Learn the correct steps to properly stabilize a Ryzen CPU overclock.
Learn the correct steps to properly stabilize a Ryzen CPU overclock.
Welcome to this guide about conducting stability tests for Ryzen CPUs! Adjusting your CPU’s multiplier is usually simple—just boost the multiplier until instability appears, then raise the voltages until you reach your target speed. This approach works well with Intel processors.
However, AMD Ryzen presents a different challenge. Its architecture requires more careful handling during overclocking. Because of how its design operates, Intel’s CPUs are straightforward, while Intel’s cores can be pushed independently of memory performance. You often have flexibility to increase CPU core ratios or memory ratios without causing conflicts.
But with Ryzen’s “more modest” IMC and Infinity Fabric design, overclocking the CPU core sometimes leads to instability in the memory controller, affecting multiple parts of the chip.
Here are the steps you should follow to stabilize your Ryzen CPU:
You’ll need a set of stress testing tools. I suggest using OCCT, Prime95 (the latest version is essential, and you may want to disable AVX instructions via a text file), and ROGRealBench.
Execute OCCT with an 8-10 hour stress test (Large Data Set).
Run Prime95 Blend for 8-10 hours.
Perform ROG RealBench for 8 hours.
Use Prime95 with a memory-heavy configuration, starting with the blend mode, then adjust these settings: 512k small FFTs, “memory in use” set to 75% of your total system memory (if this exceeds capacity, reduce it to 50%).
Continue this process for 8-10 hours.
Why these tools?
From my experience, OCCT consistently performs exceptionally well—it detects issues much faster than Prime95 and often fails first. It’s an excellent starting point.
Prime95 blend is ideal because stress tests require multiple runs; no single program should stress every component. P95 acts as a backup.
ROG RealBench is another reliable stress tester, useful as extra assurance for stability.
Prime95 with a memory-intensive setup puts significant pressure on your memory, ensuring it can handle the overclock.
When tests fail:
(Assuming you’ve already verified memory and CPU stability at stock speeds, XMP profiles enabled, and confirmed they’re stable.)
If OCCT fails during your overclock, raise both SOC and memory voltages temporarily to a safe but high level. This isolates instability to the CPU cores.
For example:
- SOC = 1.13v
- RAM = 1.4 (or 1.45v if using XMP profile)
If P95 fails, it’s usually because the vcore isn’t strong enough.
ROG RealBench failure suggests GPU or CPU issues.
P95 with memory-heavy settings points to unstable memory or IMC. Adjust voltages accordingly—either boost them or lower memory speed/loosen timings as needed.
After stabilizing, gradually reduce your IMC voltages and test again. Always stress-test each adjustment.
Once stable, consider increasing the vcore voltage by 1-2 notches above your validated safe limits. This helps ensure long-term stability. This step is particularly important for Ryzen due to its steep frequency-to-voltage curve. Intel’s curve is flatter, so such adjustments are less critical.
Good luck with your overclocking on Ryzen! It’s a complex process, but definitely worth the effort.
Is there a role for LLC in addressing vdroop and boosting stability at higher clock speeds?
8 to 10 hours of OCCT appears to be a waste of time.
If it doesn't stop within 5 minutes, I'm usually fine.
Temperature reaches nearly the maximum in just 3 to 5 minutes, even with a Closed-loop Liquid Cooler (CLC).
CLC always has an impact when under load. If voltage drops too much, you'll crash.
Almost all seasoned overclocker agree that 5 minute stress tests are meaningless. I've seen OCCT crash after just six hours. While some testers pass, others fail because of instability. That's why extended testing is essential. It really depends—some people only want a system that works halfway stable for gaming. In truth, you should run Prime95 for about 24 hours, but no one has the time for that. The improvements are minimal at best.
Many people just change their settings to achieve the fastest boot possible, aiming to avoid any performance issues. Personally, I favor consistent stability when boosting my CPU. I test my system overnight, but few can spare the time for a full 24-hour stress test.
I’d rather have reliable performance than risk a slight speed increase that might not be enough for gaming.
Absolutely, that could work well for you. It really depends on what someone prefers.
Same here, I’d prefer to fine-tune my overclocks by lowering the frequency by a few hundred hertz and cutting the voltage core down a lot. This way, I can handle all stress tests without getting hot, and it’ll feel like a very stable, rock-solid overclock that even the factory would approve of.
Great work
TechyInAZ
I find all this information really helpful. Thank you for investing the time to share. Your approach to stress testing is quite thorough compared to my own. I’ve experimented a lot with overclocking on Zen and Zen + using various CPUs, and I’d be happy to exchange some insights.
Generally, after updating the BIOS, installing Windows, drivers, and software, I run default memtest86. I usually stick to stock BIOS unless you specifically use XMP. In my experience, this step is often unnecessary with the latest BIOS versions that support Zen and Zen + chips, especially if your clock speed stays under 3200 CL16. I tend to skip it to save time, though I still consider it a solid practice—unless you’re running high-end memory with fast timings.
I disagree with the idea that raising the SOC around 1.2v always improves memory stability. It might actually lead to instability in some cases. Lowering the SOC voltage can sometimes be necessary for better consistency. I encountered this issue when trying to run stable XMP with a 1600 CPU, needing 1.2v for stability. With another CPU and same components, I required 1.2v to achieve stability. This varies by chip, and more isn’t always advantageous.
For my typical 3000/3200 memory kits running XMP on recent BIOS, leaving SOC on auto usually works fine.
I usually begin overclocking with an Intel burn test on the default stress test. It helps me quickly identify the optimal voltage and frequency range for a processor while performing a fast but challenging stress test. If it crashes immediately, I adjust voltage or clock speed. If it becomes unstable, I know I’m close. If it passes, I proceed to the next test. Having overclocked at least 20 different Ryzen CPUs, I’ve found this to be an effective initial “proof of concept.”
Next, I run a P95 26.6 small FFT for at least an hour. If no crashes occur and temperatures stay below 80°C, it’s good enough. If workers or Windows crash, I tweak voltage or clock speed until the test lasts an hour.
After that, I conduct an 8-hour real-world bench with half the system RAM. Repeat the clock/voltage tweaks if needed.
Once all three tests are passed, I run a default memtest86 and fine-tune SOC/DRAM settings as necessary.
While this process works well for quickly achieving stable overclocks, I might explore more extensive long-term testing as you suggested. I usually overclock on CPU SKUs that make sense—like 1200 or 1600 on stock coolers—so I rarely push the limits. Ultimately, software can still detect instability even with default settings, and it’s impossible to cover every scenario.
Regarding the LLC question… You should use it, but proceed with caution. Especially when pushing voltage near the safe limits of the CPU. Always monitor V Droop at full load—if you see significant droop, it’s a sign to adjust. A small FFT is helpful for checking if your LLC is properly tuned. I prefer seeing around 10 mV or more of droop; more if you’re pushing voltage higher.
The issue with LLC is that it can cause voltage spikes during load transitions, forcing the CPU to draw more current. The motherboard compensates by increasing voltage, but this happens instantly when the load changes, leading to brief spikes. This can damage your CPU if not managed carefully.