Skip to main content

What is CPU cache, and why is it so important for gaming?

AMD’s 7800X3D and 7950X3D hold the top spot in CPUs for gaming, not because they have the most cores or the highest clock speeds, but because they have the most cache. But what is CPU cache, anyway? It’s a small quantity of super-fast, rapid-access memory built into the chip itself, helping it get the data it needs for operations at blazing speed.

The returns aren’t linear, though — there’s a reason the 7950X3D doesn’t have additional cache on all of its cores. In fact, there are some downsides to having lots of extra cache to work with, even if it does help push up gaming performance. Here’s everything you need to know about CPU cache.

Recommended Videos

What is CPU cache?

AMD Ryzen 7 5800X3D chip.
Image used with permission by copyright holder

Cache is the amount of memory that is within the CPU itself, either integrated into individual cores or shared among some or all cores. It’s a small bit of dedicated memory that lives directly on the processor so that your CPU doesn’t need to fetch information from your system RAM every time you want to do something on your PC. Every processor has a small amount of cache, with smaller CPUs getting perhaps just a few kilobytes while large CPUs can have many megabytes worth of cache.

But why is cache necessary if we already have fast SSD storage, and even faster RAM? It’s all about performance. In the 1990s, RAM speed wasn’t keeping pace with CPU needs, presenting a problem for CPU designers. The solution was to add local cache to the chips themselves.

Although cache has limited capacity compared to RAM, its high speed makes up for it. The only downside is that it’s expensive to pack memory into a tiny chip, so it’s typically not been used in larger quantities. With 3D V-Cache, however, AMD found a way to make it work, and now we have higher-performing game chips to show for it.

How does cache work? L1, L2, L3 explained

The mainstream adoption of cache resulted in more nuanced implementations of cache and RAM until we ended up with the memory hierarchy, with cache at the top, RAM in the middle, and storage at the bottom. This tiered approach allows critical data for the CPU to be physically closer to the processor, reducing latency and helping your PC feel snappy.

The modern memory hierarchy.
Carlos Carvalho

Cache has its own hierarchy, or cache levels, which are split into L1, L2, and L3 cache. These are all kinds of cache, but they perform slightly different functions.

L1 cache is the first level of cache and also the smallest, usually divided into L1 instruction or L1i and L1 data or L1d. Each core within a CPU has its exclusive chunk of L1 cache, which is usually only a few kilobytes large. The kind of data stored in L1 cache is stuff that the CPU just used or expects to use imminently. If the CPU needs data that isn’t in the L1 cache, it goes to the next level: L2.

Like L1 cache, L2 cache is often exclusive to a single CPU core, but in some CPUs, it’s shared between multiple cores. It’s also much, much larger; for example, each P-core in the Core i9-12900K has 80 kilobytes of L1 cache, as well as 1.25 megabytes of L2 cache, nearly 16 times as much. However, larger caches have higher latency, which means it takes more time for communication to happen between the CPU core and the cache. When CPUs want to accomplish things in a matter of microseconds or even nanoseconds, the slightly higher latency of L2 cache does matter. If a CPU can’t find requested data within L2 cache, it asks the next level: L3.

L3 cache is a big deal: It’s shared between some or all cores within a CPU, and it’s big. The 7950X3D, for example, has 128MB of L3 cache with its bolted-on 3D V-Cache, while it only has 16MB of L2 cache. The latency of L3 cache is even worse than L2, but having a large L3 cache is really important to prevent the CPU from needing to ask the RAM for needed data. Except for storage, RAM has the worst speed and latency in the memory hierarchy, and whenever the CPU needs to access the RAM for required data, things can grind to a halt.

Some CPUs even have L4 cache, but it usually functions as RAM that’s on the CPU package. Some of Intel’s first 14nm CPUs based on the Broadwell architecture included 128MB of embedded DRAM, and the company’s Sapphire Rapids server CPUs can come with HBM2, which is kind of used like an extra level of cache.

Does CPU cache matter for gaming?

AMD CEO holding 3D V-Cache CPU.
AMD

CPU cache makes a big difference for gaming. Although single-threaded performance, instructions per clock (IPC), and clock speed have traditionally been said to be the most important factors in gaming performance, it’s become very clear that cache is probably the most important factor of all in the rivalry between AMD and Intel.

Cache is so important for gaming because of how games are designed today. Modern games have a lot of randomness, which means that the CPU constantly needs to execute simple instructions. Without enough cache, your graphics card is forced to wait on your CPU as the instructions pile up and cause a bottleneck. You can see an example of how much of a difference that makes with AMD’s 3D V-Cache technology in the CPU performance graph in gaming below.

Overall gaming performance for the Ryzen 7 7800X3D.
Jacob Roach / Digital Trends

We’ve seen a trend toward more cache for gaming in recent years. AMD has been increasing its CPU cache quantities for years, and doubled down with its 3D V-Cache technology on the Ryzen 7 5800X3D and its successors in the Ryzen 7000 generation

Intel has been playing catch-up with AMD, and its latest generation CPUs have more cache than ever before, helping them remain competitive in gaming. It seems likely that cache quantities will continue to rise in the coming years, further pushing gaming performance limits.

Matthew Connatser
Former Computing Writer
Matthew Connatser is a freelance writer who works on writing and updating PC guides at Digital Trends. He first got into PCs…
Intel’s new $249 GPU brings 1440p gaming to the masses
An exploded view of Intel's Arc A580 GPU.

Intel is trying to redefine what a "budget GPU" really means in 2024, and it's doing so with the new Arc B580 GPU. In what Intel itself described as its "worst kept secret," the B580 is the debut graphics card in Intel's new Battlemage range of discrete GPUs, and it's arriving at just $249. That's a price point that's been relegated to 1080p for decades, but Intel says the B580 will change that dynamic.

It's a 1440p GPU, at least by Intel's definition. That's despite the fact that Intel is comparing the card to GPUs like the RTX 4060 and RX 7600, both of which are more expensive than the B580 and squarely target 1080p. Intel says it can deliver higher performance than these two GPUs while undercutting the price, all in an attempt to capitalize on 1440p gamers. "1440p is becoming 1080p," as Intel's Tom Petersen put it in a pre-briefing with the press.

Read more
Intel announces sudden departure of CEO amid financial turmoil
Intel CEO Pat Gelsinger holding a chip.

Intel has announced that CEO Pat Gelsinger has retired. The executive, who first joined Intel in 1979 at 18 years old, is being replaced by David Zinsner and Michelle Johnston Holthaus. Holthaus and Zinsner will serve as interim co-CEOs while the board of directors works "diligently and expeditiously" to find a successor.

Gelsinger became CEO in early 2021. At the time, Intel was struggling to regain ground it had lost to AMD in the desktop market, as well as push a more ambitious manufacturing timeline to catch up with foreign chipmakers like TSMC. Under Gelsinger's leadership, the company made some big strides. Intel's 12th generation of processors marked a significant turning point in the company's desktop processors, and an aggressive foundry roadmap has pushed smaller nodes out of U.S.-based plants.

Read more
You could buy a gaming laptop for the price of this AMD handheld
The OneXFly F1 Pro console.

The first gaming handheld to feature one of AMD's best processors is finally here, but it's going to be a tough sell. The OneXFly F1 Pro comes with a host of impressive features, from an OLED screen to the powerful Ryzen AI 9 HX 370 APU. It could undoubtedly beat the Steam Deck and various other rivals. The downside? You might as well just buy a gaming laptop, and a good one, too -- because this one's pretty expensive.

The OneXFly F1 Pro (first spotted by VideoCardz) is a brand-new gaming handheld made by OneXPlayer, a Chinese company. The mini PC comes with all the bells and whistles you could wish for in a new handheld. There's a 7-inch 1080p OLED screen with a 144Hz refresh rate; LPDDR5X RAM with speeds of up to 7,500MHz; and up to 4TB of PCIe 4.0 solid-state drive storage.

Read more