AMD promises 13% uplift with new EXPO ‘Ultra Low Latency’ overclocking on DDR5 DIMMs — automatic memory overclocking delivers 4% improvement over standard EXPO, says AMD

AMD is bringing a new automatic memory overclocking feature out, dubbed EXPO Ultra Low Latency (ULL). The new tool promises a 13% uplift in average performance across over 30 games compared to JEDEC standard speeds for DDR5, as well as a 4% uplift compared to standard EXPO, at least according to AMD’s internal benchmarks. The company hasn’t said when ULL will be available, but it has several memory partners lined up to support the feature.

We don’t know much about how EXPO ULL works, and more specifically, where AMD is finding improvements compared to standard EXPO. Despite that, AMD shared some broad benchmarks looking at EXPO ULL performance on a Ryzen 7 9700X CPU in over 30 games. AMD didn’t share resolution (presumably 1080p) or settings details for the titles, outside of the fact that they were tuned for “best performance.” You can see the full configuration in the gallery below.

Image 1 of 3

Compared to JEDEC-standard DDR5-5600 CL46, AMD shows EXPO running at DDR5-6000 running 9% faster, and DDR5-6000 CL30 with EXPO ULL running 13% faster. Looking at 1% lows, AMD claims a performance improvement for EXPO and EXPO ULL of 11% and 15%, respectively. AMD didn’t clarify the CAS Latency of the standard EXPO memory, but it notes that it tested at CL28, CL30, and CL36 for both EXPO and EXPO ULL.

Latest Videos From

Watch full video here:

AMD simply says that EXPO ULL is “coming soon” from major memory partners, including G.Skill, Kingston, Klevv, Lexar, Origin Code, TeamGroup, V-Color, and XPG.

There are still several lingering questions about ULL, which we hope to learn more about on the ground at Computex 2026. Presumably, ULL isn’t a feature that’s being backported to older EXPO DIMMs, but rather a new standard moving forward. AMD hasn’t confirmed that’s the case, however.

We also don’t know what ULL is actually doing to find 4% extra performance compared to standard EXPO. Given the wide swath of games and CAS Latency settings AMD tested at, there’s a good chance ULL won’t show a benefit in all games. We should see better uplifts in smoothness, however, as evidenced by AMD’s 1% low results.

Tom’s Hardware is on the ground in Taipei for Computex, where we’ll be meeting with AMD to learn more about EXPO ULL, along with the newly-announced Ryzen 7 7700X3D and reintroduced Ryzen 7 5800X3D.

Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.

TOPICS

Jake Roach is the Senior CPU Analyst at Tom’s Hardware, writing reviews, news, and features about the latest consumer and workstation processors.

5 Comments Comment from the forums

Zaranthos

This has partly rolled out with BIOS updates already as my Aoros Pro Ice now supports some of this. But to really take full advantage of this will require the next generation AMD CPU's with its new memory controller and new RAM kits with EXPO ULL profiles. With the BIOS update some sub timing settings may be unlocked so people can manually push existing RAM a little more.

I've pretty much concluded there is nothing to see here for me right now other than the fact that I have a high end motherboard that should have decent support for some of the future memory kits when paired with the next generation AMD chips. Right now RAM is too expensive anyway, new kits would have to be tested and validated with next gen CPU's anyway, and even if kits become available early there is little guarantee better kits won't come out when next gen CPU's are actually available.

Lower CAS timings are great and if I don't have to sell a kidney to afford next generation RAM kits paired with a next gen AMD CPU I would likely upgrade both at the same time. We shall see.
Reply
Mindstab Thrull

All I'll say is this:
At 60 fps, 4% is 2.4 frames. At 120, it's roughly 5 frames per second.
As for "time to perform job X" tasks (eg rendering), a 10 minute job saves 24 seconds.

For gaming, the performance might not be worth it unless it's "free", ie the stuff you're picking up or has supports it anyways. For work, it might be, as that means an hour is 2m24 saved, over 8 hours that's nearly 20 minutes (19m12s) - or a job that would normally be 8h20 is now 8h.

Just my two Sarpadian coppers' worth
Mindstab Thrull
Reply
bit_user

Mindstab Thrull said:
All I'll say is this:
At 60 fps, 4% is 2.4 frames. At 120, it's roughly 5 frames per second.
As for "time to perform job X" tasks (eg rendering), a 10 minute job saves 24 seconds.

For gaming, the performance might not be worth it unless it's "free", ie the stuff you're picking up or has supports it anyways. For work, it might be, as that means an hour is 2m24 saved, over 8 hours that's nearly 20 minutes (19m12s) - or a job that would normally be 8h20 is now 8h.
You can always make this argument about small performance gains, but the way to get a big performance improvement is by stacking up a lot of small ones. If someone is intent on getting the most performance out of their rig, 4% is too big to ignore.

That said, lower latency usually means higher voltages, which means hotter memory temperatures. So, people should be advised to monitor their memory temps and try to arrange some active airflow over the DIMMs.
Reply
Mindstab Thrull

@bit_user I'm assuming that the RAM situation is the only difference, and yes, adding up 4% in three different places (assuming they all stack multiplicatively and not just adding them) is about a 12.5% difference, which is huge. Otherwise, I don't see it making sense for most people without sufficient financial incentive.
Good call on the temps btw, hadn't even considered that. (y)
Reply
bit_user

Mindstab Thrull said:
@bit_user I'm assuming that the RAM situation is the only difference,
Eh, but you can probably make similar arguments about each potential improvement. The real question is how committed you are to wringing every last drop of performance.

For my part, I run mostly stock settings. I do a slight memory overclock, but I sort of compensate by using ECC memory to give me a safety margin. I'm definitely not the poster child of overclocking and performance tuning, but I do understand enough about the theory.

Mindstab Thrull said:
yes, adding up 4% in three different places (assuming they all stack multiplicatively and not just adding them) is about a 12.5% difference,
Whether or not they're multiplicative, additive, or something less really depends on the details. It most cases, I think they tend not to be multiplicative or even additive, but a bit less. For instance, let's say you overclock a CPU that's running something where the cores are already spending a lot of time stalled on memory accesses. In this case, your speedup from overclocking is going to be sub-linear, but reducing your memory latency will produce a greater speedup than if you did it on a CPU that wasn't overclocked.

Anyway, I'm just saying that what I've learned about performance tuning is that you rarely find a huge win, all by itself. The hardware and software has already been optimized well enough that there's not giant low-hanging fruit that gives you the effect of a CPU or GPU upgrade, all by itself. Usually, you need to stack multiple gains to achieve something noticeable.
Reply

Show more comments