Increase CPU cache speed through overclocking.
Increase CPU cache speed through overclocking.
CPU performance: 11900K. MOBO model: Asus Maximus HERO XIII. Vcore setting: adaptive with +0.001 offset. XMP level 1.5GHz. All cores support AVX2 and AVX512. Offset: 0 AC/DC: 0.001 (or lowest possible). LLC: 8 VID for 5.1 at -0.15 (required only for 5.2). VID for 5.2 at -0.22.
Average vcore under load: ~1.394. Average idle vcore: ~1.403. Max vcore reached: 1.421. Max package power: 413W. Avg before AVX: 4.3W. After AVX enabled: 4.0W.
Testing without AVX: Blender running for 20 hours (with LLC4 and VID -0.04). Idle vcore was 1.51, instant 413W in under a minute full load. With the above settings, only 413W with both AVX enabled.
With ffmpeg encoding a 4K movie using libx265 at slow speed: 8 hours without changes, no problems.
Regarding uncore: I had to turn it off in BIOS to reach values above 4.0. Setting uncore to 4.3 boosted vcore to ~1.43. It didn’t stay high for long. There are other BIOS settings like System Agent, CPU VCCIO, Mem VCCIO that affect performance but mainly impact RAM.
My question: Is there a specific voltage I can adjust that would improve cache speed without increasing vcore? Or will any change still impact it? I’m unsure if higher cache always helps or if it’s more about other factors.
It aids many parallel tasks and when memory is beneficial, OC it makes sense. Stress testing isn’t exactly enjoyable. VCCSA and VCCIO support the cache a bit, boosting stability, though VCore has the biggest impact. Stick with a steady VCore voltage—it improves system steadiness and idle power use without much concern. You could gain an additional 100-200MHz on UCLK without extra effort.
I mostly work with coding, mainly in C++, and some video encoding focusing on compression with HEVC. Occasionally I handle video editing and 3D design. The voltages I see when idle are only that high because I have c-states turned off and the system is running at full CPU power in Windows at 100%. These settings are just temporary. After completing overclocking, I plan to turn on c-states and set the minimum CPU state to 0%. This would keep idle vcore below 1v. That’s why I’m avoiding a fixed vcore value. How can I reduce the vcore further? I tried adjusting VID to 5.2 but it didn’t change the vcore.
Compiling code usually gains from quick cache performance, which is definitely worth checking. I understand that point, but tweaking VID parameters adds more complexity and can cause issues. The slight reduction in power usage during idle (2-3W) doesn't justify the added instability and inconvenience of using an offset instead of a fixed one.
The reported idle CPU power of 137W seems excessive; it should align more with 50W–60W. Consider testing it and verify the actual consumption.
Currently idle with no c-states and minimum CPU state at full capacity, the lowest power package is 112. When changing the CPU state to 0%, it drops to 86. Setting C-States to auto brings it down further to 56. The idle voltage is 1.43, likely due to the LLC being at level 8.
It fits perfectly. 1.43V is fine for idle, and anything above 1.47V works as long as it doesn’t cause thermal throttling under load. You seem to be targeting the lowest possible LLC while keeping idle voltage reasonable.
It's more than enough, but I've tested chips with disabled C states at higher voltages than idle for extended periods. As long as it's not extremely high (>= 1.48V idle), I'm fine. It seems the voltage offsets were just damaged in the LLC level 8 stage. Still, I'd prefer a static voltage since it reduces transients and boosts stability significantly.