Rocket Lake Overclocking: What’s New
Intel Rocket Lake is here and just like with every new generation, a new CPU means new overclocking features. In this article we’re exploring the 18 new and improved overclocking features of the Intel Rocket Lake CPU and the ROG Maximus XIII Apex motherboard. We’ll split it up the article in four different segments.
- First, let’s quickly go over the Rocket Lake architecture
- Then, dig into the new overclocking features and tools introduced by Intel
- Then, let’s have a look at some of the new overclocking features of the ROG Maximus XIII Apex motherboard
- Lastly, let’s talk about some overclocking expectations.
Intel Rocket Lake Architecture.
Intel’s 11th generation Core products for desktop, codenamed Rocket Lake, were officially introduced by Intel during the CES 2021 tradeshow in January and arrived to the market in March 2021.
Rocket Lake is the successor to Intel’s 10th generation Comet Lake processors. Rocket Lake sports a brand new CPU core architecture while still on the vastly improved 14nm process node. The CPU core is built upon the Cypress Cove architecture which is the backported version of Sunny Cove, a core designed for 10nm Ice Lake, with some additional performance improvements.
Due to the increased core size, the flagship Core i9-11900K offers up to 8 cores and 16 threads compared to its Core i9-10900K predecessor’s 10 cores and 20 threads. It does gain support for deep learning boost and AVX-512 instructions, a new and improved cache hierarchy, and up to 19% instructions per clock improvement.
Other new features include slightly elevated default memory support up to DDR4-3200, 20 PCIe 4.0 lanes from the CPU, doubled bandwidth of the DMI link, and moving the integrated graphics to the new Xe graphics architecture.
The overclockable K-SKU processors again come in three flavors: Core i9, Core i7, and Core i5. Each of the three flavors has a -K a -KF variant. The only difference between the two variants is that the -KF comes without integrated graphics.
The frequencies of the flagship Core i9-11900K processor are nearly identical to the Core i9-10900K. So, we’ll see a base frequency of 3.5GHz, an all-core turbo of 4.8GHz, and a maximum thermal velocity boost frequency of 5.3GHz. The TDP is 125W and the CPU can boost up to 250W temporarily.
The frequencies of the Core i7-11700K processor are slightly lower than the Core i7-10700K. The base frequency is 3.6GHz, an all-core turbo of 4.6GHz, and a maximum turbo boost frequency of 5.0GHz. The TDP is 125W and the CPU can boost up to 250W temporarily.
The frequencies of the Core i5-11600K processor are also slightly lower than the Core i5-10600K. The base frequency is 3.9GHz, an all-core turbo of 4.6GHz, and a maximum turbo boost frequency of 4.9GHz. The TDP is 125W and the CPU can boost up to 180W temporarily.
Rocket Lake will work on both 500 series and 400 series motherboards, though not on B460 or H410. Obviously that means Rocket Lake fits in the LGA1200 socket.
Intel Rocket Lake Overclocking Features
Intel Rocket Lake counts 14 new or improved overclocking features. Some of these features are directly related to overclocking, others are related to performance tuning in general. We’ll go over each of the features and hopefully help you understand what it does.
Per Core Ratio Limit
A big change from previous generations of Intel CPUs on mainstream desktop is that each of the processor cores now has its own PLL. PLL stands for phase-locked loop or phase lock loop. Simply put, the PLL generates an output signal such as clock frequency. Each core having its own PLL means the processor cores can operate at independent frequencies. Prior to Rocket Lake, on mainstream desktop, all CPU cores of the CPU would run the same CPU ratio.
The per core ratio limit allows the bios to configure each physical core to its own maximum boost ratio. This enables two important new avenues for CPU overclocking.
- First, it allows users to individually overclock each of the cores and find the maximum stable frequency.
- Second, it allows users to set an aggressive by core usage overclock while constraining the worst cores. This is particularly interesting considering Rocket Lake CPUs have 2 ‘favored’ cores.
Favored cores is part of the Turbo Boost Max 3.0 technology. The favored cores are the cores that Intel deems to be the highest quality among all cores present on your CPU. Not only will these cores benefit from the highest turbo boost ratios, they will also be prioritized by the operating system when assigning workloads.
With per core ratio limit you can now prioritize one favored core to potentially run one ratio higher than the other core.
Note that while each of the CPU cores now has its own PLL (and its own V/F curve), still only one voltage is applied across all cores. The voltage applied is the highest of all active cores.
In the ASUS ROG BIOS you can access the Per Core Ratio Limit options via the Extreme Tweaker Specific Core sub-menu. In this sub-menu you can configure the maximum turbo ratio for each of the cores on your CPU. Note that the favored cores are marked with an Asterix.
In my example, I will configure a ratio limit of 53X for Core 0, then one ratio less for each successive core, all the way to 46X for Core 7. After saving the BIOS settings, I can verify these settings in the operating system using HWinfo
AVX512 Ratio Negative Offset
With the inclusion of support for AVX512 instructions, Intel has also enabled the option to reduce the CPU ratio when executing very demanding AVX512 instructions. This works exactly like the AVX offset we know from previous platforms. It is just enabled when AVX512 instructions are detected.
Both AVX and AVX512 negative ratio offsets are very useful to achieve the maximum performance for both SSE and AVX workloads. Generally speaking an offset of 2 or 3 is recommended but it is highly dependent on not only your cooling solution but also the motherboard you’re using. That is because AVX workloads are very demanding and therefore require great cooling and power delivery.
There is a single AVX512 ratio offset that is used for all cores.
In the ASUS ROG BIOS you can access this option in the Extreme Tweaker AVX Related Controls sub-menu.
AVX Per Core Offset
As mentioned in the previous segment, the AVX512 negative ratio offset is a single setting that is used for all cores. However, since the Rocket Lake CPU now supports independent ratios for each CPU core, the AVX offset is referenced against each individual core.
This is very different from previous Intel CPUs. Previously, the AVX offset would be referenced against the all-core maximum ratio. So when AVX instructions are detected, the frequency of all cores would be the all-core maximum ratio minus the AVX offset.
On Rocket Lake, the AVX offset will be applied to each core separately. That means you have to be a little careful when your overclock has different ratios for different cores.
In our example, we set the AVX2 and AVX512 offset to 5. Then we also set the Per Core Ratio Limit of Core 0 to 50, then one ratio lower for each subsequent core, ending with 43X for Core7. In the operating system we will see that the frequency under non-AVX workloads is 50X for Core0 to 43X for Core7. When we run an AVX workload, the frequency will be offset per core. The resulting frequency is 45X for Core0 to 38X for Core7.
As mentioned, you can access the AVX offset settings in the ASUS ROG BIOS by going into the Extreme Tweaker AVX Related Controls sub-menu.
AVX2 & AVX512 Disable
In addition to the ability to reduce the CPU frequency with the AVX ratios, an additional feature related to AVX is the ability to disable AVX2 and AVX512 instructions. Disabling AVX causes software to take non-AVX execution paths resulting in lower performance.
So, it doesn’t mean AVX-enabled software won’t run. It just means lower performance.
To show you the difference in performance, I ran Cinebench R20 with AVX enabled and disabled. With AVX enabled our score is over 6000 points. Also notice how CPU-Z mentions support for AVX, AVX2, and AVX512F instructions. When we disable AVX our score drops almost 700 points. Also notice how CPU-Z no longer mentions support for AVX, AVX2, and AVX512F instructions.
In the ASUS ROG BIOS you can access this option in the Extreme Tweaker AVX Related Controls sub-menu.
AVX Voltage Guard-band Override
A voltage guard-banding ensures that the effective voltage stays within the required range at all times to ensure stability. In Rocket Lake, users are given an option to change the voltage guard-band when running AVX2 and AVX512 instructions using the Voltage Guard-band Scale Factor.
The scale factor is a number between 0 and 2.0, where 0 means no scale factor is applied and 1 means the default scale factor is applied.
The resulting final guardband is as follows:
- AVX2 Final Voltage Guardband = AVX2 default guardband * AVX2 Scale Factor
- AVX512 Final Voltage Guardband = (AVX2 default guardband + AVX512 default guardband) * AVX512 Scale Factor.
Note that the value of the guard-band is arbitrary and may vary from CPU to CPU. Also, it may be a very small value so adding any scale factor would not show any or limited results.
In the ASUS ROG BIOS you can access this option in the Extreme Tweaker AVX Related Controls sub-menu.
Memory Gear Mode
One of the major improvements on Rocket Lake is the memory controller. The memory controller now supports Gear1 and Gear2 mode. Gear1 mode is the regular mode as we know it from previous platforms. Gear2 mode is the Gear-down mode.
To better understand what it does, we need to touch a little on how memory works.
On DDR4 memory, there are two bus. There’s one command/address bus and one data bus. As the name already tells, when the CPU needs to read or write data to the memory, it uses a command and address to specify what needs to be stored and where. The actual data is sent over the data bus.
The data bus is DDR, or double data rate, meaning it can send 2 signals per clock cycle. It does this on both the rising and the falling edge of the memory clock cycle.
Command and Address bus are SDR, meaning it can send 1 signal per clock cycle. It does this on the rising edge of the memory clock cycle.
As the command/address bus is the bridge between the memory controller and the system memory, both are typically synchronized.
Gear-down mode is part of the DDR4 specification and allows the address/command bus to use every other rising clock. So, practically, reduce the speed by half. This can provide some additional headroom for overclocking at the cost of a performance penalty.
In gear-down mode, the data bus is still using both rising and falling edge of the clock cycle. So, effectively gear-down mode is operating at 2T. If you would combine the gear-down and 2T, then effectively you’d be running at 4T. Performance-wise this is not recommended.
Unlikely what most enthusiast expect, it is not gear-down mode itself that is the key to increased memory overclocking capabilities.
As we mentioned, gear-down mode effectively halves the amount of commands issued by the memory controller. What means the memory controller is now running at twice the frequency compared to the DRAM address/command bus. While this is perfectly fine, rather than running the memory controller at twice the speed of the DDR address/command bus, on Rocket Lake gear-down mode also halves the memory controller frequency.
This is key to unlocking higher memory frequency headroom.
In fact, you’ll find that where on previous platforms you had to set very high Vccio and Vsa voltages for high-speed memory, this is no longer the case on Rocket Lake. In other words, Gear2 enables both higher memory frequency and does so at lower voltages. A definite win-win.
Gear1 mode is supported up to DDR4-3200 officially. From testing it looks like Gear1 may work up to DDR4-3600. For higher memory frequencies you’ll need to enable Gear2 mode. However, it’s very likely that your motherboard will automatically adjust this when selecting the higher memory ratios.
In the table below you can compare Gear1 and Gear2 mode in more detail.
In the ASUS BIOS you can change the Gear Ratio mode from Gear1 to Gear2 in the Extreme Tweaker menu under the item Memory Controller : DRAM Frequency Ratio. 1:1 is Gear1, 1:2 is Gear2.
In our example we will show the difference in frequency at DDR4-3200.
When enabling Gear1 mode, the DDR4-3200 rated memory operates both the data as well as the address/command bus at 1.6 GHz. As the Intel memory controller should be equal to or higher than the address/command bus, it is also running at 1.6 GHz.
When enabling Gear2 mode, the DDR4-3200 rated memory operates only the data bus at 1.6 GHz. However, the DRAM address/command frequency is halved and runs at 800 MHz. On Rocket Lake, by default Intel also halves the memory controller frequency to run in sync with the address/command bus at 800MHz.
DRAM Odd Ratio Mode Removed
As a consequence of enabling gear down mode, Intel has repurposed the registers which were used to enable “Odd Ratio mode” support for Comet Lake processors.
When Odd ratio Mode is enabled the memory frequency is increased by an additional 100 or 133 MHz over the resulting frequency, depending on the choice of base frequency.
However, the reference clock for memory can still be either 100 MHz or 133 MHz. That means the maximum memory frequency on Rocket Lake is DDR4-6300 with 100MHz reference clock, and DDR4-8400 with 133MHz reference clock.
Of course, you can still further increase the frequency using BCLK overclocking.
In the ASUS BIOS you can change the DRAM ratio in the Extreme Tweaker menu. The setting BCLK Frequency : DRAM Frequency Ratio lets you change between 100 MHz or 133 MHz base frequency. The option DRAM Frequency lets you pick a specific DRAM ratio.
Ring down-bin More Aggressive
While not actually an overclocking feature, Ring down-bin is an important feature to keep in mind when trying to achieve high performance.
The CPU Ring is the bus that connects all different parts of the Intel CPU to transfer data between different cores, between cores and memory, and between cores and other parts of the system.
A standard feature of the Rocket Lake CPU and previous generations is that the processor has a power management feature that reduces the Ring ratio by several steps compared to the Core ratio. In a typical situation you would see the ring down-bin of 3, meaning that if the CPU cores are running at 4.5GHz the system would configure the Ring ratio to 4.2 GHz.
The Ring has its own voltage frequency curve, meaning it will request a certain voltage when running a certain frequency. However, the CPU cores and Ring share the same voltage plane. So, if you force a specific Ring frequency which requires a higher voltage, is it very possible that would result in effectively over-volting the CPU cores. That in turn would increase the temperatures and power consumption.
On Rocket Lake, it looks like the voltage requirements for the Ring are significantly higher than past generations of CPUs. That means 3 things
- At default, the CPU will ring down-bin more aggressively to ensure stability with a given core voltage.
- When overclocking, your maximum ring frequency will likely be constraint by the CPU core voltage you select
- When performance tuning, keep the Ring frequency for last as it doesn’t have a significant impact on performance but it does have an impact on temperature and power
In the ASUS BIOS you can change the Ring Down Bin settings in the Extreme Tweaker menu. By default ring down bin is enabled. If you want to force a specific ratio, you must disable the option and set a minimum CPU Cache Ratio.
To show you the behavior of aggressive Ring Down Bin, we ensure the system is running without standard turbo boost parameters. Then we enable the ring down bin option in the BIOS.
We can now check the ring frequency in the operating system under a heavy workload.
At boot, the Ring frequency is 4.2 GHz. Immediately after starting the Prime95 workload, we see the ring frequency drop to 3.9GHz. After a while, the ring frequency will further drop to 3.1 GHz and even 3 GHz. This is about 700MHz lower than the CPU frequency.
PVD Ratio Threshold
Every PLL design has a minimum and maximum oscillation frequency (Digitally Controlled Oscillator (DCO) frequency). For an all-digital PLL the DCO operation range is between 1600MHz-~6000MHz. In order to generate lower frequencies that are within the minimum range, the PLL‘s DCO multiplies lower ratios by x2, x4 or x8 and that frequency is divided using a post divider.
This logic assumes a reference clock of 100MHz and thus the threshold to switch to a x2 post divider is 15 by default. This is called a PVD threshold. The thresholds for x4 and x8 are derived by the PLL from the x2 threshold. Another threshold is used for PLLs that work on a 133MHz ref clock (MCPLL).
When the BCLK is increased for overclocking, the logic breaks and may cause PLL banding issues.
While this is unlikely to be an issue for daily overclocking, when running BCLK in excess of 200 MHz BCLK you can set a value lower than 15 to ensure the DCO is within stable range.
In the ASUS ROG BIOS you can access the PVD Ratio Threshold in the Extreme Tweaker Tweaker’s Paradise sub-menu. All the way at the bottom there’s an option called PDV Ratio Threshold.
2x VCCIO voltage
On previous platforms the VCCIO voltage was an important vector for high memory frequencies. On Rocket Lake, Intel has split the VCCIO voltage into 2 voltage: 1 for the memory controller and 1 for the rest of the IO.
When memory overclocking, the VCCIO for the memory controller is the one that needs to increase. A voltage between 1.25v and 1.45v should be sufficient for memory speeds up to DDR4-6000 and higher.
CPU VCCIO voltage at default of 1.05v powers everything else.
In the ASUS ROG BIOS you can access the VCCIO voltages in the Extreme Tweaker menu. In the section with the voltages you’ll find CPU VCCIO Voltage and VCCIO Mem OC Voltage. It is recommended to set the former option to 1.05 and only increase the VCCIO Mem OC Voltage when overclocking the system memory.
PCH FIVR VCCIN 1.8V
Another note is that the PCH is powered by a VCCIN of 1.8v which is used as it’s FIVR input voltage.
FIVR stands for Fully Integrated Voltage Regulator. FIVR integrates the voltage regulation in the IC to simplify the overall platform power delivery design complexity. Enthusiasts will remember FIVR from the 4th generation Intel Core processors codenamed Haswell. On Haswell, FIVR consolidated 5 different voltage planes (Vcore, Vgpu, Vccsa, Vccio, and PLL) into a single voltage input (Vin).
In the ASUS ROG BIOS you can access the PCH VCCIN voltage in the Extreme Tweaker menu. In the section with the voltages you’ll find the PCH VCCIN 1.8V option. Changing this voltage is unlikely to give you any benefit when overclocking.
Delidding Rocket Lake CPU
While delidding the CPU is not impossible, it is quite difficult due to the solder used and the positioning of the resistors on the CPU PCB.
I was sent a picture of a delidded Rocket Lake CPU by a good friend. If you were thinking to pop the hood off your CPU just to see what’s underneath the heatspreader, I saved you the trouble.
Realtime Memory OC
Real-Time Memory OC enables the ability to switch between non-OC and OC frequency (XMP1) at runtime in the operating system. This features uses the SAGV technology. SAGV stands for System Agent GeyserVille. GeyserVille is the original codename for the power-saving technology that would eventually be launched as SpeedStep on the Intel Mobile Pentium 3 CPUs.
So, it is not a surprise that SAGV dynamically tunes the system agent voltage and clock frequencies depending upon power and performance needs. SAGV itself has been around for several generations of Intel Core processors but is now used to enable real-time memory OC.
At the moment of recording, it’s not entirely clear what’s the benefit of using this feature on a desktop system.
In the ASUS ROG BIOS you can enable real-time memory oc in the Extreme Tweaker menu. First, make sure to enable XMP. Then access the DRAM Timing Control sub-menu and enable the option called Realtime Memory Overclock. Then save and exit the BIOS.
During the reboot, the CPU will train both the standard JEDEC specification and the XMP specification. As it’s double the work this can take up to 2 minutes. So you’ll need some patience.
OC Thermal Velocity Boost OC
After Turbo Boost 2.0 and Turbo Boost Max 3.0, Thermal Velocity Boost is a third extension of the Intel’s pursuit of maximizing system performance via turbo boost. Introduced in 2018 along with the Core i9-8950HK Coffee Lake mobile processor, Thermal Velocity Boost is available on select Core i9 desktop and mobile CPUs. It enables one additional ratio over the Turbo Boost Max 3.0 ratio when the CPU is running below a certain temperature threshold. On Rocket Lake desktop CPUs, that temperature threshold is 70 degrees centigrade.
OCTVB or OC Thermal Velocity Boost was introduced along with the Intel Cryo Technology not so long ago. OCTVB simply allows the user to manually configure the Thermal Velocity Boost triggers and ratios
We extensively used OCTVB in SkatterBencher #19 and explained the configuration in a video titled “Intel Thermal Velocity Boost Explored: TVB OC”. If you want to understand how this feature works, check out that video
While OCTVB is included in all Rocket Lake documentation and should be supported, at the moment of recording the feature has not been enabled yet. That means we’ll likely come back to this topic in a future article.
When the feature is eventually supported you’ll be able to access the settings in the ASUS ROG BIOS. Specifically, in the Extreme Tweaker Thermal Velocity Boost sub-menu.
Legacy Overclocking Features
Of course Rocket Lake also supports the legacy overclocking features listed below:
- Turbo Boost Technology 2.0
- BCLK overclocking
- Turbo Boost Max Technology 3.0
- Thermal Velocity Boost
- TjMax Offset
- Per Core HT Disable
- Advanced Voltage Offset (VF point offset)
- Extreme Tuning Utility
- Extreme Memory Profile
ASUS ROG Maximus XIII APEX Overclocking Features
The ASUS ROG Maximus XIII Apex counts about 4 new or improved overclocking features. Some of these features are directly related to overclocking, others are related to performance tuning. We’ll go over each of the features and hopefully help you understand what it does.
VLatch is a brand new feature on the Maximus XIII Apex and Extreme that allows enthusiasts to detect the true Vmin and Vmax of the CPU core voltage without having to purchase an expensive oscilloscope.
Explaining the VLatch function and purpose gets complicated very quickly but I’ll try to keep it as simple as possible.
In one of my YouTube videos I talked about Intel Adaptive Voltage mode. In that video I showed this drawing to explain how the CPU and motherboard VRM interact to provide a certain voltage. There are 3 basic steps:
- The motherboard communicates about its board design to the CPU via the ACDC Loadline
- The CPU requests a certain voltage by using a VID which is a combination of the factory-defined V/f curve and the ACDC loadline
- The voltage provided by the motherboard VRM is the requested voltage minus adjusted by the VRM loadline.
It’s the VRM loadline that many enthusiasts wish to have control over to when trying to find the maximum stable settings. There are two parts of the loadline that are important to system stability.
First, the Vdroop. This is the decrease in voltage when the CPU goes from idle to load. Obviously, you want your CPU to be stable in all scenarios so knowing what’s the lowest voltage the CPU will run at is very important. After all, the if the voltage is too low then the overclock will be unstable.
Second, the Undershoot and Overshoot. These are very brief spikes that occur when the CPU switches from idle to load, or from load to idle. These spikes cannot be measured easily and usually require an expensive oscilloscope to detect. I can highly recommend ElmorLabs article titled VRM Load-Line Visualized if you want to see a great picture of undershoot and overshoot in action.
While undershoot and overshoot are very temporary spikes, an undershoot that’s too low will also cause instability.
With VLatch you can now detect the undershoot and overshoot. On the Maximus XIII Extreme and Maximus XIII Apex there’s a proprietary circuit design that allows the embedded controller to capture and report the true Vmin and Vmax. You can use either HWinfo or the OLED display to see these voltages.
If you want to use VLatch to identify the best load-line setting for your system, I suggest using the following procedure.
First, we must configure the system such that the frequency behavior is dynamic yet controllable. To do this, we
- Unlock the turbo boost power limits, as this will force the CPU to run between 4.8 and 5.3 GHz
- Disable AVX2 and AVX512, as this will prevent the high temperature from causing frequency throttling
- Disable Ring Down Bin, as this will ensure a consistent CPU Ring frequency
- Set Max Ring to 40X to force the ring frequency at 4 GHz.
- Use CPU adaptive voltage mode
- Change the VRM load Line calibration setting in the Digi+ VRM sub-menu
In the operating system, we do the following:
- Open HWinfo to log the information.
- Start SuperPI 32M to run in the background. This will ensure that there’s always minimum 1 core active and prevent the VLatch from reporting the minimum voltage when all cores are idle (which would be around 0.6V)
- Run 1 loop of ROG Realbench which will take about 9 minutes
When done, export the CSV data into excel and highlight the following information:
- VLatch: Min and Max
- Vcore: Min and Max
Then calculate the Undershoot as the difference between VLatch Min and Vcore Min, and calculate the Overshoot as the difference between Vcore Max and VLatch Max. To find the optimal loadline setting, compare the sum of undershoot and overshoot across all loadline settings.
Then, after about two hours of testing you can pour yourself a glass of whisky when you realize that it’s the same setting as the ASUS ROG BIOS had already recommended to you.
As reported by several media leading up to the ROG Maximus XIII series availability, MemTest86 is now integrated in the BIOS. This allows users to test memory overclocking without risking damaging their operating system or corrupting system files.
In the ASUS ROG BIOS you can access the integrated MemTest86 at the very bottom of the Tool menu. Click Activate MemTest86 to start the test. After the test finished, it will also store a test report. You can access those in the same menu under MemTest86 Reports.
Since Rocket Lake has a significantly increased memory frequency overclocking capability, I’m sure you’ll see people running MemTest at DDR4-5866 or even higher than DDR4-6000 using ambient cooling.
I was fortunate enough to be able to try a prototype G.SKILL kit rated at XMP-5866. Here’s the kit running MemTest86 at XMP speeds with no manual adjustments to either DRAM timings or voltages. The motherboard auto-rules set the DRAM voltage to 1.65V according to the XMP, sets the VCCIO voltage to 1.448v, and the System Agent voltage to 1.472v. It shows very well how amazing the memory overclocking capabilities of Rocket Lake is.
MCE & Package Temperature Threshold
ASUS MultiCore Enhancement is a single BIOS option that removes all limits constraining the Turbo Boost 2.0 algorithm. Effectively, it allows the CPU to run at maximum turbo boost frequencies indefinitely. On Z590 motherboards, ASUS has enabled this option by default.
For years, hardware reviewers and consumers alike have been very outspoken against motherboard vendors unlocking the Turbo Boost power limits by default. Their key argument is that the CPU is not running at Intel specification. While I understand this argument, I disagree with it.
The Turbo Boost 2.0 algorithm consists of three key parameters. PL1, PL2, and tau. PL1 determines the long-term power consumption, PL2 determines the peak power consumption, and tau is a parameter used for calculating how long the Turbo Boost can be active.
These parameters are chosen to work together with Intel’s minimum specification guidelines for motherboard vendors and third-party cooling solution providers. Effectively, Intel promises the customer a guaranteed minimum performance level even if the customer pairs the CPU with a motherboard that has a minimum specification power delivery design and a cooler that has a minimum specification cooling capacity.
However, we all know that there are plenty of high-quality motherboards with powerful power delivery designs and solutions like custom loop water cooling that can dissipate well over 300W of thermal heat.
In this scenario, the customer has purchased the equipment to run the Intel CPU at maximum specification. The maximum specification is determined by Intel’s factory set voltage frequency curve, the maximum Turbo Ratio Limits, and TjMax.
In case of the Core i9-11900K, a maximum specification Core i9-11900K will run stably at*:
- 5.3 GHz, if only the 2 favored cores are active and the package temperature is below the thermal velocity boost threshold of 70 degrees and below TjMax of 100 degrees
- 5.2 GHz, if only the 2 favored cores are active and the package temperature is above the thermal velocity boost threshold of 70 degrees and below TjMax of 100 degrees
- 5.1 GHz, if 2 cores non-favored cores are active and CPU package temperature is below TjMax of 100 degrees
- 4.9 GHz, if all cores are active and the package temperature is below the thermal velocity boost threshold of 70 degrees and below TjMax of 100 degrees
- 4.8 GHz, if all cores are active and the package temperature is above the thermal velocity boost threshold of 70 degrees and below TjMax of 100 degrees
(* before Adaptive Boost Technology)
Therefore, as a performance enthusiast I fully approve of motherboard vendors enabling unlocked power limits by default. Especially on motherboards aimed at customers who will likely purchase high-end cooling solutions. For example, customers who purchase motherboards with the Z-series chipset.
Whether I need 125W or 350W to achieve the maximum performance level is beside the point. It is up to the customer to choose the motherboard and cooling solution that fits their use-case.
That said, however, I do think it would be smart for motherboard vendors to implement solutions that can ensures system stability in the widest range of scenarios. That’s why I also very much approve of ASUS’ decision to by default enable the Package Temperature Threshold feature.
We previously used this feature in SkatterBencher #19. To better understand how this feature helps regulate the power consumption to target a specific CPU temperature, you can check out the video titled “ASUS AI Cooling Explored: Package Temperature Threshold”.
The long story short is that the package temperature threshold will reduce the CPU power consumption if the temperature exceeds a manually defined threshold. The default configuration for the Maximus XIII series motherboards is 90 degrees centigrade.
In the ASUS ROG BIOS you can enable or disable the ASUS MultiCore Enhancement in the Extreme Tweaker menu. You can configure the package temperature threshold in the Extreme Tweaker AI Features submenu.
To demonstrate the behavior at default settings, we leave all BIOS settings as default. Then we go in the operating system and run a heavy workload like Prime95 Small FFTs with AVX. We use HWinfo to track the operating temperature and CPU-Z to check the CPU frequency.
We can see that the at first the CPU is running all cores at 4.9 GHz. That’s because all cores are active and the CPU temperature is below the 70 degree thermal velocity boost threshold.
However, almost immediately the temperature increases to over 80 degrees and the frequency is reduced to 4.8 GHz.
As the CPU Package power hits over 270W, the temperature starts increasing slowly. For the purpose of demonstration, we disconnect the radiator fans. This will further increase the operating temperature.
Not even 2 minutes later the CPU hits 90 degrees centigrade. This is the package temperature threshold as configured by ASUS.
Moments after the CPU temperature hits 91 degrees centigrade, we see the CPU ratio drop one or two ratios.
ASUS’ implementation will very gradually decrease the CPU ratio to target a maximum CPU temperature of 90 degrees centigrade.
Almost 3 minutes later and still running without any fans spinning, the CPU frequency barely dropped to 4.5 GHz.
Fast-forward another 6 minutes and the system is still running a respectable 4.2 GHz at 90 degrees centigrade.
Overall, I think ASUS has come up with a brilliant solution that truly maximizes the performance level for enthusiast K-SKU users.
Extended voltage cap
The last item on the list is the extended voltage cap options which you can find in the Extreme Tweaker Auto Voltage Caps submenu. There are options to cap the CPU Core, CPU VCCIO, VCCIO Mem OC, and CPU System Agent voltage.
Not much more to add here.
Intel Rocket Lake Overclocking Expectations
In terms of the overclocking expectations we can split it up in two parts: ambient cooling and extreme cooling.
|1 core max stable||8 core max stable||8 core max w/ AVX stable||Ring max stable||Memory max stable|
For ambient overclocking, I expect the following maximum overclocking results when using high-end custom loop water cooling.
- 5.5 GHz for 1-core maximum stable frequency
- 5.1 GHz for all-core maximum stable frequency
- 4.8 GHz for all-core AVX maximum stable frequency
- 4.3 GHz for maximum stable ring frequency
- DDR4-6133 for maximum stable memory frequency
This is almost the overclock that I managed during one of my early recorded test sessions.
So, compared to Comet Lake:
- Similar single core frequency range
- 100MHz less for all core frequency
- 200 MHz less for all core with AVX frequency
- 300-400 MHz less for the ring frequency
- Vastly improved memory frequency capabilities
These overclocking margins are also in line what we can see for extreme overclocking.
To end this article with I’d like to share with you the Rocket Lake CPU Frequency record set by the ASUS ROG team last month. As is tradition, they used liquid helium to squeeze the most out of a new CPU architecture on launch day. The team consisted of Elmor from ElmorLabs, Shamino from ASUS ROG, and SafeDisk also from ASUS ROG.
While Elmor focused on obtaining the CPU Frequency validation, SafeDisk ran a series of benchmarks with liquid helium. In the end, they pushed the Core i9-11900K past 7.3 GHz with all eight cores enabled.
I won’t describe all the footage but I’ll try to annotate the video to give you a little more insight in how a liquid helium session works. I hope you enjoy the bonus footage. As per usual, if you have any questions or comments, feel free to drop them in the comment section below.