---
title: "DDR4 vs DDR5 on servers: bandwidth, real latency and what changes | StreetHosting"
description: "Compare DDR4 and DDR5 on servers: bandwidth, real latency in nanoseconds, subchannels, and the impact on games, databases and compiling, without the marketing."
url: "https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers"
type: "page"
language: "en-US"
---

Infrastructure · 7 min · Intermediate

Published on Nov 13, 2026 · Updated on Nov 13, 2026

# DDR4 vs DDR5 on servers: what the memory swap really delivers

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.

By [Equipe StreetHosting](https://streethosting.com.br/en/autores#equipe-streethosting) · StreetHosting infrastructure and support team

[Latency and ping](https://streethosting.com.br/en/guides/topics/latency) [Network, DNS and domains](https://streethosting.com.br/en/guides/topics/networking) [Hardware and datacenter](https://streethosting.com.br/en/guides/topics/hardware) [Costs, plans and billing](https://streethosting.com.br/en/guides/topics/costs)

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%2Fddr4-vs-ddr5-servers.%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%2Fddr4-vs-ddr5-servers.%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%2Fddr4-vs-ddr5-servers.%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%2Fddr4-vs-ddr5-servers.%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%2Fddr4-vs-ddr5-servers.%20Highlight%20the%20step-by-step%20instructions%2C%20the%20prerequisites%20and%20the%20most%20common%20mistakes. "Perplexity")

Share:

[](https://x.com/intent/tweet?text=DDR4%20vs%20DDR5%20on%20servers%3A%20bandwidth%2C%20real%20latency%20and%20what%20changes&url=https%3A%2F%2Fstreethosting.com.br%2Fen%2Fguides%2Finfrastructure%2Fddr4-vs-ddr5-servers "Share on X") [](https://www.facebook.com/sharer/sharer.php?u=https%3A%2F%2Fstreethosting.com.br%2Fen%2Fguides%2Finfrastructure%2Fddr4-vs-ddr5-servers "Share on Facebook") [](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fstreethosting.com.br%2Fen%2Fguides%2Finfrastructure%2Fddr4-vs-ddr5-servers "Share on LinkedIn") [](https://wa.me/?text=DDR4%20vs%20DDR5%20on%20servers%3A%20bandwidth%2C%20real%20latency%20and%20what%20changes%20https%3A%2F%2Fstreethosting.com.br%2Fen%2Fguides%2Finfrastructure%2Fddr4-vs-ddr5-servers "Share on WhatsApp")

For agents: Copy as Markdown [.md](https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers.md)

In this guide 7 sections

* [What changed from DDR4 to DDR5](https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers#o-que-mudou)
* [Bandwidth: where DDR5 wins](https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers#banda)
* [Real latency in nanoseconds](https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers#latencia)
* [Channels, subchannels and capacity](https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers#canais)
* [Games, databases and compiling](https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers#cargas)
* [How to check and measure](https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers#medir)
* [DDR4 and DDR5 on StreetHosting](https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers#onde-rodar)

Quick answer

**DDR5** delivers a lot more bandwidth than **DDR4**, splits each module into two independent subchannels and allows higher capacity modules. Real latency, measured in nanoseconds, stays close to that of a good DDR4, because DDR5's higher CL comes along with a faster clock. In games, databases and compiling, the CPU difference usually weighs more than the memory one. The StreetHosting Xeon VPS uses DDR4 and the Ryzen 9 9950X VPS uses DDR5.

## What changed from DDR4 to DDR5[](https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers#o-que-mudou)

| Feature                  | DDR4                                             | DDR5                                                   |
| ------------------------ | ------------------------------------------------ | ------------------------------------------------------ |
| Typical transfer rate    | 2133 to 3200 MT/s                                | 4800 MT/s or more, above 6000 on enthusiast kits       |
| Voltage                  | 1.2 V                                            | 1.1 V, with the voltage regulator on the module itself |
| Channels per module      | 1 channel of 64 bits                             | 2 independent subchannels of 32 bits                   |
| Burst length             | 8                                                | 16                                                     |
| Banks per chip           | Up to 16 (4 groups of 4)                         | Up to 32 (8 groups of 4)                               |
| Internal ECC on the chip | No                                               | Yes, on every chip                                     |
| Capacity per module      | Up to 32 GB on desktop, 64 GB or more on servers | 48 GB and 64 GB on desktop, much more on servers       |

Every row in that table serves the same problem: keeping memory feeding processors with more and more cores. DDR5 bets on more parallelism, with subchannels and more banks, and on more bandwidth, and pays with tighter timing and a higher CL.

## Bandwidth: where DDR5 wins[](https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers#banda)

The theoretical bandwidth of a 64 bit channel is the transfer rate multiplied by 8 bytes. The math shows the gap between generations:

| Memory    | Rate      | Bandwidth per 64 bit channel |
| --------- | --------- | ---------------------------- |
| DDR4 2666 | 2666 MT/s | 21.3 GB/s                    |
| DDR4 3200 | 3200 MT/s | 25.6 GB/s                    |
| DDR5 4800 | 4800 MT/s | 38.4 GB/s                    |
| DDR5 5600 | 5600 MT/s | 44.8 GB/s                    |
| DDR5 6000 | 6000 MT/s | 48.0 GB/s                    |

DDR5 6000 delivers almost double the bandwidth of DDR4 3200 per channel. Those numbers are the ceiling. In practice, a program only takes advantage of that gain if it is limited by bandwidth, which is the case for workloads that sweep large amounts of data: analytics, scientific computing, memory copies and some databases on heavy queries.

## Real latency in nanoseconds[](https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers#latencia)

The CL number (CAS Latency) alone misleads, because it counts clock cycles, and the clock changes between generations. The real latency, in nanoseconds, is:

`latency (ns) = CL x 2000 / rate in MT/s DDR4 3200 CL22 = 22 x 2000 / 3200 = 13,75 ns DDR4 3200 CL16 = 16 x 2000 / 3200 = 10,00 ns DDR5 4800 CL40 = 40 x 2000 / 4800 = 16,67 ns DDR5 5600 CL46 = 46 x 2000 / 5600 = 16,43 ns DDR5 6000 CL30 = 30 x 2000 / 6000 = 10,00 ns`

A mid frequency DDR5, the kind seen on servers, usually has a CAS latency between 14 and 17 ns, a little above a good DDR4. Only enthusiast DDR5 kits reach 10 ns. The difference is a few nanoseconds, and that is only one part of the total.

The latency a program feels when fetching data that is not in cache is the sum of several stages: the path inside the processor, the memory controller, the bus and the chip. On a real machine the total is in the range of a few tens of nanoseconds, and CL accounts for only part of it. That is why a processor's large cache weighs so much: it avoids the trip to memory. The guide on [CPU L3 cache on game servers](https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server) shows this effect.

The rule for reading an ad: ignore the CL on its own and divide by the clock. A kit with a higher CL and a higher frequency can have the same real latency as a kit with a lower CL and a lower frequency, and still deliver more bandwidth.

## Channels, subchannels and capacity[](https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers#canais)

A DDR4 module has a single 64 bit channel. A DDR5 module has two 32 bit subchannels that operate independently, each with its own burst of 16 transfers. Each burst still delivers 64 bytes, the size of a cache line, but now two different requests can be served at the same time on the same module. With many cores asking for different data, this cuts the queues.

* **More channels on the platform:** what sets the total bandwidth is the number of channels times the speed of each one. Many server processors have 4, 6, 8 or more channels, depending on the generation, against 2 on desktop processors such as Ryzen. A machine with 8 channels of DDR4 can have more total bandwidth than one with 2 channels of DDR5.
* **Capacity:** DDR5 allows larger chips and higher capacity modules, which helps anyone who needs a lot of RAM in few slots.
* **ECC:**DDR5's internal ECC is not the same as module level ECC, as the guide on [ECC memory on servers](https://streethosting.com.br/en/guides/infrastructure/ecc-memory-servers) explains.

## Games, databases and compiling[](https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers#cargas)

The effect of memory depends on how much the program sits waiting for it.

* **Game server:**the game's main loop runs on one or a few cores and reads and writes lots of small pieces of data. What weighs most is single core performance and cache. Memory speed shows up, but it usually weighs less than the CPU, and the effect varies a lot with the game and the processor. The comparison between [one strong core and many cores](https://streethosting.com.br/en/guides/infrastructure/single-thread-vs-multi-thread-cpu) explains why that dominates.
* **Database:** small, frequent queries, like those of a website or an API, depend on latency, page cache and disk. Analytical queries that sweep large tables depend on bandwidth, and that is where DDR5 and more channels help. The guide on [a VPS for databases](https://streethosting.com.br/en/guides/vps/choose-vps-for-database) covers general sizing.
* **Compiling:** compiling is CPU work, with lots of reading of files that the system keeps cached. More cores and more frequency shorten a build far more than the memory generation does.
* **Ordinary web apps, bots and APIs:** they are almost never limited by memory, but by CPU, disk or network.

Comparing a DDR4 machine with a DDR5 one almost always also compares the processor generation, the core count, the cache and the frequency. Whoever concludes that DDR5 gave some performance gain is usually measuring a newer CPU. To isolate memory, the only fair way is the same processor with both generations, and that is only possible on the few platforms that accept both.

## How to check and measure[](https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers#medir)

You can see the configured memory speed and run a simple bandwidth test. On a VPS, what you see depends on what the hypervisor shows, and the test measures the bandwidth your virtual machine can get, which is what matters for your use.

`# memory type and speed (on a physical machine) sudo dmidecode -t memory | grep -E "Type:|Speed:|Configured" # memory bandwidth test sudo apt install -y sysbench sysbench memory --memory-block-size=1M --memory-total-size=20G run`

sysbench reports the rate in MiB per second. By default it uses a single thread; to measure with every core, add `--threads=$(nproc)` to the command. Run it a few times and compare the average, because the result varies with the load on the machine. To check how much CPU and RAM is left during use, see the guide on [how to check CPU, RAM, disk and network on Linux](https://streethosting.com.br/en/guides/vps/check-cpu-ram-disk-network-linux).

## DDR4 and DDR5 on StreetHosting[](https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers#onde-rodar)

The two [VPS](https://streethosting.com.br/en/vps) families use different memory generations, and choosing between them is mainly a choice of processor.

* **[Xeon VPS](https://streethosting.com.br/en/vps/xeon):** uses DDR4 and starts at R$ 26.00 per month, with 2 vCPU, 2 GB of RAM, 20 GB NVMe and a 10 Gbps uplink. It suits websites, APIs, bots, small databases and services that do not depend on single core performance.
* **[Ryzen 9 9950X VPS](https://streethosting.com.br/en/vps/ryzen):** uses DDR5 and starts at R$ 43.00 per month, with 1 vCPU, 2 GB and NVMe, with a 1 Gbps uplink. It is the pick for what depends on per core performance, such as a game server.

Whoever is deciding between the two should look at what the service does with the CPU before looking at the memory generation. The full comparison is in [Ryzen VPS vs Xeon VPS](https://streethosting.com.br/en/guides/vps/ryzen-vs-xeon-vps). All of them are in São Paulo, with root access, AntiDDoS included and up to 5 TB of traffic per month.

In this guide

* [What changed from DDR4 to DDR5](https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers#o-que-mudou)
* [Bandwidth: where DDR5 wins](https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers#banda)
* [Real latency in nanoseconds](https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers#latencia)
* [Channels, subchannels and capacity](https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers#canais)
* [Games, databases and compiling](https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers#cargas)
* [How to check and measure](https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers#medir)
* [DDR4 and DDR5 on StreetHosting](https://streethosting.com.br/en/guides/infrastructure/ddr4-vs-ddr5-servers#onde-rodar)

## Frequently asked questions

Is DDR5 always faster than DDR4?

In bandwidth yes, by a wide margin. In real latency not necessarily: DDR4 3200 with CL16 is about 10 ns, and many mid frequency DDR5 kits land between 14 and 17 ns. Final performance depends on the processor and on what the program does with memory.

Does DDR5's higher CL mean it is worse?

No. CL counts clock cycles, and DDR5's clock is faster, so each cycle is shorter. To compare, convert to nanoseconds: CL multiplied by 2000 and divided by the rate in MT/s.

Is it worth moving from DDR4 to DDR5 to improve a game server?

The memory alone rarely changes much. Game servers depend mostly on the performance of a single CPU core and on cache. The gain you see when changing platform comes mainly from the newer processor that uses DDR5.

Can I put DDR5 memory on a DDR4 board?

No. The modules have a different notch, and the board and processor have to be the right generation. Going from DDR4 to DDR5 means changing platform.

What matters more, more memory channels or DDR5?

It depends on the workload. For work that reads a lot of memory in parallel, such as data analysis, more channels weigh heavily. Servers with 4, 6 or 8 channels of DDR4 can have more total bandwidth than a desktop with 2 channels of DDR5.

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 Xeon VPS Xeon VPS for steady workloads, automation and long-running projects.](https://streethosting.com.br/en/vps/xeon)

## Related guides

[VPS Beginner Ryzen VPS vs Xeon VPS: which CPU to choose for your server Choosing between a Ryzen VPS and a Xeon VPS comes down to the workload, not the brand on its own. This guide compares the actual processors behind the two lines on clock speed, per-core performance, stability and ECC memory, with practical scenarios and the price of every plan. 10 min Read guide](https://streethosting.com.br/en/guides/vps/ryzen-vs-xeon-vps) [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 CPU L3 cache: why it matters for 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. 8 min Read guide](https://streethosting.com.br/en/guides/infrastructure/cpu-l3-cache-game-server)

[← Back to the Guide Center](https://streethosting.com.br/en/guides)
