Intel Core i9-12900K Overclocking, Power and Thermals
The Alder Lake chips are easily the best overclocking chips we've seen from Intel in several generations. As you'll see on our gaming page, that results in up to double-digit performance increases in gaming performance in our test suite.
Unless stated otherwise, our overclocking settings are always geared to emulate what we expect a standard enthusiast can achieve with a reasonable type of conventional cooling for 24/7 use (we use a 280mm Corsair H115i AIO as our default cooler). As such, you can tune more aggressively or employ better cooling solutions to attain even higher results, but as always with overclocking, your mileage may vary.
We initially tuned the Core i9-12900K's P-Cores to 5.2 GHz and set the E-Cores to 4.0 GHz (all-core on both), but while these settings proved perfectly stable, the chip overwhelmed our 280mm Corsair H115i under certain heavy loads — a single core sporadically reached 100C. We feel that a 360mm AIO water cooler, or a custom loop, could handle the thermal output of a 5.2 GHz overclock. There could also be slightly more frequency headroom as well: Thermals, not silicon quality or ability, are the bottleneck with this chip.
To keep temperatures in check, we dialed in a 5.1 GHz P-Core and 3.9 GHz E-Core overclock with a 1.29V CPU vCore. We also assigned a -2 GHz AVX offset to align with the same all-core AVX clock speeds we observed with the chip at stock settings. This means the chip runs at 4.9 GHz when it encounters AVX workloads, allowing us to reach higher with other, more common types of work (like gaming) while staying within a comfortable thermal envelope during taxing AVX-heavy work.
We also dialed in a 4.2 GHz fabric clock, which is a whopping 1.6 GHz above the stock setting of 2.6 GHz. This is also 200 MHz beyond what we've been able to achieve with Intel's prior-gen chips, marking a big improvement to fabric capabilities. Finally, we assigned memory to DDR4-3800 with 14-14-14-34 timings in Gear 1 mode. Experience has taught us that staying in Gear 1 mode provides the best overall performance in most applications. It is noteworthy that we've seen several results with DDR4-4000 in Gear 1 for everyday usage, which is a significant improvement over the Rocket Lake chips that tapped out at DDR4-3800 in Gear 1. You might have to adjust the VccIn (or equivalent) to enable this slightly higher level of performance, but we stuck with DDR4-3800 to emulate a bog-standard overclock.
As you can see above, the chip easily ran at a 5.1 GHz all-core overclock while staying under the 100C threshold through a series of Handbrake, POV-Ray, Y-cruncher, and Blender runs (these are our most stressful real-world applications, and all but Blender employ heavy AVX workloads). Notably, the 12900K can hit 5.3 GHz on two 'preferred' cores during normal operation, so we're sacrificing 100 MHz of peak clock speeds with this config. That manifests as a slight performance decline in a few entirely single-threaded workloads, but the impact was rare. It's also worth mentioning that very few modern workloads are truly single-threaded.
The 12900K peaked at 277 Watts, but we only saw those peaks during the multi-threaded y-cruncher tests that heavily employ taxing AVX instructions. Temperatures peaked at 97C for a brief moment, but largely hovered in the low-90's during the most stressful workloads. Again, thermals are the limiting factor, but you can do more complicated overclocks that prioritize a few cores with higher peak frequencies or step up to a beefier cooler to unlock higher overclocking potential.
Intel Core i5-12600K Overclocking, Power and Thermals
The Core i5-12600K is a tuner's delight: Core i5-12600K comes with quite a bit more thermal headroom due to its lower core count, meaning there's more room to play, and it also has plenty of frequency headroom.
We dialed in an all-core 5.0 GHz P-core and 3.9 GHz E-Core overclock with a 1.33V vCore (notably, that's slightly higher than the 1.29V we used for the 12900K). We also employed a -3 GHz AVX offset, 4.2 GHz fabric, and dialed in a DDR4-3800 Gear 1 memory overclock with 14-14-14-34 timings.
With our H115i 280mm watercooler, albeit with the fans cranking away at full speed, we peaked at 96C for a single one-second measurement during our stress tests. The chip hovered in the 75C to 85C range during the majority of the tasks. We could absolutely wring more out of this chip, and the 5.0 GHz overclock is above the 12600K's peak 4.9 GHz speed at stock settings, which is becoming increasingly rare with newer chips.
Alder Lake's Thread Director technology works best with Windows 11, so we tested with a fresh install in addition to our standard Windows 10 test image. We updated to newer versions of our benchmarks for Windows 11, where applicable, and also added a few new application and gaming benchmarks.
In accordance with AMD's official guidance, we assured that our clean-install Windows 11 test system had all patches for a recent AMD L3 cache bug that impacted AMD processors. AMD says the patches resolve the L3 issues, and our own in-depth testing has also confirmed that the patches are successful. As an additional level of caution, we re-installed the chipset drivers every time we swapped chips and retested L3 cache latency for each chip both before and after each series of tests.
We also tested with secure boot, virtualization support, and fTPM/PTT active to reflect a properly configured Windows 11 install. This can drop performance as well, on both AMD and Intel CPUs.
Given that Alder Lake will also be used with the less-than-optimal Windows 10, we also tested with our existing test image (build 19041.450). This version of Windows isn't the latest, but Intel confirmed that it has the same Lakefield scheduling optimizations as newer versions of Windows 10, meaning that our testing is representative of trends you'll see in the real world. However, even though we have a few identical tests with both operating systems, you shouldn't interpret our results as being fully indicative of Windows 10 versus Windows 11 performance.
Our historical Windows 10 gaming results for Ryzen processors were dated. Several of the games have since had performance-impacting updates, not to mention that numerous chipset and BIOS updates have been issued in the interim. As such, we updated the chipset drivers and motherboard BIOS to reflect the current state of play and then retested all of the game benchmarks.
With both Windows 10 and 11 covered, we also wanted to measure the difference between DDR4 and DDR5 performance on both operating systems. We used the MSI Z690 Carbon WiFi as our DDR5 platform and the decidedly lower-end MSI Z690-A WiFi for DDR4 testing.
We're sticking with our standard policy of allowing the motherboard to exceed Intel's recommended power limits, provided the chip remains within warrantied operating conditions. Our tests use the default lifted PL1 and PL2 restrictions. Almost all enthusiast-class motherboards come with similar settings, so this reflects the out-of-box experience with a high-end motherboard. Naturally, these lifted power limits equate to more power consumption, and thus more heat, as we'll cover in detail later in the review. We also have a full breakdown of the test system configurations at the end of the article.
|Intel Socket 1700 DDR5 (Z690)||Core i9-12900K, Core i5-12600K|
|Row 1 - Cell 0||MSI Z690 Carbon WiFi|
|Row 2 - Cell 0||2 x16GB G.Skill Ripjaws S5, DDR5-5200 @ DDR5-4400 36-36-36-72|
|Intel Socket 1700 DDR4 (Z690)||Core i9-12900K, Core i5-12600K|
|Row 4 - Cell 0||MSI Z690A WiFi DDR4|
|Row 5 - Cell 0||2x 8GB Trident Z Royal DDR4-3600 - Stock: DDR4-3200 14-14-14-36|
|Intel Socket 1200 (Z590)||Core i9-11900K, Core i7-11700K, Core i5-10600K|
|Row 7 - Cell 0||MSI Z590 Godlike|
|Row 8 - Cell 0||2x 8GB Trident Z Royal DDR4-3600 - Stock DDR4-3200/2933 Gear 1|
|AMD Socket AM4 (X570)||AMD Ryzen 9 5950X, Ryzen 9 5900X, Ryzen 7 5800X, Ryzen 5 5600X|
|MSI MEG X570 Godlike|
|Row 11 - Cell 0||2x 8GB Trident Z Royal DDR4-3600 - Stock: DDR4-3200 14-14-14-36|
|All Systems||Gigabyte GeForce RTX 3090 Eagle - Gaming and ProViz applications|
|Row 13 - Cell 0||Nvidia GeForce RTX 2080 Ti FE - Application tests|
|2TB Sabrent Rocket 4 Plus|
|Row 16 - Cell 0||Open Benchtable|
|Row 17 - Cell 0||Arctic MX-4 TIM|
|Windows 10 Pro version 2004 (build 19041.450)|
|Row 19 - Cell 0||Windows 11 Pro|
|Cooling||Corsair H115i, Custom loop|