I recently upgraded from a Core i5-14600K with an ASUS TUF B760-Plus WiFi board to a Ryzen 9850X3D with an X870E Aorus Elite revision 1.1. I'm looking for a stable, mostly set-and-forget configuration that provides good performance without unnecessary voltage, heat, or long-term degradation.
So far I've tried a 1.15–1.18 V SoC voltage, 1.25 V VDDIO, and 1.35 V VDD/VDDQ for DDR5-5600 at 32-36-38-38-68-114 timings. I'm using a 65K tREFI because there's no dedicated fan over the memory, FCLK is set to 2133 MHz, UCLK is synchronized with MEMCLK, and Curve Optimizer is set to negative 15 on all cores.
I set the CPU thermal limit to 88°C instead of allowing it to reach the default 90–95°C range. PBO is enabled in advanced mode, but its limits and scalar are still on Auto. Boost override is disabled because I want to understand the safe limits before trying to increase boost clocks. I have not changed LLC or used Curve Shaper yet.
I'd appreciate recommendations for safe PBO, voltage, and per-core Curve Optimizer settings, along with reliable software for testing the CPU and memory together and separately. I'm also unsure how to interpret occasional voltage spikes and whether DDR5-5800 or DDR5-6000 would be worthwhile. A DDR5-5800 configuration passed separate CPU and memory tests but produced errors in a combined CPU-plus-memory test, so I returned to DDR5-5600.
2 Answers
If longevity is the priority, avoid leaving the important PBO controls on Auto. A reasonable starting point is scalar 1x with stock-style limits around 162 W PPT, 120 A TDC, and 180 A EDC. The processor’s normal temperature limit is designed to be safe when voltage and current remain within the intended limits, so lowering it to 88°C is optional rather than essential.
Boost Override does not force the CPU to run at that frequency constantly; it only raises the maximum target when temperature, voltage, and current headroom allow it. The motherboard’s default LLC is generally the best choice unless monitoring shows a genuine overshoot problem. For tuning, per-core Curve Optimizer is usually a more practical compromise than Curve Shaper, and Ryzen Master can be used for testing changes without repeatedly entering the firmware.
Use HWiNFO to observe effective clocks, CPU voltage behavior, temperatures, and power rather than relying on a single sensor reading. Also test the memory and CPU together, since separate tests can pass while the combined memory-controller workload still produces errors. If the kit can run DDR5-6000 reliably, that is typically a better target, but stability matters more than the headline frequency. Memory subtimings such as tRC, tRFC, tRRD, and tFAW can have a larger gaming impact than simply raising frequency.
DDR5-6000 is worth trying if the memory controller and kit can handle it, but do it incrementally and validate every change. A small increase from 1.35 V toward roughly 1.38–1.40 V may help some kits, though that is not automatically appropriate for daily use—especially with Samsung memory, limited airflow, and higher ambient temperatures.
Use a combined workload such as OCCT CPU plus memory, along with longer memory testing, because a configuration that passes CPU-only and RAM-only tests can still fail when the integrated memory controller is stressed at the same time. If tighter DDR5-5600 or DDR5-5800 timings provide lower latency and better real-world results, there is no reason to chase 6000 MHz just for the number.
That makes sense. I’ll compare latency and gaming performance instead of assuming DDR5-6000 is automatically faster, and I’ll prioritize a configuration that remains error-free under combined testing.
I’ll switch to HWiNFO and check the readings more carefully. The DDR5-5800 setup passed CPU-only and memory-only tests but failed OCCT’s combined test within about 40 seconds, so I’m treating that as unstable for now. I’m hesitant to raise memory voltage to 1.40 V because the modules are Samsung-based, there’s no direct airflow over them, and the room can get warm.