I upgraded from a Core i5-14600K with a B760 motherboard to a Ryzen 9850X3D and an X870E Aorus Elite. I'm looking for a conservative, set-and-forget configuration that improves efficiency and performance without unnecessarily increasing voltage, heat, or long-term degradation.
My current settings are VSOC at either 1.15V or 1.18V, VDDIO at 1.25V, and memory voltage at 1.35V. The DDR5-5600 kit is running at CL32-36-38-38-68-114 with 1/2/0 Nitro settings. I set tREFI to 65K because the memory has no dedicated airflow, FCLK is 2133MHz, and UCLK is synchronized with MEMCLK. Curve Optimizer is set to negative 15 on all cores.
I lowered the thermal limit to 88°C, while PBO is set to Advanced with automatic power limits, scalar set to Auto, and boost override disabled. I have not changed LLC or used Curve Shaper because I'm still learning how those settings affect stability and voltage behavior. I'd like to maintain roughly 5.6–5.7GHz boosts without excessive voltage or heat, and I'm wondering whether per-core Curve Optimizer values would let me undervolt further.
I also tried DDR5-5800 with different timings and an FCLK of 2000MHz. CPU-only and memory-only tests passed, but OCCT's combined CPU-and-memory test produced errors within about 40 seconds, so I returned to DDR5-5600. What settings and testing tools would you recommend for safely tuning the CPU and memory together?
2 Answers
For a longevity-focused setup, avoid leaving scalar and the PBO power limits completely open. A reasonable starting point is scalar 1X with stock-like limits such as PPT 162W, TDC 120A, and EDC 180A. The default 95°C temperature limit is generally designed for sustained use when the normal voltage and current protections are active, so lowering it to 88°C is optional rather than essential.
Boost override does not force the processor to run at a higher voltage all the time. It only gives the boost algorithm additional headroom when temperature, voltage, and current conditions permit it. A modest +100MHz setting can be tested, but stability and temperatures matter more than the number itself. Leave LLC at the motherboard default unless reliable monitoring shows a specific problem; aggressive LLC can create larger voltage transients.
Use HWiNFO rather than relying on a single reading from a simpler monitor, and test with tools such as OCCT, Cinebench loops, y-cruncher, and separate memory tests. Ryzen Master can be used for experimentation, although BIOS settings are usually better for the final configuration. For gaming performance, tuning secondary memory timings can matter more than simply raising memory frequency.
DDR5-6000 is worth testing on this platform, but don’t assume it will work with the same timings or voltage as a faster memory kit. You may need slightly more DRAM voltage, perhaps around 1.38–1.40V, along with looser timings and carefully tuned subtimings. However, memory temperature and the specific memory IC matter, especially without direct airflow.
The combined CPU-and-memory failure is useful information: it suggests the system may be marginal when the memory controller and CPU are loaded together, even though each component passes by itself. That can point to the memory-controller voltage, fabric settings, timings, or simply an unstable overclock. Returning to DDR5-5600 is sensible until the whole configuration passes extended combined testing. A stable, tighter DDR5-5600 or DDR5-5800 setup can easily be preferable to DDR5-6000 with loose timings.
That makes sense. The kit uses Samsung memory chips, and my room can get warm, so I’m hesitant to apply 1.40V without airflow. I’d rather keep a stable 5600–5800 configuration if 6000 requires looser timings that hurt latency.
I’ll switch to HWiNFO to check the actual telemetry and try per-core Curve Optimizer values instead of assuming every core can handle the same negative offset. I was mainly concerned about brief voltage spikes and repeated BIOS restarts while testing.