What's your ideal performance level?
What's your ideal performance level?
I agree that Ryzen Master is a user-friendly program, and I find it helpful. For stability, I think it's better to use Ryzen Master first, then once you reach a stable overclock, adjust your BIOS accordingly. The 3900x isn't really meant for gaming; it's designed for workstations. You won't get the full performance from a gaming version of that chip—that's just a truth, not a weakness.
When it comes to overclocking, I've repeatedly emphasized that 1.35 volts isn't safe on Ryzen 3000 models. The confusion started with misinformation from AMD during the early days, but that's not accurate. The actual safe voltage varies by individual chips. You should test it—run P95 for a long time, check small FFT results, and see the safe limit for your specific processor. I've noticed chips in the 1.35V range, some higher, and others lower than that.
Some users on Reddit tried overclocking two 3600s and saw significant damage after even a small voltage spike. While I don't say you're doing it wrong, it's rare for owners to know this and many just rely on vague "safe" numbers.
Regarding gaming, new Ryzen CPUs are built to their maximum potential out of the box. Overclocking them is challenging, but performance between 3700x and 3900x isn't noticeable. Keeping your CPU stock and using PBO + AutoOC lets it boost cores automatically during light tasks, improving efficiency.
Sticking with stock settings and enabling PBO + AutoOC is usually best unless you need maximum performance for demanding work. Running it at full stock with PBO can cause instability or heat issues, potentially shortening its lifespan. It's crucial for every Ryzen user to understand this before experimenting, as many have ruined their gear without realizing the risks.
I used AutoOC and got a core speed of 4.750 in RM, but Cinebench on one core showed a max of 4.330. With PBO it was 4.600 in RM and stayed under 4.475. Based on my profile it was around 4.300. All tests showed big fluctuations, then recovery, while PBO and AutoOC kept staying above 4.3k most of the time. That inconsistency affects performance, especially during dips. The drops in PBO and AutoOC are usually still above 4.3k, which is better than setting all cores to 4.3k. I wasn’t very impressed with the single-core boost, and running at 1.42v didn’t really help much. The temperatures were manageable at 75°C max. Thanks for the chat—it was cool talking this through, especially since I’m not sure many of my friends understand CPU stuff or overclocking.
I really like it, it's quite engaging, which is why I wrote the novels. I'll check later when I'm near my main rig, but in my bios I think I managed to turn on PBO and then enabled the AutoOC offset (+200mhz or something similar was an option). Then I just let her run. Concerning your 1.42V on a single core, the main point is to be cautious with voltage. For light workloads, Ryzen can safely raise the voltage to support those extra cores or threads. That's why browsing or doing light tasks usually sees the CPU automatically adjust to 1.4+ volts—it's safe. It only poses a risk under heavy loads when you apply too much voltage to all cores.
Thanks for the suggestion! I recall seeing PBO earlier but wasn’t sure about it until I looked again. Gamers Nexus is currently overclocking a 3100x to 5k on LN—it's really fun to watch.
I was experimenting with something comparable on my 3600, aiming for optimal performance per watt. It would face tough cooling challenges, so I pushed it to around 1.075-1.125v where performance per watt remained consistent across stable clocks. I reached only 4GHz at 1.11v, though the results seem weaker from what I understand. PBO could boost it to 4.05GHz on a single core but would cause a sharp drop in multi-core performance. I’m satisfied with the current setup.
He did the same action with the 3900x, but I don't believe he succeeded.
I've had to sell my 3900X stock. It seems I missed out on the silicon chance in every possible way except CPU performance. At 4.1Ghz with 1.35v it's not very stable and will overheat to around 95°C under full load. I'm using Dark Rock Pro for cooling. The stock will go through the same test at 95°C around 4Ghz, so I'll keep it as is.
Have you experimented with undervolting the unit? You mentioned 1.35v, but have you considered reversing the approach? I noticed significant temperature drops. Progress came from gradual adjustments. I began at 1.25v and 3.8k, then lowered to 1.20v at 3.8k once stability was confirmed in Cinebench. I moved to 3.9k, then reduced to 1.15v at 3.9k and further to 1.15v at 4k, etc. I changed one parameter at a time—either voltage or clock speed—to keep variables minimal. I relied on temperature and clock data, aiming for a balance between performance and thermal comfort. I was surprised by the stability at 4.3k with just 1.1v across all cores at 70°C. It’s not flawless; longer-term testing is needed, but it seems more reliable than pushing high voltages. You’ll likely encounter a point where voltage drops too low and the system crashes, so I increased it incrementally. Eventually, I settled around 1.05v and 4.1k, then raised it to 4.2k for better consistency. Overall, I found a reasonable compromise between speed and heat.