Intel’s Z690 platform with DDR5 includes many new features and overclocking possibilities for those who aren’t scared of the bios. If you’re the type who wants to get the most out of your new platform, memory overclocking is worth checking out. Overclocking your memory is safe, fun, and it can give you up to a 20% increase in performance for some tasks. But it’s not always a simple process, and adjusting the memory timings can be incredibly daunting if you have no reference or guide. However, with our simple instructions, we will demystify memory overclocking and give you the tips and tricks to get started.
DDR5 vs. DDR4
We want to take a minute and highlight some of the main differences between DDR4 and DDR5. For generational memory changes, such as DDR3 to DDR4, the primary focus has been increasing the frequency. Generational leaps typically mean lower power usage, higher density, and increased latency. However, when it comes to DDR5, we don’t simply see an increase in frequency like previous generational leaps. This generation sees a redesign of the entire data bus, channel architecture, and power delivery systems. As expected, the new memory also brings higher die density, lower power usage, and increased timings to the table.
The JEDEC rating for DDR4 ranged from 1600 MT/s up to 3200 MT/s at the end. The new DDR5 standard starts at a JEDEC rating of 4800 MT/s. We know that there will be higher JEDEC ratings to be released as DDR5 matures. Right now, we have seen ratings up to 7000 MT/s marked for the future of DDR5.
The highest possible die density for DDR4 was 16 GB. In recent times, Samsung released 256 GB LRDIMM and RDIMM modules based on their 16 GB die. The new DDR5 standard doubles the die size to 32GB for UDIMM. Based on the 32 GB die size and using Samsung’s example of the 256 GB DDR4 module, it’s not a stretch to imagine that we will see 512GB of memory on a single module in the future.
The very best that DDR4 had to offer, in terms of JEDEC rating at least, was 3200 MT/s running at CL22-22-22. Although there are lower data rates in the JEDEC specification for DDR5, the primary rating for desktop users at launch will be 4800 MT/s running at CL40-40-40.
Another fundamental change with DDR5 is the way it gets its power. Until now, all previous generations of DDR memory received its voltage, pre-regulated, from the motherboard. The motherboard was responsible for taking the 12 V input and reducing that to a usable voltage for the memory controller, such as 1.35 v. Because of this, high-end motherboards generally provided better voltage control and thus resulted in better memory overclocking capabilities. The new DDR5 specification moves the voltage regulation from the motherboard to the individual memory stick. The motherboard provides 12 V host input to the respective memory modules, and the module takes care of the voltage control. Known as the PMIC, the new memory voltage controller is responsible for ‘bucking’ the voltage down to what’s required by the memory. We’ll get into this more later on.
The changes in the functional architecture are what make DDR5 unique from all other memory releases. Up until now, one stick of memory contained one internal memory channel. If you wanted to run dual-channel memory, for instance, you would need two memory modules. Mainstream motherboards, such as Intel’s Z590, had a maximum of 2 DIMMs per channel (2 DPC). Therefore, if you outfitted the board with 2 or 4 modules, you would be running dual-channel at maximum. The significant new change is that DDR5 contains two memory channels within one module. Therefore, with two sticks running in a Z690 motherboard, you get quad-channel memory. More memory channels dramatically increase performance for memory-intensive workloads. Mainstream motherboards still support only 2 DPC, but now that means you will be running quad-channel with 2 or 4 memory sticks.
We typically only see the Error Correction Code (ECC) implemented in the server and workstation world. Traditional performance DDR4 UDIMM does not come with ECC capabilities. The CPU is responsible for handling the error correction. Due to the frequency limitations, we’ve only seen ECC on lower-spec kits. DDR5 changes everything in this regard because ECC comes standard on every DDR5 module produced. Therefore, the system is then unburdened as it no longer needs to do the error correction.
It all starts with understanding your hardware. Knowing what DDR5 IC you are dealing with is a good indicator for setting up the memory. There are a few ways to determine the IC. The simplest method is by opening up the memory tab of CPUz, which will tell you what the brand is, but it won’t tell you anymore. You simply cannot attempt overclocking without this piece of software, so download it now if you don’t have it installed already.
For a more detailed look at your memory, we’d recommend using Thaiphoon Burner. This software reads the SPD values, showing the die revision, PCB layout style, and rank structure. We want to express a word of caution. The information you receive via software tools depends on the SPD being programmed correctly from the factory, which doesn’t always happen. Currently, Thaiphoon Burner is not compatible with DDR5, but we expect to see updates shortly.
So now I know the IC brand name; how does that help me? Below is a table of basics to help you understand what’s possible. Please keep in mind that the information below is highly subjective and may dramatically change as we progress into the subsequent phases of DDR5.
|Preliminary DDR5 Launch Overclocking
|Frequency Scaling||4800 – 7000||4800 – 5600||4800 – 7000|
|VDD Voltage Scaling||1.10 V – 1.40 V||1.10 V – 1.25 V||1.10 V – 1.40 V|
|VDDQ Voltage Scaling||1.10 V – 1.50 V||1.10 V – 1.35 V||1.10 V – 1.50 V|
XMP – Memory Overclocking Profile
If you’re reading this guide and attempting to overclock your memory, then you likely have a memory kit that comes with an XMP 3.0 rating; not all of them do. When you first install the memory and don’t configure the bios, it will run at the JEDEC rated speed. For most DDR5 UDIMM at launch, the JEDEC rating is 4800 CL40-40-40. The JEDEC standard should work with all platforms and all possible configurations with no bios editing necessary. The drawback of that excellent compatibility, it’s slow.
If you’re running JEDEC memory speeds, then your first step is to apply the XMP profile. Enter the bios and enable the XMP profile. Go into the operating system, open CPUz, and verify the rated speed using the ‘Memory’ tab.
If your XMP rating is 5200 MT/s, and CPUz shows a DRAM frequency of 2600 MHz, don’t worry; that’s the correct value. DDR stands for ‘Double Data Rate,’ which means data transfers occur on both the rising and falling edges of the clock signal. In other words, the actual frequency of 2600 Mhz is 5200 MT/s.
Making your memory go faster is excellent, but it’s easy to take things too far. To take the XMP training wheels off and overclock the frequency or timings, you will need to test each change you make. Whether it’s the final minutes of a DOTA 2 battle or a virtual video job interview, the last thing you want is for the computer to crash because of an aggressive memory overclock. Testing is critically important, don’t skip this step.
The objective is to make one change in the bios and then enter the OS and run a simple stress test. Going through this process will inevitably find that some memory profiles appear to work correctly, but running a quick stress test will fail. We recommend running the AIDA64 stress test with ‘Stress system memory’ enabled for about 10 minutes.
Please keep in mind that just because your memory profile passes 10 minutes of the AIDA64 stress test, it still might be unstable. The ultimate test for your memory is running MemTest. There are a few different versions of this age-old software, but we’re going to focus on what can be done quickly in the OS.
We recommend downloading and running HCI MemTest, a free memory stressing program that puts an exceptional amount of load on the memory and the CPU memory controller. If there are any issues with the overclock, it should be evident within the first 20 minutes of running this stress test. As seen below, an aggressive overclock failed within the first minute of MemTest. We used a wrapper program called RunMemtestPro 4.0, which uses the HCI MemTest program and upgrades the user interface. We won’t link the wrapper program for security reasons, but a simple Google search should bring it up quickly if you’d like to try it out.
Before we go further, please understand that you may irreversibly damage your components from increased voltage and overclocking stress. Some memory vendors expressly void the warranty if you overclock the memory, so please proceed if you assume all responsibility. It’s especially concerning with DDR5 because they haven’t been in the hands of overclockers long enough to expose any inherent weakness or long-term degradation.
So you’ve applied the XMP profile, and it’s working perfectly. By this, we mean that you have gone through stability testing and basic functionality testing without any issues. The XMP profile is a good start, and you’re already more advanced than most computer users out there. But you’re here because you want to go beyond the XMP profile, and we’re here to encourage you to do so.
Step 1: Frequency
The first step on your overclocking journey should be increasing the DRAM frequency. With the XMP profile still applied, go into the bios and select the subsequent available frequency beyond the XMP. For example, our XMP rating is 6400, so we will choose 6600 as the initial test.
In the picture below, there’s more to the story than just setting 6600. There are two other factors to consider. First, you need to decide which BCLK/DRAM ratio to use. The two options are 100:100 or 100:133. For Alder lake and Z690, there is no reason to use the 133 ratio over the 100 one. Previously, on Z590 and Rocket Lake, choosing the 133 ratio significantly impacted top-end overclocking frequency, but we haven’t seen that effect yet on Alder Lake. When you are just starting, we recommend choosing the 100:100 ratio to keep things simple.
Next, you will need to choose the gear setting represented in the picture below by ‘G2’ or ‘G4’. The gear setting is the memory controller ratio to the DRAM frequency. In ‘G2’ mode, or Gear 2, the memory controller frequency will be exactly half of the DRAM frequency. As you may have guessed, the ‘G4’ setting makes the memory controller frequency one-quarter of the DRAM frequency. What does this mean for overclocking? Well, the idea is that if you run the memory controller at lower clock speeds, it will allow higher memory clocks speeds. The reality is that Gear 2 is all that you need to max out DDR5 frequency on air. We recommend choosing the G2 option when you are just starting.
So now you know that your first move with overclocking should be to choose ‘6600 (66×100.00×1.00) G2’. Make that change, hit F10, and attempt to enter the OS and check for memory stability. If that proves to be stable, choose the next highest frequency available to you, and try again. Repeat this cycle until you find the system is no longer stable. At this point, it’s time to add some voltage, so move on to step 2.
Step 2: Corse Voltage Tuning
So you know the maximum frequency that your kit will run without increasing the voltage, this will be around 5400-5600 MHz for Micron-based kits, and for Hynix, it’s in the 6000-6600 MHz range. Most of you have already extracted extra MHz from your kit, and you didn’t even need to increase the voltage, so congratulations are in order and a good place to stop for many beginners. However, if you want to push things even further, it’s time to add some voltage.
The next step is to increase the voltage beyond XMP and repeat step 1. The goal is to increase the frequency more with the addition of voltage. All told, five voltage levels affect the overclocking capabilities. Tuning the voltage levels can be time-consuming and challenging, but even a tiny increase in voltage can dramatically affect the overclocking results.
There are a few different approaches, but beginners usually have the best experience with a top-down approach. In this method, you set subjectively high voltage levels, lower them until the system becomes unstable, and then restore them to the last stable level. There are no established guidelines for safe voltage levels, but the following voltages are representative of the maximum safe voltage levels for DDR5 at this point. However, we would not recommend running the system 24/7 with this profile; the voltage levels are too high, especially CPU SA.
Step 3: Primary Timings
So you’ve found the maximum stable frequency with no voltage applied, and you increased the voltage to gain stability with more frequency; what’s next? If you want to improve performance, the next step is to massage the primary timings. The process is relatively straightforward on Alder Lake, but it’s incredibly time-consuming. We say it’s easy because we’ve found that the system can go from 100% stable to completely failing with one timing movement. There’s not much you can do to fix it—a slight departure from previous generations, which had more flexibility with timing stability.
The process is simple. Reduce one primary timing by two and attempt to run a quick stability test in the OS. If that proves to be unstable, increase it by one and repeat. If, for instance, you reduced the tCL from 40 to 38 and it’s stable in the OS, repeat the process and set 36. This trial-error method is time-consuming, but it’s the best way to approach the problem because you also learn the behavior of timings this way. It’s tempting to use a profile you may have found online, but we urge you to follow this slow method of adjusting one timing at a time. Below is a picture of a typical timing profile for Hynix.
Intel has a running list of the XMP 3.0 profiles that are available on the market. Below is an essential list, but it gives you an idea of the current capabilities: DDR5 Memory Profiles.
Here is the result of our early testing concerning memory profiles and frequency scaling. It should go without saying, but we will repeat it. We have a preliminary result, and the timings shown here are highly subjective. Your results may be vastly different.
|Preliminary 24/7 Stable Timing Profiles
Step 5: Secondary and Tertiary
We want to keep this review simple and easy to follow; therefore, we won’t be doing a deep-dive down the timings rabbit hole. However, there are a few sub-timings that are too influential and need attention.
Some of them are perhaps more influential than even primary timings or frequency overclocking. That’s a bold statement, but the reality is that each motherboard sets sub-timings differently, which can lead to wildly different efficiency results. Taking the time to edit these timings below is an essential step if you are looking to max out the performance.
All of the timings are influential and important; however, there’s a significant degree of difference in efficiency and benchmark scoring. We’ve done the leg work, so you don’t have to. Here are the heavy-hitters for DDR5.
|The Most Influential DDR5 Sub-Timings|
|tWR||32-72||Try to keep tWR the same as tRAS. Lower is better|
|tRFC2||300-600||Setting this too low can limit the top-end frequency|
|tRFCpb||180-400||Setting this too low can limit the top-end frequency|
|tFAW||18-32||You generally want this set as low as it will go|
|tREFI||6240-65535||You generally want this set as high as it will go|
|tRDRD_dg||7-15||Value of 8 if possible, or as low as it will allow|
|tWRWR_dg||7-15||Value of 8 if possible, or as low as it will allow|
Step 4: Fine Voltage Tuning
The final step is to tune the voltage again after you’ve finalized all of the timings. Once you’ve gone through and tuned your memory frequency and timings, we’d suggest that you take one final step regarding the voltages. It’s a good idea to run the lowest voltages possible to extend the life of your products. The best approach to the final voltage tune is to test one voltage option at a time and run 100% MemTest for each change you make. Testing like this is a slow process, but it can make a massive difference in the long run.
Does The Motherboard Effect Overclocking?
We know there’s a vast difference in overclock-ability with the DRAM modules themselves, but what about the motherboard? Overclockers always flock to 1 DPC motherboards, but is there a vast difference? What is 1 DPC, and why does it matter? One DIMM per channel (1 DPC) motherboards only have 1 DIMM slot for channel A and one for channel B. They are easy to spot because they only have 2 DIMM slots, compared to the standard 4.
Furthermore, they are usually at the top end of the performance motherboard product tack. In general, these motherboards overclock memory higher because the distance to CPU is shorter, and there’s no empty unterminated DIMM slot leading to a phenomenon called parasitic inductance. Additionally, motherboard manufacturers devote extra time and resources such as BIOS optimizations to achieve higher overclocking potential on flagship 1 DPC motherboards.
In this section, we’re going to compare the overclocking potential of our MSI Z690 MPG WiFi with the ASRock Z690 AQUA OC. The MSI board has 4 DIMM slots and is 2 DPC, while the ASRock board has 2 DIMM slots and is 1 DPC.
The Micron results came as a bit of a surprise. Both motherboards topped out at 5600 MHz with Micron-based DDR5. All of the potentials of the Z690 AQUA OC didn’t do much for the overclocking capabilities of Micron memory. I think we can safely assume that Micron’s 5600 Mhz limitation is so low that motherboard optimization is not a significant factor. It’s worth noting, however, that the ASRock board managed tighter timings for the same voltage levels.
Here we see the power of 2-DIMM motherboards. When the frequencies rise above 6000 MHz, the motherboard becomes a much more significant factor. With our Hynix-based kit, the MSI 4-DIMM board topped out at 6400 CL36 @ 1.35 V. That may seem like a great result, but it’s the XMP frequency of our Team Group test kit. We’ll review the Team Group kit at a later date. It’s important to realize that the bios have not had time to mature yet, so there will likely be improvements in the future. Harnessing the power of the ASRock Z690 AQUA OC, we managed 6666 CL32 at the same voltage.
So the motherboard does indeed play a vital role in memory overclocking. We compared a high to midrange Z690 4-DIMM board to one of the very best 2-DIMM boards. Imagine the results; it would have been much worse if we used a low-tier Z690 4-DIMM board.
We want to cover memory cooling; briefly, a topic of much debate as DDR5 now has a 3-phase PMIC voltage controller on each memory module. The PMIC adds additional heat to the DRAM module, so there’s more cause for concern. That said, our experience is that the heat output is nominal and not worth special attention. In a worst-case scenario test, with no airflow and 10 minutes of stress, we’ve seen a maximum temperature of about 40 °C.
What can you gain from aftermarket memory heat sinks? In our experience, there’s absolutely no overclocking increase from a better cooling solution. Our suggestion is to make sure you have a decent fan blowing directly on the memory during all of your overclocking endeavors. The only reason to mount aggressive memory heat sinks, such as the Bitspower ones you see below, is to use an extreme cooling container.
If there’s one takeaway from this article, it’s that memory overclocking cannot be rushed. It can be a slow process, and you must be willing to devote some time to the bios if you want to take the XMP training wheels off. It’s essential to test everything along the way and use your best judgment. If increasing the frequency 200 MHz requires a voltage increase from 1.20 V to 1.35 V, then it’s probably not worth the added system stress to do so. When done right, though, most memory kits have enough overclocking headroom to increase performance with minimal voltage increase. The most crucial aspect of overclocking is to have fun with it and enjoy the process. We encourage you to enter the bios and explore the wonderful world of memory overclocking!
David Miller – mllrkllr88