2 Phases 1 Raptor Lake
I’m demonstrating all you need to run a 13700K Raptor Lake CPU to its limit are two water-cooled VRM phases.
You can see the Z690 Torpedo EK X motherboard with all but two CPU Vcore phases disabled. We’re running Prime95 on a Core i7-13700K with all P-cores and E-cores enabled and Turbo Boost 2.0 power limits unlocked. Admittedly, the VRM runs a bit hot at 103 degrees Celsius.
In this short blog post, I’ll explain how I set up this test and why it works the way it does.
Table of Contents
Honestly, this video is long overdue. I’ve wanted to tackle this subject since I published the 13700K SkatterBencher guide in November last year.
I used the Z690 Torpedo EK X motherboard in that guide, which features an EK Liteblock. If you missed that guide, a liteblock is like a monoblock because it cools both the CPU and the VRM. However, unlike the traditional monoblocks, the VRM isn’t actively cooled by the liquid. Instead, the water block connects to the stock thermal solution with a thermal pad. That connection ensures sufficient heat exchange to vastly improve the VRM temperatures.
As I showed in SkatterBencher #34 with the 12900KF and SkatterBencher #50 with the 13700K, the theoretical and practical VRM thermal performance improvement over a standard passive heatsink was clear from the internal test data and the data from my overclocking guides. While the Liteblock doesn’t perform as well as a typical monoblock, the thermal improvement is still significant.
One topic I didn’t cover in these guides is that the improved thermal performance also opens the door for VRM cost optimization.
VRM & Power Phases
VRM is short for voltage regulator module. It typically includes several power MOSFETs managed by a dedicated controller. The VRM ensures the right power is provided from the power supply to the CPU or GPU.
The power delivery design of motherboards and graphics cards is vital for stability, reliability, performance, efficiency, and safety reasons.
- Stability: The VRM is responsible for providing stable and consistent power to the CPU and other components on the motherboard. If the VRM is poorly designed, it can result in voltage fluctuations, which can cause instability, crashes, and other performance issues.
- Reliability: A well-designed VRM can improve the reliability and longevity of the CPU and other components. Poor voltage regulation can cause components to wear out more quickly, leading to premature failure.
- Performance: The VRM can also significantly impact the performance of the CPU and other components. A high-quality VRM can provide cleaner and more stable power, improving performance and overclocking potential.
- Power efficiency: A good VRM can also improve power efficiency, reducing the input and output power gap and maintaining lower operating temperatures. That can help to prolong battery life and improve overall system cooling.
- Safety: Properly designed VRMs can also incorporate safety features to protect the CPU and other components from overvoltage, overcurrent, and other abnormal conditions, helping to prevent damage to the system and potential safety hazards.
Engineers must carefully consider several elements during the motherboard design phase to design the right VRM. Those elements include power requirements, thermal management, electrical component selection, protection features, PCB layout, and EMI mitigation.
- Power requirements: The VRM must be designed to meet the power requirements of the components it supplies power to. This includes determining the number and type of power phases, the maximum current and voltage levels required, and the overall power efficiency of the design. For enthusiast motherboards, it’s essential to design the VRM so that it can support overclocking too.
- Thermal management: The VRM generates heat which must be effectively dissipated to prevent overheating and ensure the longevity of the components. So, the engineers must select appropriate cooling solutions, such as heat sinks and fans, and optimize the heatsink thermal design and surrounding components.
- Component selection: The components used in the VRM must be carefully selected for their quality, reliability, and performance characteristics. This includes choosing the appropriate MOSFETs, inductors, capacitors, and other components based on their specifications, tolerances, and cost.
- Protection features: The VRM should include protection features to prevent damage to the components in case of overvoltage, overcurrent, or other abnormal conditions. This includes incorporating features such as overcurrent protection, overvoltage protection, and thermal shutdown.
- PCB layout: The layout of the VRM on the PCB is critical to its performance and reliability. It should be optimized for the specific components used. It should minimize the impedance between the power delivery components and the CPU/GPU.
- Electromagnetic interference (EMI) mitigation: The VRM can generate EMI that can interfere with other components on the motherboard or nearby devices. The design should include appropriate EMI mitigation techniques such as filtering and shielding.
Finally, and perhaps most importantly, the VRM design must be cost-effective too. The VRM components are usually the most significant chunk of a motherboard’s bill of materials (aside from the chipset). Any dollar spent too much could mean being less price-competitive in the market.
In previous SkatterBencher guides, I’ve discussed how thermal limitations can sometimes be a bottleneck for overclocking. For example, in SkatterBencher #40, where I overclock the NVIDIA GeForce GT 1030, I note that “the GPU VRM struggles to operate sufficiently cool with the increased voltage. It’s so bad that at 1.4V, we measure temperatures of 150C and can smell a slight burning going on [and] an overheating VRM is critically limiting our voltage headroom.”
Z690 Torpedo EK X CPU VRM Design
The CPU VRM design of the Z690 Torpedo EK X consists of the following major components:
- An MPS MP2960 digital multi-phase controller, driving both the VccCore and VccGT voltages
- Sixteen (16) MPS MP87992 70A Intelli-Phase DrMOS.
DrMOS is a power delivery solution that integrates the MOSFET driver directly on-chip with the power FETs. Funny tidbit: Intel wrote the DrMOS 1.0 specification back in 2004.
Each CPU power phase consists of two DrMOS components. The MP87992 is rated to operate continuously up to 70 amps and 125 degrees Celsius. The configuration for this motherboard has overcurrent protection set at 52A per phase, so 416A in total when all 8 phases are enabled. If we’d use a CPU voltage of 1.2V, that would translate into about 500W output power.
13700K CPU VRM Phase Testing: Setup
Anyway, back to the main topic of this video.
I wanted to check how much the EK Liteblock and water cooling could impact the VRM operating temperature when pushed to its limits. To test this, I use the following hardware.
|CPU||Intel Core i7-13700K||410|
|Motherboard||EK-Mana MSI MAG Z690 Torpedo EK X||280|
|CPU Cooling||EK-Pro QDC Kit P360||840|
|Fan Controller||ElmorLabs EFC-SB |
|Power Supply||Antec HCP 1000W Platinum|
|Storage||Kingston SSDNow V300 120GB||30|
|Chassis||Open Benchtable V2||199|
A quick note on the thermal solution.
As usual, I use the EFC-SB to map the radiator fan curve to the water temperature. Without going into too many details: I have attached an external temperature sensor from the water in the loop to the EFC-SB. Then, I use the low/high setting to map the fan curve from 25 to 40 degrees water temperature. The main takeaway from this configuration is that it gives us a good indicator of whether the cooling solution is saturated.
To test the VRM capability, I run Prime95 Small FFTs without AVX on the Core i7-13700K with unlocked Turbo Boost 2.0 power limits. All P-cores and E-cores are enabled and run at 5.3 and 4.2 GHz, respectively. The average CPU Package Power throughout the test is about 260W.
I change the number of active phases in the motherboard BIOS by using a test BIOS provided by MSI.
I also hooked up the EVC2 device to the motherboard, allowing me to talk to the MP2960 CPU VccCore voltage regulator. This was necessary for two reasons:
- To set the number of active phases to 3 and 2. While the setting is available in the BIOS, the board didn’t always boot up for some reason. So, alternatively, I could adjust it at run time in the operating system using the EVC2 software.
- To increase the OCP level at runtime. The default configuration sets the OCP at 52A multiplied by the number of active phases. The Prime95 workload uses about 180A, so when we’re down to 3 phases, it will exceed the configured OCP of 156A. Fortunately, we can adjust the OCP level to 127A per phase.
Then I monitor the input power using the ElmorLabs PMD and a range of temperatures and voltages using HWiNFO. What I’m looking for is three things:
- How many phases can I disable before the system gets unstable?
- What is the VRM temperature?
- What is the power efficiency (input power vs. CPU package power)?
Let’s have a look at the results.
13700K CPU VRM Phase Testing: Results
From this chart, we can derive the answers to our questions.
First, I can reduce the number of CPU phases from 8 to 2, and the system remains stable.
Second, when we change the phase count from eight to two, the VRM temperature doubles from 50 to 100 degrees Celsius. While that may seem exceptionally high, remember that the VRM components are rated to operate safely well above 100 degrees Celsius. So, thermally, this configuration is perfectly acceptable.
Third, the power efficiency drops from about 85% to 77%. While that may sound like a lot, at a CPU package power of 300W, that’s a difference of 37W input power to the power supply. So, it’s not like it will drastically increase your power bill.
Pushing the Limit of 2 CPU VRM Phases
This wouldn’t be a good video if we didn’t try to push our 2-phase configuration to its limits. I also tried to run Prime95 with AVX and AVX2 enabled to significantly increase the current draw. In addition to the settings I changed previously, I also
- Increased the CPU TjMax to 115 degrees Celsius,
- maxed out the per-phase OCP to 127A,
- maxed out the allowed VRM temperature to 150 degrees Celsius.
Here are the main differences in power consumption between the three 2-phase configurations.
By enabling AVX2, the CPU package power increases about 28%. That translates into a 25% increase in per-phase current draw and a 37% increase in total CPU power consumption. The power efficiency decreases from 77.1% to 72.4%, and the VRM temperature now exceeds 110 degrees Celsius.
If we consider the peak measures, it becomes even crazier. The most we can squeeze out of this 2-phase system is 505W CPU input power for a 351W CPU Package Power and 121 amps per phase. Admittedly, this is not a sustainable configuration, given both the CPU and the VRM are triggering the thermal protection mechanisms. Also, our EFC-SB-regulated thermal solution is entirely at its limits. It’s pretty wild the motherboard doesn’t outright explode on us.
So, what is a sustainable system, then? Let’s try the same settings with 3 phases enabled and run Prime95 Small FFTs with AVX2 again.
The CPU Package Power is still 340W on average, with the CPU Input Power peaking at 484W. However, the per-phase current now only peaks at 81A, and the maximum MOS temperature is 105 degrees Celsius. The efficiency is only 77.6%, down from our original 85%, but not too bad, considering we’re relying on less than half the phase count.
Alright, let’s wrap this up.
First, I’d like to thank my friends at MSI and ElmorLabs, who helped me set up this demonstration. While I know at a basic level how to use the EVC2 device to communicate with digital VRM controllers, I still needed much help to do it properly.
Second, despite its relatively basic design, I am genuinely surprised by how well the EK Liteblock design cools the VRM. I hoped to run this Z690 Torpedo EK X at half the phase count, but getting Raptor Lake to run full blast with only 2 phases is beyond my expectations.
Thirdly, I’m excited to see how VRM liquid cooling can be used to cost-optimize VRM design or to extract additional performance from basic designs. I know there’s much more at play when designing a VRM than the thermal solution. And, of course, liquid cooling isn’t a catch-all solution. But still, maybe something cool will come out of this.
Lastly, water cooling the VRM has become an exciting topic throughout my SkatterBencher overclocking guides. In most cases, the VRM thermal solution is sufficient to squeeze the most out of a CPU or GPU. However, I’ve encountered the VRM temperature as sometimes a bottlenecking system feature. I’ve seen this on low-end graphics cards like the GT 1030 or Arc A380, where cost-down VRM designs limit the overclocking range due to the thermal solution. But I’ve also seen high-end solutions for many-core processors, where the VRM would reach near its thermal limit when maxing out the CPU, even with a monoblock.
Soon, when I try the 56-core Sapphire Rapids Xeon w9-3495X, I’m confident I’ll see something similar. Hopefully, I can also test this with a water-cooled VRM.
Anyway, that’s all for today!
I want to thank my Patreon supporters for supporting my work. If you have any questions or comments, feel free to drop them in the comment section below.
See you next time!