DDR5 for gaming: speed, CL and EXPO/XMP

A DDR5 box shows two numbers that look like a robot's postcode: 6000 and CL30. Knowing which one matters, and which one only looks impressive, saves you from paying for the wrong one.

The short answer

How fast does DDR5 need to be for gaming? Not as fast as the box suggests. Any DDR5 kit runs your games [10], and the gain from faster memory depends on your processor and on whether the processor is the bottleneck. In the one dated test we can cite, a kit at DDR5-6000 CL30 was 4 percent faster on average than a DDR5-5200 CL40 kit on a Ryzen 7 7800X3D, and 12 percent faster on a Ryzen 7 7700X [11]. Compare kits by two numbers together: the data rate and the first-word latency in nanoseconds. Switch on the EXPO or XMP profile in the BIOS, and test that it is stable. Our assessment: a kit around DDR5-6000 with CL30 to CL36 is the sensible middle on AMD's AM5 platform, and a kit that costs much more rarely earns it back in a game.

How it works

1. What the numbers on the box mean. "DDR5-6000" is the data rate in million transfers per second (MT/s), not a clock speed. The memory clock of a DDR5-6000 module ticks at 3000 MHz and moves data on both edges, hence 6000 [10]. "CL30" is the CAS latency: the number of clock cycles between the processor asking for data and the memory starting to answer [10]. Because cycles get shorter as the clock rises, a higher CL does not automatically mean slower.
2. First-word latency: the formula. To compare two kits, turn the CL into time: first-word latency in nanoseconds = CL × 2000 ÷ data rate in MT/s [10]. A DDR5-6000 CL30 kit gives 30 × 2000 ÷ 6000 = 10.0 ns. A DDR5-4800 CL40 kit gives 40 × 2000 ÷ 4800 = 16.7 ns. Engadget reports that JEDEC's standard timings keep this value near 14 ns from DDR5-3200 to DDR5-8800 [10]. We could not check that claim, because JEDEC sells its standard for a fee. The table after this list is our arithmetic from the labels. It is a rough yardstick: real access time also depends on other timings and on the processor, so benchmarks decide, not this number alone.
3. What JEDEC sets as the base. JEDEC is the standards body for memory. It published the first DDR5 standard on 14 Jul 2020 and said DDR5 was expected to launch at 4.8 Gbps, against 3.2 Gbps at the end of DDR4's life [1]. Its update JESD79-5C of 17 Apr 2024 defines timing parameters up to 8800 Mbps. The previous version covered timings only up to 6400 and part of the core timings up to 7200 [2]. A module without a profile starts at the standard settings. Intel says XMP modules boot with the default JEDEC settings first [9], and Corsair says that without a profile the speed falls back to the default of the DDR5 standard [13].
4. What AMD and Intel officially support. These are the speeds the processor makers list on their own pages, as of 25 Oct 2026. The table after this list collects them. AMD lists the values per module count and rank: the Ryzen 7 7800X3D page shows DDR5-5200 for two single-rank or two dual-rank modules and DDR5-3600 for four [3]. In AMD's listing, more modules mean lower speed.
5. What EXPO and XMP do. Both are profiles stored on the memory module. Intel XMP overclocks DDR5 modules to speeds above the standard settings, with preset speed, timing and voltage that the memory vendor tested; XMP 3.0 for DDR5 supports up to five profiles, three from the vendor and two that you can rewrite [9]. AMD EXPO is the equivalent for AM5 boards and is described as a way to make memory overclocking user-friendly [7]. Neither maker calls it anything other than overclocking. A 2022 report says EXPO and XMP are not the same and not compatible [14], and Corsair notes that the option names depend on the platform [13]. Buy a kit whose label names the profile your platform reads, and check the motherboard's compatibility list.
6. Why 6000 shows up everywhere on AMD. AMD's own game-performance footnotes name DDR5-6000 kits: a G.Skill DDR5-6000C30 kit with EXPO on the Ryzen 7 7800X3D page (the test systems there use a Ryzen 9 7950X and 7950X3D) [3], and 2 × 16 GB of DDR5-6000 for the Ryzen 7 9800X3D [19]. Third parties explain why: TechSpot says DDR5-6000 lets Zen 4 processors keep the memory and the memory controller at a 1:1 ratio [11], and Engadget says the controller can follow the memory clock up to roughly 3000 MHz, above which it drops to half speed and adds delay [10]. Engadget adds a case: a DDR5-6000 CL30 kit finished ahead of a DDR5-6400 CL32 kit in Tom's Hardware's test, although both have 10.0 ns first-word latency [10]. We did not check that test. The lesson holds on its own: equal latency on paper does not mean equal results.
7. What it does for frame rates. Here are dated numbers with their platform. TechSpot (26 Jul 2024) tested 13 games at 1080p with a GeForce RTX 4090, a setup chosen to make the processor the bottleneck [11]. The table after this list shows the average over 13 games compared with the base memory. TechSpot points out that the 3D V-Cache part reacts less to memory than the 7700X, and that the 1080p setup shifts the load to the processor, while a graphics card that is the bottleneck at your resolution takes the gain away [11]. It did not recommend Intel's 13th and 14th generation at the time because of stability issues and included the 14700K "more out of interest" [11]. The test is over two years old at 25 Oct 2026, so it does not reflect later BIOS updates or newer processors. Engadget (4 Oct 2026) says no game requires a particular DDR5 speed and that DDR5-7200 and DDR5-8400 kits only marginally improve frame rates in processor-bound play [10]. PCWorld (3 Oct 2025) says up to 10 percent from a tuned DDR5-6000 kit with low latency, without naming a test setup [12]. We cannot verify that figure, so treat it as an opinion. Prices change quickly, and Engadget notes unusually high RAM prices in October 2026 [10]. Compare kits on the day you buy, with the price and its date.
8. Stable, or just booting. A profile that loads is not a profile that is stable. Update the BIOS first, because new versions improve memory compatibility [11]. Use one kit of two modules: NZXT's support page says profiles are meant for a single kit and that mixing kits risks instability, and it recommends Profile 1 when a kit shows several [15]. Check your motherboard's qualified vendor list (QVL) [12][15]. Then test. MemTest86 from PassMark boots from a USB stick and supports DDR5 with XMP 3.0 and EXPO profiles [16]. A PC builder's support page says its default run is four passes and takes several hours, and that even one error means a fail [17]. If errors appear, switch the profile off and run again: a pass at the standard settings points at the profile, not the module [17]. PassMark's manual advises to leave a margin and not to settle on the highest setting that barely passes [18].
9. Warranty and risk. AMD states that overclocking memory, including altered timings and voltage, voids any applicable AMD product warranty, even when enabled through AMD hardware or software [7][8]. Intel says that enabling XMP is a type of memory overclocking and may void the processor warranty when used beyond the given specifications [9]. NZXT says that enabling these profiles does not affect its own system warranty, but it cannot guarantee stability [15]. These are statements of the companies, not legal advice, and which one applies to you depends on your processor, board, memory and seller. Read the terms of each.

First-word latency at a glance

Official memory speeds, as of 25 Oct 2026

A dated test: what faster memory did

TechSpot, 26 Jul 2024, 13 games at 1080p, GeForce RTX 4090 [11].

A worked example: a Ryzen 7 7800X3D build

This is an example, not a report on our systems. We make no statement about the memory in our homelab or our workstations (as of 25 Oct 2026). We did not install the kits named here.

Say you build around a Ryzen 7 7800X3D. AMD's page lists DDR5-5200 as the official speed with two modules and DDR5-3600 with four [3]. You pick a 2 × 16 GB kit labelled DDR5-6000 CL30 with an EXPO profile. Its first-word latency is 30 × 2000 ÷ 6000 = 10.0 ns, while a DDR5-5200 CL40 kit gives 40 × 2000 ÷ 5200 = 15.4 ns. That is a gap of 5.4 ns on paper. In TechSpot's dated test, the same pair of kits gave +4 percent on average with a 1080p, RTX 4090 setup [11]. If your games run at 1440p or 4K on a graphics card that is already the limit, expect less. You update the BIOS, enable EXPO, let the board reboot, run MemTest86 overnight and look for zero errors [16][17]. AMD says that this is overclocking and that the warranty position changes [7][8]. Whether a few percent is worth the price difference on the day you buy is your call, and with a 3D V-Cache part the data say the premium has less to buy [11].

Checklist

• Read your processor's page at AMD or Intel and write down the official memory speed for two modules and for four.
• Choose two modules, not four, unless you need the capacity, and accept the lower official speed if you pick four.
• Convert each candidate kit to nanoseconds with CL × 2000 ÷ data rate and compare the results side by side.
• Pick a kit whose label names the profile your platform reads (EXPO for AMD boards, XMP for Intel boards) and look it up in the motherboard's QVL.
• Compare benchmarks only with their date, processor, graphics card and resolution, and ignore percentages that name none of these.
• Update the BIOS before you enable the profile.
• Enable the profile in the BIOS and, if the kit lists several, start with Profile 1.
• Run MemTest86 for the full default run and treat any error as a failed test.
• Switch the profile off and test again if errors appear, then decide between a lower speed and a different kit.
• Read the AMD or Intel warranty statement and your seller's terms before you rely on a profile, and keep the receipt.
• Compare prices on the day you buy, with source and date, because we list none.

Related pages: Memory / RAM; CPU / processor; ECC memory (gaming kits are not ECC modules).