Has anyone else experienced odd 13900K performance with an OC at 4.5 GHz?
Has anyone else experienced odd 13900K performance with an OC at 4.5 GHz?
I completed the setup for my friend’s identical machine. The key components were the CPU, motherboard slot, cooling solution, and possibly RAM and power supply. -i9-13900K -Asrock Z790 Taichi (BIOS updated) -64 GB Kingston Fury RGB DDR5 (2x32) 5600 MHz (XMP Stable) -Corsair HX1200i PSU with Custom Cablemod 24 Pin + 2x8 Pin for CPU -EKWB Quantum Performance S480 mm Radiator (roughly 30-35 mm thick), Pump & Res: EKWB TBE 300 D5, EKWB Quantum Velocity 2 Block, EKWB Quantum Torque 12/16 fittings, EKWB Clear Flexible 12/16 tubing, Corsair XL5 Clear Liquid -Around 16 fans. Front: 4x 120mm Bequiet Lightwings, radiator fans (top exhaust) 4x 120mm Bequiet Lightwings Highspeed, on PSU cover just sitting 2x Fractal Design 140 mm that came with the case, on GPU Backplate 2x Noctua iPPC 3000 rpm 140 mm, on side next to the pump & res another Fractal + Noctua iPPC 140mm 3000 rpm, back of the case for exhaust Noctua iPPC 140mm 3000 rpm (Yes, a lot of ippcs, but I got them for free so maybe just use them). Standard temperatures in my friend’s build were around 95-96°C with Cinebench running without power caps. My setup: 96-97°C after just one minute of Cinebench testing, reaching about 40.000–40.100°C. No power limits were applied. With Intel’s power limits enabled at 253W, it dropped to 87°C and 38.900–39.000°C. I opted for an undervolt approach since the board required it. This meant adjusting voltage offsets—negative bias on core, E-core, and ring voltages. I experimented with many settings but never achieved stability at -120 mV. Eventually settled on -110 mV across all parameters. On Cinebench rendering, performance remained steady. Maximum temps stayed around 85°C; cores peaked near 71°C, while P-cores hit 5.5 GHz. Power consumption was 245W and CBR reached 40,000 points. I’m curious—are there other tweaks I missed? Could adjusting clock speeds or voltages further help? I’m aiming for a 4.5 GHz run if possible without exceeding 90°C.
The 13700K runs at 5.6GHz P cores, 4.5GHz E cores, and a stable 4.8GHz ring in OCCT small FFTs, making the 4.5GHz E cores quite manageable from a 13900K perspective—especially if it holds up under Cinebench stability. It’s important to note that Cinebench isn’t a true stress test; I can maintain those voltages for extended periods during Cinebench R23 without issues, though OCCT will likely fail soon afterward. If you still wish to continue overclocking or undervolting, examine the PLL settings and experiment cautiously. These parameters usually don’t impact CPU temperatures much and can lower required voltages, potentially allowing deeper undervolting. Just keep in mind they tend to perform best just below the optimal range, not right at it. You’ll need to check your BIOS for specific values, as these parts are typically found on LGA 1700 boards—most of my testing has been with MSI models.
Sorry for the delayed reply, thank you for your time—I really appreciate it! Cinebench works well as a stress test for my workflow, especially with video, animation, and 3D rendering. The temperatures tend to be higher compared to other software like Premiere, Davinci, Blender, Maya, and 3DsMax. I checked the PLL values on my Asrock Z790 Taichi and found them under "OC Tweaker" → "Voltage Configuration." There’s a section labeled "PLL Voltage Configuration" with several options: -P-Core PLL Voltage Offset, E-Core PLL Voltage Offset, Ring PLL Voltage Offset, System Agent PLL Voltage Offset, Memory Controller PLL Voltage Offset, and GT PLL Voltage Offset. Each description notes a range from 0 to 15 bins, with 15mV per bin. Adding more than five bins can expand the frequency range during overclocking, though the optimal bins vary per processor. You’ll need to experiment to find the best settings for your specific chip. Regarding the System Agent, it might require a fresh CMOS reset and you’re avoiding memory controller changes—so if you set offsets for P-Core, E-Core, ring, and GT, you’d likely use a value like 1 (15mV offset). This adds about 15mV to the CPU voltage and can help with further undervolting.
It isn't a true stress test yet—I've noticed random crashes even when settings work for Cinebench. It works well as a full-core thermal load, but beyond that it's limited. You probably don't need to tweak the GT PLL; it's mainly useful for iGPU and real power usage isn't high. The SA PLL could be worth adjusting, though I haven't seen it prevent POST issues. It does seem to help with ring clock performance. You might try entering 1 and using intervals of 1. I don’t have an ASRock Z690/Z790 board to experiment with, so I’m not sure, but based on the description it seems important.
Thank you. I'll check SA, Ring, P-Core and E-Core at 1. Also, will I use a negative voltage offset for SA Voltage (FIVR) instead of adjusting the PLL, keeping the voltage unchanged? Just like with the P-Core, E-Core and ring, I applied -110mV. Would adding an SA Voltage offset further reduce the voltage? This assumes I don't modify the PLL values.
I probably wouldn’t experiment with an undervolted SA because instability shows up in odd locations (like random blue screens on the desktop and GPU problems). The SA doesn’t consume much power—around 10 to 20 watts total—so undervolting it wouldn’t make a big difference, likely saving only about 5 watts. In the context of a 300W CPU, that’s essentially negligible.
Based on your understanding, the method would involve entering intervals of 1 and adjusting the P-Core, E-Core, and Ring offset voltages one step at a time. For instance, setting the PLL Offsets (P-Core, E-Core, Ring and SA) to 1 and then the FIVR offsets to -120mV on those same points. This approach aims to lower voltage without affecting temperature too much. Disabling undervolting protection in the BIOS might be a good idea if you're confident about the process.
The approach seems to have some theoretical basis, though real-world results may vary. You might achieve improvements of around 10-20mV, but testing it is likely the best course. Aim to keep PLL offsets under +350mV or adjust based on your board’s specs—likely you’ll find a suitable range. The small voltage drop from PLL won’t cause excessive power draw compared to the extra 20mV from VCore, so it should still be beneficial. It probably won’t affect protection circuits significantly, though if you notice unexpected crashes, consider disabling it temporarily. For brownout protection, leaving it on is reasonable.
I'm ready to try it. What does +350mV or +24 mean in your setup? You're referring to a +24 on the bin level, with each bin around +15mV—so roughly two bins? Should I limit myself to two entries for PLL offsets? Also, your board shows -20mV as a significant change. From my tests, even small differences like -10mV can mean 2-3c variations. If it stays stable at -20mV, that could save up to 4-6c, which at 84°C would be crucial. If I can keep it under 80°C on full loads, I'm really happy. Even trying an OC like 5.6-5.7 (slightly ramped) might help, though my main goal is to run it cool right out of the box (5.5 GHz P-Core).