Introduction to Overclocking and The Flow Chart
Overclocking has become very common in the past few years as it's an easy method for increasing the performance of your PC to remove any bottlenecks that might exist. We present a quick guide on how to get started tuning your CPU to its maximum, and the good news is that the new 9900K CPUs are actually very similar to the last four generations CPUs, so overclocking is much the same. While Kaby Lake (7th Generation) CPUs were basically Skylake (6th Generation) CPUs, but with a better process (14nm+) and higher clocks, the 9900K (9th Generation) adds more cores and minor process improvements (14nm++).
When you add cores you also increase the chance of one core not overclocking to higher levels, and it only takes one core to decrease highest all-core overclock, and that is where the process improvements help. However, here is the crazy news. Intel's 9th Generations S-series CPUs don't statically assign cores to turbo bins, so the one core that needs to hit 5GHz during single core turbo is picked at random. That means that if you can cool the CPU well enough, you should be pretty much guaranteed 4.9-5GHz on all cores. Motherboard power delivery will also play an important role in this endeavor.
The Flow Chart
There are differences between the chart here and the one designed in our Skylake Overclocking Guide. For starters, the new CPUs do consume a bit more power, but they also can take a few more millivolts and maintain the same temperatures. Our starting voltages have increased a bit. No vendor has found a way to overclock non-K SKUs with BCLK. So, we are focusing much more on multiplier overclocking since it's more straightforward and there is basically no need for BCLK overclocking for the majority of people.
The basics of overclocking have not changed, you increase multipliers, and then increase voltages to help maintain stability. You hit a wall when your temperatures go over 80C under stress testing, which means you cannot add more voltage unless you increase cooling, so you can't add another multiplier and remain stable. With the 9900K you can also increase Tjmax levels, which will increase the throttle point of your CPU.
Our starting points have changed as well; you should start at 4.9-5.0GHz, as most CPUs can do that with ease at 1.3v or less (aim for under 1.3v to keep thermals in check with an CPU that isn't de-lidded). Our CPU was only able to hit 5.0GHz stable in most cases, but six cores at 5.0GHz with a reasonable VCore is better for our CPU than pumping it up.
Disclaimer and Before You Begin Overclocking
Disclaimer Overclocking your CPU technically voids your warranty. However, if you want to overclock and still be covered, Intel does provide an aftermarket overclocking warranty for about $19 located here: (PTPP). Overclocking can also damage your CPU, especially if done incorrectly. This guide is about how to overclock, but doesn't take responsibility for damages that could occur; you bare sole responsibility for any damages that may arise.
Have you Overclocked Before?
If you have overclocked before and understand hardware selection and the basics of overclocking, you should skip to the next page. The first part of this guide is for those who want to know what to do before overclocking.
Where do I start?
9900K Overclocking Guide Systems
- Motherboard: Intel changed power supply guidelines when switching between the Z370 and Z390 chipsets, and even though both chipsets will support overclocking, you will need a Z370 motherboard or Z390 motherboard with a very high-end VRM to get out every little bit of performance. Even at stock settings, power supply guidelines are much higher.
- DRAM: Intel and motherboard vendors have greatly improved DDR4 compatibility and speed potential, at least compared to Skylake. While we still recommend buying a kit off your motherboard's Qualified Vendors List (QVL), up to 3600MHz is a good target for easy stability. You don't need RAM that fast, but if you can afford it, you shouldn't have a problem getting it to run at full speed. Four sticks are harder to overclock than two, and if you get a four stick kit, you can't expect more than 3200MHz with ease (especially if it's a 64GB kit).You can find your QVL in the support section of your motherboard vendor's website or inside the manual (if differs by brand).
- Cooler: High-end air coolers are recommended, but most people get all-in-one water-cooling coolers since they offer the best of both worlds. They are easy to install, safe, and perform well. Your cooling is your ultimate limitation on your overclock when you are at above-ambient temperatures, so don't cheap out on a cooler.
- PSU: I would leave about 200-300W aside for a nice overclock on an 9900K/KF.
You enter the BIOS/UEFI by hitting "delete" or F2 (on most boards) during boot up. For most boards you have basic and advanced modes, I always skip to the advanced mode and tend to navigate with the keyboard. To enter a setting, you either type (or delete and then type), use +/- keys, or you click and scroll. Then you have to "Save & Exit" the BIOS/UEFI for the settings to apply (typically F4 or F10 key).
CPU Multipliers and Voltages
CPU core, AVX offset, FCLK, and Cache/Uncore multipliers allow you to overclock the CPU. The base clock of 100MHz is multiplied by each multiplier (ratio) and results in the final frequency. We have gone ahead and documented what each setting is called in each motherboard that might feature it.
MultiCore Enhancement is an easy way to overclock all cores to the maximum turbo frequency, which in this case is 5GHz. The issue is that multi-core enhancement adds too much voltage most of the time, and that overheats the CPU. However, you might get lucky and get a good CPU or motherboard that doesn't over volt the CPU. The stock all core turbo clock on the 9900K is 4.7GHz, enabling multi-core enhancement will increase that to 5GHz.
Start with 49x or 50x, and you can just type it in on most motherboards. On the Supermicro motherboard, you can set core six setting to what you want the rest of the cores to be at if you don't want to set each core individually. On ASUS, the board will auto-change to sync all cores if you set XMP to enable as it is enabling multi-core enhancement all the way.
While most games and other software don't use Intel's AVX instruction set, many stability testing programs and some software programs do. AVX uses parts of the CPU typically not used and as such will greatly increase power consumption, temperatures, and it could require more voltage to remain stable. The AVX Offset setting mas added a few generations ago to allow for different levels of overclocking depending on if AVX is detected in use. It allows you to set a number of multipliers for the CPU core to drop down to if AVX is engaged.
For example, if you set 50x for the Core ratio and -2 or 2 for the AVX offset, you will get a 5GHz overclock while playing games (no AVX) and 4.8GHz when running HandBrake (AVX). MSI lets you set a frequency for the CPU to shift to instead of a number of multipliers to go down when AVX is enabled.
The CPU cache, ring, and uncore are all the same thing, which mainly controls the frequency of the cache section of the CPU. Some vendors have turned cache up to 4.4+GHz by default, it depends on the CPU model. While you won't be able to match the CPU core multiplier most of the time, you should be able to stay 3-5 multipliers below the CPU core multiplier without requiring too much extra voltage. I would start with 4.4GHz cache ratio, overclock the CPU up to its maximum and then lower it a few to find a cache multiplier that doesn't require more VCore to stay stable. I should mention that VCore also provides power to the cache region and isn't a separate voltage rail like it was in the past.
VCore is your main voltage for stabilizing the CPU core and cache overclocks. Every vendor offers the ability to set the VCore to override mode, and it is the default on a few brands (like GIGABYTE), otherwise you have to choose. There are two other modes on most motherboards; adaptive and offset. If you run your CPU at maximum speed all the time, then override is what you want, but if you are going to let the CPU multiplier go up and down according to load, then you will want to set adaptive or offset mode. I will go over CPU multiplier dropping on the next page. The adaptive mode allows you to set a VCore you want as maximum and will drop down the VCore if the multiplier drops. It is better to use adaptive than offset, and it is easier to setup. Adaptive was introduced more recently, and before that we only had offset.
The offset mode allows you to set some millivolts to be added or subtracted to the CPU's default VCore for each multiplier. The new CPUs have different VCore levels for each multiplier (called VID), and every CPU has its individual default voltage levels. By default, Intel specification requires the VCore to drop by a certain amount when load and multiplier are increased. Load Line Calibration (LLC) reduces or reverses this default voltage drop, and it helps a lot when you are trying to stabilize the CPU.
If you want to set offset on a GIGABYTE motherboard, you need to type "normal" to unlock offset mode (offset is called DVID in their UEFI). Many motherboards will auto increase VCore beyond Intel specifications based on "auto-rules", so it's best to always set VCore, try 1.28-1.3v when at around 4.9GHz.
By default, motherboards should set FCLK to a multiplier of x8 (800MHz), but many vendors have set 1GHz by default. If your motherboard has it set at x8/800MHz, you should go and increase it to 1x or 1GHz, as it could help GPU performance slightly. FCLK typically does not produce instability when increased.
Power Settings and DRAM Overclocking
Power Saving Settings
Enhanced Intel Speed Step (EIST or Speed Step) and Intel C-States are the two mechanisms by which your CPU can down clock itself when idle to save power. Don't disable any of these settings and leave your Windows Power Plan to "Balanced" if you want to save power and have the CPU frequency drop at idle. You should also set your VCore to Adaptive or Offset mode, so it drops as well. MSI has an option (CPU Mode: Fixed or Dynamic) that will change the settings for you, but it's not hard to do on your own.
If you want your CPU multiplier to stay at maximum at all times, on some boards all you have to do is set Windows Power Plan to High Performance, on others you also need to disable C-States. EIST is part of how Turbo Mode works, which is how we are able to overclock, so on some boards if you disable it, you can't overclock the CPU. Window's High-Performance power plan and disabling C-States should allow you to run full speed all the time, but if it does not, you can also try disabling EIST (it depends on how the vendor has implemented the setting).
While we have power saving settings that we can manipulate, we also have power limitation settings that we might also need to increase. If performance is not scaling (try running CINEBENCH every multiplier step to see if performance increases), power restrictions are probably to blame. These power limits are set in place by default, and on some motherboards, these settings are automatically changed behind the scenes when you overclock, but not all. It is best to increase these limits.
On Supermicro motherboards, setting "0" will maximize the power limit. These settings are found in multiple places in most UEFIs (I mashed them up in the image above); check advanced PWM features, CPU configuration (sometimes not on the OC page), voltage, and CPU overclocking menus.
We always recommend using XMP memory profiles for easy memory overclocking. XMP is easy to use and will take the memory to speeds it was designed to operate at. You can also manually set the DRAM multiplier and change between 1.00x and 1.33x reference clock multipliers to expand the granularity of memory speed settings (so memory speed can be set in 100MHZ and 133MHz increments; i.e. 2600MHz, 2666MHz, 2800MHz, 2933MHz).
If you manually set the multiplier, you must also manually set DRAM timings and voltage, as XMP sets those for you. Memory timings are a bit more difficult to conquer as there are so many of them, but you can mess around with the primary timings (like CAS latency). Some motherboards like ASUS's ROG boards also have built-in memory overclocking profiles made by their extreme overclockers.
There are three main voltages for overclocking DRAM. DRAM voltage is the voltage that the DRAM is fed, and on most motherboards when you set XMP, the motherboard will automatically increase DRAM voltage. Some motherboards also have auto-rules for two other voltages, System Agent (VCCSA) and System IO (VCCIO). These two voltages can be set up to 1.3-1.35v, but I would start with around 1.2v on both of these if default voltage levels don't work for your memory overclock or if they work and the auto rules took them too high.
Be careful; some motherboards have auto-rules that will increase both VCCIO and VCCSA to 1.3-1.35v by default, essentially maximizing them but also increasing temperatures and power consumption. On those motherboards you can try setting them a bit lower.
Intel Max Voltages and Stability Testing
There are only two things you can do to help stabilize higher clocks. The first is to raise voltages, and the second is to improve cooling. Improving cooling is the best method to increase clocks and stability, as it only helps. Improving cooling is difficult and costs money, but increasing voltages is free. However, increasing voltage can also increase instability as the added heat and heat noise can affect stability. It's a double edged sword.
You can delid your CPU, use a custom water cooling loop, change thermal interface material, and even go sub-zero. You might find that reducing voltages lowers CPU temperature enough, so something like 1.3v could be stable while 1.4v would cause instability at the same frequency because of the added heat.
Intel recommends a maximum operating voltage of 1.52v VCore, but that's just way too high. I don't know of any watercooler that could cool down a CPU being fed 1.52v, but of course, that figure from Intel doesn't account for LLC since it's not part of Intel default specifications. I recommend 1.45v as maximum VCore with non-subzero cooling, but that doesn't matter since you will probably hit a thermal limit before you hit that voltage limit. The VCCSA (System Agent) is 1.05v stock, and the VCCIO (System IO) is rated 0.95v stock.
Even though Intel says maximum VCCSA can be 1.52v (which is a new addition to the 8th/9th generation datasheet) I recommend no more than 1.35v on either VCCSA not VCCIO. I honestly wouldn't run much more than 1.25v for 24/7 use. If you hit an issue with DRAM overclocking, try many different levels from 1.1 to 1.35v. I have seen memory controllers that don't like anything over 1.25v, so try increasing VCCSA first and then VCCIO, and it's fine if they are different levels.
Here is our 9900K chiming in at 5.2GHz with a 0-AVX offset, which when combined with Handbrake, which uses AVX, produces a pretty heavy load. For Prime95 testing we did after we did use an offset of two, because Prime95 is much more stressful. However, it's easy to run Handbrake first to see if the overclock even stands a chance. Blender also has a benchmark, which takes a bit longer than Handbrake, but which should be a heavier load and should be more indicative of overall stability. In the end Prime95 is what you will need to use to get the most out of your CPU
Prime95 has a few different tests you can run. The default test is a blend test, which tests most everything, but isn't going to demolish CPU core instabilities as Small FFTs would. There is some documentation on how to disable AVX use within Prime95 by adding in a string in the local.text file created when the program runs, but from my testing, that method didn't work.
AIDA64 also has a built-in test, and you can choose what you test (core, FPU, etc.) but I would increase the amount of RAM used in that test as the default value is a bit low. AIDA64 is considered a "safe" test by many, as it's not designed to beat your CPU up, but many believe you can't become a man until you get a really good beating. When you run Prime95 keep an eye on temperatures, as you can see I am in the danger zone (80-90C) already with only 1.3v and 5GHz (5.2GHz with -2 AVX offset)
If your CPU is thermal throttling you can use the new TJmax offset setting added to the 9900K, which will allow you to increase throttle point. Some suggest adding 15C, but we think it's a bit unsafe to use this offset to that extent. The good thing about Handbrake is that it spits out an average work encode speed in FPS, so you can gauge if your performance is increasing or not.
Overall, overclocking the 9900K is quite easy, but you should keep in mind that overclocking can degrade your CPU faster than naturally intended, but most enthusiasts would agree with the long lifespan of modern CPUs and the overall feeling of accomplishment make it worth it.