Manual memory tuning can deliver meaningful performance gains, but it is also one of the most tedious forms of PC optimization. AMD's new EXPO Ultra Low Latency profiles aim to provide many of those benefits without the hours of tweaking, rebooting, and stability testing.
G.Skill is among the first manufacturers to support the new standard with DDR5 kits that include additional tuned timings beyond a conventional EXPO profile. In theory, that means lower latency and better performance straight out of the box.
You can, of course, manually tune many of these timings on almost any memory kit. The advantage of having a manufacturer such as G.Skill do it is validation. These modules use binned memory chips that have been tested and verified to run tighter timings at lower voltages while remaining stable.
In other words, EXPO ULL offers a shortcut around both the "silicon lottery" and the usual tuning grind. Finding the absolute floor for your latencies requires adjusting hexadecimal values, rebooting, and running memory stress tests for hours, or even days, to ensure the system will not crash. For those who simply want to game, this process is not exactly a fun time.
For Science, Not a Buying Recommendation
Speaking of fun... we have a few quick disclaimers for those who might find themselves triggered by this review. First and foremost, we strongly recommend that you do not purchase DDR5 memory right now, or possibly any DRAM, if you can avoid it. When purchasing memory is unavoidable, buy the absolute cheapest kit that meets your needs.
The point of this article is not to sell you on this particular memory, at least not right now. Rather, we are simply looking at what AMD's new EXPO Ultra Low Latency kits have to offer. So, if you are upset that we are testing a new DDR5 memory kit in the current climate, think of this as a test conducted purely "for science" and nothing more.
What EXPO ULL Actually Changes
With that out of the way, what does EXPO ULL memory offer over regular EXPO memory?
Put simply, it is a new memory profile standard being introduced as part of the EXPO 1.2 update. It builds upon AMD's existing EXPO technology by offering even tighter memory timings, with the goal of reducing latency at sweet-spot frequencies by around 5 to 7 nanoseconds.
AMD has achieved this by allowing memory manufacturers such as G.Skill to tighten subtimings including tREFI, tRRDS, and tWR.
tREFI is the refresh interval, and this value has been increased substantially. By extending the time between mandatory memory refreshes, the RAM spends less time performing maintenance and more time actively processing CPU requests, which can contribute significantly to lower latency.
tWR is the write recovery time. Lowering this timing improves how quickly the memory can transition from a write operation to a precharge command.
The third new option memory manufacturers can adjust is tRRDS, which represents the row-to-row delay for short operations. This subtiming is reduced to allow faster consecutive row activations within the same bank group.
There is also a fourth SPD setting that can be adjusted: VDDP voltage. By locking the VDDP voltage, companies such as G.Skill can help ensure stability for the memory controller inside Ryzen processors when running these more aggressive subtimings.
The kits we have on hand for testing are the G.Skill Trident Z5 NeoX RGB DDR5-6000 CL36 and Trident Z5 NeoX RGB DDR5-6000 CL30. The CL36 kit features matte-black heat spreaders, while the CL30 kit uses classier-looking piano-black, or gloss-black, heat spreaders.
Memory Timings EXPO vs ULL
The important part, of course, is the memory timings. Here is how the sweet-spot DDR5-6000 CL30 memory we normally use for testing compares with the new EXPO ULL kits.
First, G.Skill has taken this opportunity to tighten tRAS, reducing the value from 96 to 76 for the CL36 kit and to just 32 for the CL30 kit.
In addition to tREFI, tRRDS, and tWR, a number of other timings have been tuned, including tRC, tRRDL, tFAW, and tWTRL. The timings used here look very similar to what Buildzoid refers to as his lazy SK Hynix timings, and all three kits use SK Hynix memory.
G.Skill DDR5-6000 Memory Timings
| Timing | EXPO CL30 | CL36 ULL | CL30 ULL |
|---|---|---|---|
| RFCns | 294.6667 | 266.6667 | 166.6667 |
| Frequency | 6000 | 6000 | 6000 |
| Cmd2T | 1T | 1T | 1T |
| CL | 30 | 36 | 30 |
| RCDWR | 38 | 36 | 38 |
| RCDRD | 38 | 36 | 38 |
| RP | 38 | 36 | 38 |
| RAS | 96 | 76 | 32 |
| RC | 134 | 112 | 68 |
| RRDS | 8 | 4 | 4 |
| RRDL | 15 | 8 | 8 |
| FAW | 32 | 20 | 20 |
| WTRS | 8 | 6 | 6 |
| WTRL | 30 | 20 | 16 |
| WR | 90 | 72 | 48 |
| RDRDSCL | 8 | 8 | 8 |
| WRWRSCL | 23 | 23 | 23 |
| CWL | 28 | 34 | 28 |
| RTP | 23 | 12 | 12 |
| RDWR | 21 | 20 | 22 |
| WRRD | 8 | 8 | 8 |
| RDRDSC | 1 | 1 | 1 |
| RDRDSD | 8 | 8 | 8 |
| RDRDDD | 8 | 8 | 8 |
| WRWRSC | 1 | 1 | 1 |
| WRWRSD | 15 | 15 | 15 |
| WRWRDD | 15 | 15 | 15 |
| STAG | 7 | 7 | 7 |
| STAGsb | 0 | 0 | 0 |
| MOD | 42 | 42 | 42 |
| MODPDA | 32 | 32 | 32 |
| MRD | 42 | 42 | 42 |
| MRDPDA | 32 | 32 | 32 |
| RFC | 884 | 800 | 500 |
| RFC2 | 480 | 480 | 402 |
| REFI | 11677 | 46707 | 46707 |
| XP | 23 | 23 | 23 |
| PHYWRD | 6 | 6 | 6 |
| PHYWRL | 16 | 22 | 16 |
| PHYRDL | 36 | 36 | 36 |
| WRPRE | 3 | 3 | 4 |
| RDPRE | 3 | 4 | 4 |
| RDPOST | 1 | 1 | 1 |
| WRPOST | 1 | 1 | 1 |
| REFIns | 3892.333 | 15569 | 15569 |
AMD claims that EXPO ULL can improve gaming performance over previous EXPO configurations by 4% when using non-X3D CPUs such as the Ryzen 7 9700X. Based on that, we can assume there will be little to no improvement for those using X3D CPUs, at least in most games. That is not really shocking, as we already know that X3D processors are typically less sensitive to memory performance.
Still, for this initial test, we decided to test both the Ryzen 7 9800X3D and Ryzen 7 9700X across five games at 1080p and 4K.
The extreme resolution testing has been included simply to demonstrate what happens when the results become more GPU-limited, which we realize will be obvious to many of you. Although we find GPU-limited CPU testing misleading in a traditional CPU benchmark, this is more of a system-level benchmark than a CPU review. Since we are only testing a handful of games, including the additional data is not a problem.
With that, let's get into the results.
Test System Specs
| CPU | AMD Ryzen 7 9800X3D AMD Ryzen 7 9700X |
| Motherboard | MSI MAG X870E Tomahawk Max (BIOS 7E59v5A32) |
| Memory | G.Skill Trident Z5 RGB 32GB DDR5-6000 CL30 G.Skill Trident Z5 NeoX RGB DDR5-6000 CL36 G.Skill Trident Z5 NeoX RGB DDR5-6000 CL30 |
| Graphics Card | Asus ROG Astral RTX 5090 OC Edition |
| ATX Case | MSI Prospect 700R |
| Power Supply | Kolink Regulator Gold ATX 3.0 1200W |
| Storage | TeamGroup T-Force Cardea A440 M.2 PCIe Gen4 NVMe SSD 4TB |
| Operating System | Windows 11 25H2 |
| Display Driver | Nvidia GeForce Game Ready Driver 610.74 WHQL |
Benchmarks
Cyberpunk 2077: Phantom Liberty
First up, we have the Cyberpunk 2077: Phantom Liberty results using the Ultra Ray Tracing preset at 1080p. With the 9800X3D, the new ULL memory kits offer virtually no additional performance in this game. The CL30 Ultra Low Latency kit delivers just one extra frame, amounting to a mere 1% increase.
However, when we look at the 9700X, we see a much more substantial performance improvement from both ULL kits. The CL36 kit was 7% faster, while the CL30 kit was 8% faster. These are not massive margins, but they do bring the non-X3D Ryzen 7 part much closer to the 9800X3D, which is admittedly GPU-limited at these quality settings.
That being the case, there is not much to see when we switch to the 4K data. The takeaway is that if you are happy gaming at less than 60 fps, your CPU and RAM combination does not matter much because you will almost always be heavily GPU-limited. With that, let's move on.
Marvel Rivals
Moving over to Marvel Rivals, we will again start with the 1080p data, this time using the Medium preset. Looking at the Ryzen 7 9800X3D, it is a familiar story. Throwing tighter memory timings at the 3D V-Cache chip barely moves the needle.
The CL30 Ultra Low Latency kit manages to deliver just six additional frames on average over the standard EXPO profile, which amounts to an insignificant 2% increase. There is a slight improvement in the 1% lows, which rise from 198 fps to 209 fps, but overall, it is nothing to write home about.
Things become much more interesting when we move to the standard Ryzen 7 9700X. Here, the tighter subtimings deliver a highly respectable performance uplift. The CL36 ULL kit boosts average frame rates by around 9%, while the CL30 ULL kit extends that to a 10% gain over the base EXPO configuration. Although it does not come close to eliminating the large gap to the 9800X3D, it is a very solid performance improvement for non-X3D owners.
Finally, increasing the resolution to 4K shifts the bottleneck almost entirely to the RTX 5090. As you can see, the margins disappear. Whether you are using the 9800X3D or 9700X, with standard EXPO or one of the tuned ULL kits, performance flatlines at around 130 fps on average, with 1% lows hovering near 110 fps.
At this resolution, the memory and CPU configuration are largely irrelevant, so let's move on to the next title.
Assetto Corsa Competizione
Moving over to Assetto Corsa Competizione, we tested using the Epic preset at 1080p, and the results are quite revealing. Starting with the Ryzen 7 9800X3D, the story remains largely unchanged. The large 3D V-Cache does much of the heavy lifting, meaning tighter memory timings provide little additional value.
The CL30 Ultra Low Latency kit delivers just a three-frame improvement on average over the standard EXPO kit.
However, with the standard Ryzen 7 9700X, subtiming tuning has a strong impact on frame rates. The CL36 ULL kit delivers a solid 8% uplift, while the CL30 ULL kit improves average frame rates by an impressive 10%, increasing performance from 181 fps to 199 fps.
More importantly, the 1% lows improve by 11%, rising from 153 fps to 170 fps. It does not catch the 9800X3D, but it extracts a substantial amount of additional performance from the non-X3D architecture.
What makes this title particularly interesting is what happens when we switch to the 4K data. Unlike other games, where the margins disappear almost entirely at 4K, Assetto Corsa Competizione is notoriously CPU- and memory-intensive, leaving us heavily CPU-limited even at 2160p when using the RTX 5090.
With the 9800X3D, the CL30 ULL kit improves average frame rates by 3%, while the 1% lows increase by 9%, rising from 177 fps to 192 fps.
Meanwhile, the 9700X sees an impressive 11% increase in average frame rates with the CL30 ULL kit, scaling from 178 fps to 198 fps. Because this game places such a heavy load on the processor, the 9800X3D retains a substantial 23% lead over the 9700X even at 4K.
For hardcore sim racers, it is clear that CPU and memory configuration matter considerably, regardless of the resolution.
Baldur's Gate 3
Moving on to Baldur's Gate 3, we tested using the Ultra preset at 1080p. Looking at the Ryzen 7 9800X3D, we see a break from the familiar pattern. Unlike the previous titles, the 3D V-Cache part responds fairly well to the tuned memory.
The CL30 Ultra Low Latency kit boosts the average frame rate from 199 fps to 217 fps, representing a solid 9% increase over the standard EXPO profile. The 1% lows also improve nicely, rising from 135 fps to 146 fps.
Moving down to the standard Ryzen 7 9700X, the memory scaling is substantial. Switching from the standard EXPO kit to the CL30 ULL profile increases the average frame rate from 121 fps to 139 fps, an impressive 15% performance gain. The 1% lows also climb from 91 fps to 107 fps.
It still cannot catch the 9800X3D, which remains well ahead, but that is a significant amount of additional performance for 9700X owners.
Cranking the resolution up to 4K, Baldur's Gate 3 proves to be remarkably CPU-intensive, meaning the performance margins do not disappear entirely, as they did in some of the other games tested.
With the 9800X3D, the CL30 ULL kit still manages to deliver a little more performance, increasing the average from 155 fps to 162 fps. Meanwhile, the 9700X continues to scale well with tighter timings at 4K, rising from an average of 120 fps to 135 fps with the CL30 ULL kit, representing a 13% uplift. The 1% lows also improve from 89 fps to 100 fps.
Marvel's Spider-Man 2
Moving over to Marvel's Spider-Man 2, we tested using the Very High preset with Ultimate Ray Tracing enabled at 1080p. Here, we see another example in which reduced memory latency improves performance for the 9800X3D.
Typically, the large 3D V-Cache helps the processor overcome memory bottlenecks, but the heavy ray tracing workload in this title appears to make it particularly sensitive to subtimings. The CL30 Ultra Low Latency kit provides the 9800X3D with a significant 11% performance uplift over the standard EXPO profile, increasing the average frame rate from 107 fps to 119 fps. The 1% lows improve by the same 11% margin.
Turning to the standard Ryzen 7 9700X, the memory tuning continues to deliver meaningful gains. The CL36 ULL kit improves average frame rates by 9%, while the CL30 ULL configuration delivers a 10% increase, taking the average from 89 fps to 98 fps. The 1% lows also receive a welcome 12% boost.
Interestingly, because the 9800X3D scales so well in this title, the tuned 9700X does not close the gap to the flagship chip. The 9800X3D remains firmly ahead, demonstrating how important both cache capacity and memory latency can be when handling demanding ray tracing workloads.
However, at 4K, the rendering burden shifts almost entirely to the RTX 5090. With Ultimate Ray Tracing enabled, the GPU reaches a hard limit, and nearly every configuration delivers the same 77 fps average with 1% lows of 50 fps.
Apart from a negligible two-frame drop with the standard 9700X configuration, the choice of CPU and memory is practically irrelevant at this resolution, as the GPU is doing nearly all the heavy lifting.
Memory Tuning Made Easy
For those running a standard Zen 4 or Zen 5 processor, the takeaway is crystal clear: memory tuning can improve performance by a meaningful margin when CPU limited. Across our 1080p testing, moving from a standard EXPO profile to the tightly tuned CL30 ULL kit increased average frame rates by 10% in Spider-Man 2, Marvel Rivals, and Assetto Corsa Competizione, with an especially impressive 15% uplift in Baldur's Gate 3.
Although it does not entirely close the gap to the 9800X3D, optimizing memory subtimings is essential for anyone looking to extract every last drop of performance from a non-X3D processor. None of this is particularly new, of course. Around three years ago, we published an in-depth article examining memory tuning and found strong gains across a number of titles.
The feedback we received from many of you was essentially, "That is all well and good, but I do not want to spend time tuning and validating my memory for a 10% to 15% improvement when CPU-limited." That is entirely fair, and we completely understand. This is why these new EXPO ULL memory kits could be particularly appealing to those of you shooting for maximum performance, without all the fuss.
For those using an X3D processor, the story is a little more nuanced. In many titles, such as Cyberpunk 2077, Assetto Corsa Competizione, and Marvel Rivals, the large L3 cache acts as a substantial buffer, largely masking memory latency bottlenecks and making the ULL kits virtually irrelevant.
However, in exceptionally demanding CPU workloads such as Baldur's Gate 3 and Spider-Man 2, the 9800X3D still delivered respectable performance improvements of 9% and 11%, respectively, when paired with the CL30 ULL kit. When playing heavily CPU-bound games, the premium memory can still do some meaningful work.
Then there is the 4K data. When playing graphically demanding titles and prioritizing visual quality over frame rates, you will almost always run headfirst into a GPU bottleneck, even with a monster such as the RTX 5090. In Cyberpunk 2077, Spider-Man 2, and Marvel Rivals, frame rates were effectively flat across all configurations, making the choice of CPU and memory timings largely irrelevant.
As always, however, there are exceptions. Hardcore sim racers playing titles such as Assetto Corsa Competizione, along with those playing other simulation-heavy or esports games, can still encounter CPU and memory bottlenecks at 4K. In those scenarios, the ULL kits continued to provide measurable and worthwhile scaling for both the 9700X and 9800X3D.
The obvious question, then, is whether the performance uplift is worth the price. AMD's David McAfee has stated that the new AMD EXPO Ultra Low Latency kits would be introduced at pricing similar to standard AMD EXPO memory kits. However, early reports suggested substantial premiums for the new memory, in some cases reaching as high as 80%.
G.Skill explained to us that this was due to the recent increase in DRAM pricing, which immediately affects newly released products. Existing memory kits, meanwhile, were purchased by retailers at earlier and somewhat lower prices. Those prices are still massively inflated, but they are not quite as bad as the latest pricing.
Essentially, G.Skill is saying that DRAM pricing was already bad and has only continued to worsen. Early reports pointed to EXPO ULL kits priced at around $1,100. However, kits such as the CL36 NeoX model we tested can now be purchased for $530, just $30 more than a comparable G.Skill CL36 EXPO kit. That's a 6% premium, which is not unreasonable. Realistically, though, the entire situation is bad, as paying $500 or more for 32GB of DDR5 is horrendous.
Still, this does suggest that EXPO ULL memory may command only a relatively small premium. It also makes sense that buyers would pay somewhat more for memory that has been validated and guaranteed to operate with tighter timings.
Circling back to the disclaimer from the beginning, we strongly recommend avoiding DDR5 purchases right now wherever possible. It is unclear how long pricing will remain this absurd, but personally, I would hold out.
Alternatively, you could purchase two G.Skill Flare X5 8GB DDR5-6000 CL30 modules for $240. If I had to buy memory right now, that is what I would do. For the most part, you can still get by reasonably well with 16GB of memory.