider the hardware you're pairing RAM with. We have recommendations for both DDR4 and DDR5 that cover all the best options.
One final and often overlooked thing to consider is memory channels, which can boost the bandwidth of DDR memory when using a certain number of sticks. Most platforms generally offer two, four, or eight channels, and if the system has the same amount of RAM sticks as channels (double will also work), the bandwidth will get a nice boost. For most users, this means installing two or four sticks of memory in order to activate dual-channel memory, which is what you get on non-professional hardware. You definitely want dual-channel memory too, as single-channel mode ruins performance in pretty much everything, including games. For example, if you want 16GB of RAM, it's better to buy two 8GB sticks instead of a single 16GB one. Some professional hardware may also support quad-channel memory.
The future looks boring for DRAM
Although the field was quite crowded for DRAM several years ago, today the only kinds of DRAM that really matter are JEDEC's DDR memory, like DDR5 and GDDR6. Of JEDEC's few challengers in recent years, such as Intel's Optane, practically none have succeeded. Even HBM, which started as a project between SK Hynix, Samsung, and AMD, was eventually adopted by JEDEC, leaving pretty much the entire ecosystem fairly simplified and unfragmented.
Nothing lasts forever, especially in computing, but it looks like DRAM will be a fairly static field for the foreseeable future. Perhaps there's room for new types of memory tailored to the needs of specific applications, but it's incredibly unlikely we'll see a shakeup more significant than that. It's much more likely companies will keep making DDR memory for years.