--- title: "CPU L3 cache: why it matters for game servers and databases | StreetHosting" description: "Learn the L1, L2 and L3 cache hierarchy, why a big L3 speeds up game servers and databases, the Ryzen X3D case and how to check your cache with lscpu on Linux." url: "https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server" type: "page" language: "en-US" --- Infrastructure · 8 min · Intermediate Published on Nov 13, 2026 · Updated on Nov 13, 2026 # CPU L3 cache: what it is and why it helps game servers and databases RAM is slow next to the processor. Cache is the small, fast memory inside the chip, and on workloads with many random accesses, like game servers and databases, more L3 changes the result. By [Equipe StreetHosting](https://streethosting.com.br/en/autores#equipe-streethosting) · StreetHosting infrastructure and support team [Hardware and datacenter](https://streethosting.com.br/en/guides/topics/hardware) [Databases](https://streethosting.com.br/en/guides/topics/databases) Summarize with: [](https://chat.openai.com/?q=Summarize%20the%20key%20points%20of%20this%20StreetHosting%20guide%3A%20https%3A%2F%2Fstreethosting.com.br%2Fen%2Fguides%2Finfrastructure%2Fcpu-l3-cache-game-server.%20Highlight%20the%20step-by-step%20instructions%2C%20the%20prerequisites%20and%20the%20most%20common%20mistakes. "ChatGPT") [](https://claude.ai/new?q=Summarize%20the%20key%20points%20of%20this%20StreetHosting%20guide%3A%20https%3A%2F%2Fstreethosting.com.br%2Fen%2Fguides%2Finfrastructure%2Fcpu-l3-cache-game-server.%20Highlight%20the%20step-by-step%20instructions%2C%20the%20prerequisites%20and%20the%20most%20common%20mistakes. "Claude") [](https://www.google.com/search?udm=50&aep=11&q=Summarize%20the%20key%20points%20of%20this%20StreetHosting%20guide%3A%20https%3A%2F%2Fstreethosting.com.br%2Fen%2Fguides%2Finfrastructure%2Fcpu-l3-cache-game-server.%20Highlight%20the%20step-by-step%20instructions%2C%20the%20prerequisites%20and%20the%20most%20common%20mistakes. "Google AI Mode") [](https://x.com/i/grok?text=Summarize%20the%20key%20points%20of%20this%20StreetHosting%20guide%3A%20https%3A%2F%2Fstreethosting.com.br%2Fen%2Fguides%2Finfrastructure%2Fcpu-l3-cache-game-server.%20Highlight%20the%20step-by-step%20instructions%2C%20the%20prerequisites%20and%20the%20most%20common%20mistakes. "Grok") [](https://www.perplexity.ai/search/new?q=Summarize%20the%20key%20points%20of%20this%20StreetHosting%20guide%3A%20https%3A%2F%2Fstreethosting.com.br%2Fen%2Fguides%2Finfrastructure%2Fcpu-l3-cache-game-server.%20Highlight%20the%20step-by-step%20instructions%2C%20the%20prerequisites%20and%20the%20most%20common%20mistakes. "Perplexity") Share: [](https://x.com/intent/tweet?text=CPU%20L3%20cache%3A%20why%20it%20matters%20for%20game%20servers%20and%20databases&url=https%3A%2F%2Fstreethosting.com.br%2Fen%2Fguides%2Finfrastructure%2Fcpu-l3-cache-game-server "Share on X") [](https://www.facebook.com/sharer/sharer.php?u=https%3A%2F%2Fstreethosting.com.br%2Fen%2Fguides%2Finfrastructure%2Fcpu-l3-cache-game-server "Share on Facebook") [](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fstreethosting.com.br%2Fen%2Fguides%2Finfrastructure%2Fcpu-l3-cache-game-server "Share on LinkedIn") [](https://wa.me/?text=CPU%20L3%20cache%3A%20why%20it%20matters%20for%20game%20servers%20and%20databases%20https%3A%2F%2Fstreethosting.com.br%2Fen%2Fguides%2Finfrastructure%2Fcpu-l3-cache-game-server "Share on WhatsApp") For agents: Copy as Markdown [.md](https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server.md) In this guide 7 sections * [The L1, L2 and L3 hierarchy](https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server#hierarquia) * [Why L3 helps game servers and databases](https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server#por-que-ajuda) * [The Ryzen X3D case](https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server#x3d) * [The Ryzen 9 9950X and its 64 MB](https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server#9950x) * [Cache on a VPS is shared](https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server#vps) * [How to see the cache with lscpu](https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server#lscpu) * [Where to run it at StreetHosting](https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server#onde-rodar) Quick answer **CPU cache** is a small, very fast memory inside the processor, in three levels: **L1** and **L2** are private to each core and **L3** is larger and shared. It saves the trip to RAM, which is dozens of times slower. Game servers and databases make many random accesses to data that repeats, so they benefit from a big L3. That is why Ryzen X3D chips stand out in games and why the Ryzen 9 9950X, with 64 MB of L3, is a good base for this kind of load. You can see the sizes with `lscpu`. ## The L1, L2 and L3 hierarchy[](https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server#hierarquia) A CPU runs instructions at billions of cycles per second, but RAM answers in a time equal to hundreds of cycles. If the core had to wait for RAM on every read, it would sit idle almost all the time. Cache fixes that by keeping a copy of recently used data close to the core. It is organized in layers, and each one trades capacity for speed: | Level | Typical size | Approximate latency | Who uses it | | ----- | ---------------------------------- | -------------------- | ------------------ | | L1 | Tens of KB per core | A few cycles | Only its own core | | L2 | From 512 KB to a few MB per core | About a dozen cycles | Only its own core | | L3 | From 16 to over 100 MB on the chip | A few dozen cycles | Shared among cores | | RAM | Tens of GB | Hundreds of cycles | The whole system | The exact numbers change from model to model, but the proportion holds: each step down is bigger and slower. When a core needs a piece of data, it looks in L1, then L2, then L3, and only then goes to RAM. Finding it in cache is called a _hit_ and not finding it is called a _miss_. What counts is the hit rate: a few more misses already turn into a lot of waiting time. ## Why L3 helps game servers and databases[](https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server#por-que-ajuda) Cache works well when the program reuses data. Two kinds of pattern favor it, and game servers and databases have both. * **A main loop on one thread:** a game server runs the simulation in fixed cycles, the ticks. On every tick, a single thread walks through entities, players, collisions and rules. That thread is the bottleneck, and every cache miss leaves it waiting on RAM. More L3 keeps more of the world state nearby. How this relates to choosing cores is in [single thread or multi thread CPU for a server](https://streethosting.com.br/en/guides/infrastructure/single-thread-vs-multi-thread-cpu). * **Many random accesses:** a database looks up records by index, walks trees and follows pointers. Each step jumps to a different address, with no sequential pattern the hardware can predict. If the most used index pages fit in L3, the database answers without touching RAM. The opposite also holds. A task that reads a large file from start to finish, like encoding a video, uses each piece of data once and moves on. The CPU prefetcher anticipates that sequential pattern, and the size of the L3 barely changes the result. What decides the gain is the hot data set, the part of the state that is touched all the time. If it fits in the cache, the jump is large. If it is much bigger than the cache, extra L3 helps little, and then RAM speed and latency take over. To compare generations, see [DDR4 or DDR5 on a server](https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers). ## The Ryzen X3D case[](https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server#x3d) AMD created the X3D line precisely to exploit this. Instead of making the chip bigger sideways, it stacks an extra layer of L3 cache together with the core chiplet, a technology called 3D V-Cache (on top of the chiplet in the 5000 and 7000 series, underneath it in the 9000 series). A Ryzen 7 7800X3D, for example, reaches 96 MB of L3, against 32 MB on the regular equivalent model. The effect shows up in games and in simulations with many random accesses, which is why these processors lead game benchmarks even with a slightly lower clock. On a game server the same logic applies: the tick thread barely waits on RAM. The gain depends on the game and the world size, so treat benchmark numbers as a hint and not as a promise. A big cache does not fix everything. X3D models tend to have a lower or more limited clock, and workloads that depend on raw frequency or on many cores may perform the same or worse. Choose by the profile of your load, not by the name of the processor. ## The Ryzen 9 9950X and its 64 MB[](https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server#9950x) The Ryzen 9 9950X is not an X3D, but it already comes with plenty of L3: 64 MB in total, split between two core chiplets (CCDs) with 32 MB each. That split matters. Each core only uses its own CCD's L3 as its cache. When the data it needs sits in the other CCD's L3, the copy comes over the Infinity Fabric, the link between the chiplets, and takes much longer. In practice, a thread that always runs on the same CCD has that CCD's 32 MB within reach. For a game server, where one main thread dominates, that is good: it sits on a fast core with a large L3 within reach. Each core also has its own L2, which reinforces the same idea. ## Cache on a VPS is shared[](https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server#vps) On a dedicated server, all of the machine's cache is yours. On a VPS, it is not. KVM virtualization splits the physical machine's cores among customers, and the L3 is shared by every core of the same chiplet, no matter who they are assigned to at that moment. That means your VPS's effective cache depends on how much the neighbors use. If another VPS on the same machine runs a load that sweeps through a lot of memory, it pushes your data out of the L3, and yours starts to miss more. There is no guarantee of a fixed slice of cache per customer. That is why measuring matters more than reading the spec sheet: run your real load and compare, as in the guide on [checking CPU, RAM, disk and network on Linux](https://streethosting.com.br/en/guides/vps/check-cpu-ram-disk-network-linux). For anyone who needs fully predictable cache, a dedicated server is the option. ## How to see the cache with lscpu[](https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server#lscpu) On Linux, the `lscpu` command lists the caches at the end of its output: `lscpu | grep -E 'L1d|L1i|L2|L3' # one line per level, with the size of each instance and the total lscpu -C # straight from the file system, for core 0 cat /sys/devices/system/cpu/cpu0/cache/index3/size` On Ubuntu 24.04 the lscpu summary groups the caches under the `Caches (sum of all)` section, with the total for each level and the number of instances. An L3 of 64 MiB in 2 instances, for example, is two blocks of 32 MiB, one per CCD. `lscpu -C` shows the size of one instance in the `ONE-SIZE` column and the total in `ALL-SIZE`. L1 shows up split into data (d) and instructions (i). Inside a VPS what you see is what the hypervisor exposes, and the values may reflect the whole physical machine and not your share, or differ from what the CPU really has. Use it as a reference for the model, not as your quota. ## Where to run it at StreetHosting[](https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server#onde-rodar) The [Ryzen 9 9950X VPS](https://streethosting.com.br/en/vps/ryzen) runs on that processor, with NVMe and a 1 Gbps uplink, and starts at R$ 43.00 per month with 1 vCPU and 2 GB of DDR5. It is a good base for a game server or a database that depends on one fast thread, keeping in mind that the cache is shared with other customers on the same machine. Every VPS is KVM, with root access, in São Paulo, with AntiDDoS included and up to 5 TB of traffic per month. When the load is heavy enough to need the whole machine, the [dedicated servers](https://streethosting.com.br/en/dedicated) in São Paulo give you the entire machine, without sharing cache or cores with other customers, with delivery in up to 6 days. In this guide * [The L1, L2 and L3 hierarchy](https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server#hierarquia) * [Why L3 helps game servers and databases](https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server#por-que-ajuda) * [The Ryzen X3D case](https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server#x3d) * [The Ryzen 9 9950X and its 64 MB](https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server#9950x) * [Cache on a VPS is shared](https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server#vps) * [How to see the cache with lscpu](https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server#lscpu) * [Where to run it at StreetHosting](https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server#onde-rodar) ## Frequently asked questions What is CPU L3 cache? It is the last layer of fast memory inside the processor, before RAM. It is shared among several cores and holds data that was used recently or is likely to be used next, avoiding a trip to RAM, which is much slower. Does more L3 cache make any server faster? No. The gain shows up on workloads whose hot data fits in the cache and that make many random accesses, such as game servers, databases and in memory caches. Tasks that read data sequentially, like video encoding, gain little. What is the 3D V-Cache on Ryzen X3D chips? It is an extra layer of L3 cache stacked together with the core chiplet. It greatly increases the total capacity, for example to 96 MB on models like the Ryzen 7 7800X3D, which benefits games and workloads with many random accesses. How much L3 cache does the Ryzen 9 9950X have? 64 MB of L3, split into 32 MB for each of its two core chiplets (CCDs). Each core uses the L3 of its own chiplet as its cache, and data sitting in the other chiplet's L3 arrives much more slowly. On a VPS, do I get the cache all to myself? Usually not. A VPS is a slice of a physical machine, and the L3 is shared among all cores of the same chiplet. If other customers on the same machine use a lot of cache, less is left for you. Inside the VPS, lscpu shows what the hypervisor reports, which is usually the processor's cache and not your share. Next step See Ryzen VPS Ryzen 9 9950X VPS in São Paulo with root access, NVMe and gamer Anti-DDoS. [See Ryzen VPS](https://streethosting.com.br/en/vps/ryzen) [See dedicated servers Exclusive hardware in São Paulo with NVMe and Anti-DDoS.](https://streethosting.com.br/en/dedicated) ## Related guides [Infrastructure Intermediate Single thread vs multi thread CPU: what matters on a server A game server can lag with the CPU at 25% while a database puts every core it can find to work. Learn where that difference comes from, how to tell which side your application is on, and what kind of vCPU to buy. 11 min Read guide](https://streethosting.com.br/en/guides/infrastructure/single-thread-vs-multi-thread-cpu) [Infrastructure Intermediate DDR4 vs DDR5 on servers: bandwidth, real latency and what changes DDR5 has much more bandwidth and about the same real latency as good DDR4. In practice, the CPU weighs more than the memory in almost every server workload, and the right math is in nanoseconds, not just the CL number. 7 min Read guide](https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers) [Infrastructure Advanced ECC memory: what it is and why database and storage servers use it ECC memory stores extra bits to correct a one bit error and detect a two bit error. It prevents silent data corruption on machines that stay on for months, such as database and storage servers. 9 min Read guide](https://streethosting.com.br/en/guides/infrastructure/ecc-memory-servers) [← Back to the Guide Center](https://streethosting.com.br/en/guides)