All of the following can lead to failure of computer components. Neither the editors nor the author bear any responsibility. You act at your own peril and risk!
Today is a special day — the release of the long-awaited ClockTuner for Ryzen (CTR) project, which can affect the performance of systems based on Ryzen processors with Zen 2 microarchitecture. Before starting a tour of CTR and recommendations, I would like to touch on the backstory, the very thing that inspired me to create this software. There will be no superfluous information, so I hope that you will read absolutely everything. This will avoid a number of problems and misunderstandings.
I think many of you remember the release and the first reviews of Ryzen processors based on the Zen 2 microarchitecture. They brought both a new level of performance and a price reduction for Intel processors. The growth of AMD shares and unprecedented hype on the forums. The delight of users and another portion of fuel to the fire in the form of a cool bonus — Precision Boost Overdrive (additional automatic overclocking). In short, PBO is a technology that allows you to change the limits of power limits that limit the boost of the processor, and thereby increase the frequency in auto overclocking. In addition, the user is offered the opportunity to change the frequency curve by a certain amount. For those who have not seen the presentation of this technology, you can familiarize yourself with it here:
Users were promised an additional processor frequency depending on the «success» of the sample, the cooling system and the VRM capabilities of the motherboard. Many even believed that smart overclocking did exist and that it would bring results. On the first UEFI firmware, some users reported that PBO is indeed capable of raising the frequency by an impressive amount up to 200 MHz, but only on Ryzen 5 3600 processors and at the cost of huge power consumption. The rest of the CPUs were left out, they were not allowed to change the frequency curve. After some time, it became clear that no one is working on this technology and other processors will not receive the functionality that Robert Hallock told us about.
As many of you already know, the new manufacturing process is a series of complexities at all stages of production and processing, so TSMC’s 7nm is no exception to the rule. One of the most important points is the sorting of crystals, which allows you to create both very energy-efficient processor lines and high-frequency gaming solutions. In addition to determining the energy characteristics of the crystal as a whole, the energy characteristics of each nucleus are also determined individually. That is, one crystal can contain both a number of very successful nuclei capable of conquering a very high frequency, and a number of less successful nuclei. In theory, this allows the processor manufacturer to use all the cores at maximum efficiency and vary the processor frequency depending on the number of cores loaded. Sounds cool, doesn’t it? In practice, things are a little different. Since the process of energy performance assessment is complex, and the time frame for assessment is limited (time is money), a simpler way of assessing energy performance is used. As a result, we have a certain sample of the processor, with information about its approximate capabilities. Accordingly, in order for each sample to work as it should, the operating voltage will be selected according to the worst sample. The disadvantages are incorrect core labels, excessive power consumption (and heat dissipation) if the user comes across a successful processor instance. In other words, such a sample has a hidden reserve.
Fortunately, AMD was also able to provide a counterbalance to these popular shortcomings (intel has exactly the same shortcomings) — individual overclocking of each CCX. Let me remind you that CCX (Core CompleX) is one of the structural units of the processor, which can include up to 4 cores, caches and other related modules. The Ryzen 3000 series line has products where the number of CCXs can range from 1 to 16 (Threadripper 3990X). However, with such a powerful tool as overclocking with CCX, most users (95%) will have no idea what frequency to set for each CCX and what voltage should be. Processors with four CCX or more can shock the novice enthusiast.
Since this is all a huge complexity for many, and the reserve of processors is not used, I came up with the idea to create software that can help users easily and easily get free performance, if such, of course, is available.
System Requirements and Getting Started
For the program to work properly, the system must meet the following requirements:
- AMD Ryzen processor with Zen 2 architecture (Renoir is temporarily unsupported);
- UEFI with AGESA Combo AM4 126.96.36.199 (and newer);
- SVM (Virtualization) — it is recommended to disable (optional);
- CPU Voltage — Auto (UEFI);
- CPU Multiplayer — Auto (UEFI);
- stable RAM overclocking or stable XMP mode;
- Windows 10 x64 1909–2004 build;
- .NET Framework 4.6 (and later);
- Ryzen Master 2.3 (using monitoring driver).
The next set of requirements is also mandatory and applies to UEFI settings. Since the success of CTR is highly dependent on the VRM capabilities of the motherboard (I highly recommend reading this chapter), we need to make some kind of foundation in UEFI in order to secure the system tuning process from BSOD.
The most important setting is LLC (Load Line Calibration), my recommendations are as follows:
- ASUS — LLC 3 or LLC 4;
- MSI — LLC 3;
- Gigabyte — in most cases Turbo, but can also be Auto;
- ASRock — Auto or LLC 2; An important point, CTR has mediocre compatibility with ASRock motherboards, since all LLC modes show abnormally high Vdroop, i.e. strong voltage drop;
- Biostar — Level 4+.
For owners of ASUS motherboards, it is recommended to use additional settings:
- Phase mode — Standard;
- Current capability mode — 100%.
How to set CTR:
- Download (this is the only official source) the CTR.7z archive and unpack it in a place convenient for you.
- Download the Cinebench R20 archive and extract the contents of the archive to the «CB20» folder (located in the CTR folder).
- Launch Cinebench R20, accept license agreement, close Cinebench R20.
- CTR is ready to go.
Behind absolutely any project, in addition to the developer and the author of the idea, there are people who also made their contribution. I want to express my gratitude to all those who took part in testing, advised on technical issues or found important information that could improve CTR. In particular, special thanks to:
@A_z_z_y (Vadym Kosmin), Martin Malik, @CodeZ1LLa (Oleg Kasumov), @lDevilDriverl (Oleksii Baidala), Sami Makinen (AMD), Danny Ordway (ACI), @Spaik (Alexey Savitski), @datspike (Alexey Elesin), Keaton Blomquist, @tsa, @cluster_edge, @Anem (Anton Emashov), @GUN’G’STAR, @CapFrameX, @irusanov, @stormpand.
CTR also contains third-party modules:
- Ryzen Master SDK from AMD is the main monitoring module.
- LibreHardwareMonitorLib — cpu svi2 and soc svi2 monitoring.
- Maxon’s Cinebench R20 is everyone’s favorite benchmark.
- Prime95 by George Woltman is the best comprehensive CPU stress test.
- Reverse engineering version of Shamino’s CCX Work Tool — as the basis for access to SMU.