What is the most effective way to overclock an AMD system?
What is the most effective way to overclock an AMD system?
I completely agree. Prime95 demands a significant amount of processing power. Another point I overlooked is that Prime95 won’t halt just because a core fails; it will display messages like "Worker thread terminated," indicating instability. It’s important to monitor all cores while using Prime95. Another test I frequently perform is OCCT, for example, running it for about five minutes. I believe it relies on Linpack.
However, this is mainly for stability checks and ensuring temperatures remain within safe limits. For actual performance measurements, I conduct benchmarks in my regular activities—playing a few games or using CPU tools like Cinebench. If performance declines, there must be an issue to investigate. It’s likely related to the recent changes you made. Since only a few adjustments were done, it should be clear.
I use Hwinfo64 on a separate monitor to track temperatures and clock speeds. Particularly during PBO tests, I check which of the three PBO values is limiting and what the other two reached. I recall observing a drop from 105 watts to 140 watts on the CPU with minimal performance gain. That didn’t seem worthwhile for me.
During game benchmarks, I run MSI Afterburner to monitor CPU and GPU metrics concisely. I always keep MSI AB and Hwinfo64 open during tests, as they can influence FPS. Having them running consistently should help maintain consistent results. If something feels off in one run, I repeat the benchmark. Perhaps Windows is checking for updates or something similar.
I suspect overheating from OPs cooling might be limiting performance. The 3900x model is quite demanding. It would also be wise to inspect the pasteboard—if it’s dry and hard, it could be clogged and needs replacement every 2-3 years. The fact that Prime95 is losing clocks suggests the cooling system may not be keeping up.
I tested cinebench r23 on loop ten times. The score stayed over 18000 and the clock temperature never exceeded 76c during that run. I also used OCCT, which gave similar results. I adjusted a few parameters and set PBO+AutoOC with an offset of 0.0625, keeping PPT/TDC consistent and EDC on auto. I wasn’t sure whether to enable Fmax, but didn’t change it either. After trying many offsets or disabling it completely, the results were successful.
I ran prime95, OCCT, cinebench loop, and Aida64 for about 30 minutes each last night, and everything stayed up until 3am. I experimented with different settings and ended up achieving 4.275 to 4.3ghz across all tests. The only exception was the Prime test, which experienced thermal throttling at around 87c after 20 minutes. This suggests that the default cooler settings might have limited my performance.
Thanks both of you! You really helped me a lot. I’ve gained a lot of knowledge about Ryzen OC for future builds.
I remember my first build was in 2017 with a 7700K and I started overclocking after reading about it. After about 5 hours of testing, I got a 5ghz result, which probably ruined the experience. So you both found a solution for me. Thanks again! Appreciate your help.
I've come across information regarding an EDC issue. If you enable it to 1, it should be compatible with 3000-series models and possibly earlier BIOS versions. Feel free to check if you're interested.
Regarding the pump speed, I typically operate at full capacity. I can't detect any noise. My fans reach their maximum at 2300 rpm but usually stop around 1400-1500, which isn't very loud. In fact, when playing games, the fans on an Nvidia 2080 tend to be louder than those in radiator fans, especially with headphones on.
However, your situation could differ. Since it's your PC, you have the freedom to adjust settings as you see fit. That's what makes it flexible.
When it comes to overclocking, you can refer to guides and other users' configurations. I do it mainly to establish a range of values for testing. Perhaps PPT between 80 and 200, EDC at 1-150, and voltage around 1.25-1.325v—just numbers I came up with. Each chip behaves differently, so yours will too.
Avoid copying others' settings; they're tailored to their specific chips. My main concern is whether they thoroughly tested the system. Would they return in six months to report any instability issues? I read about this exactly yesterday on Reddit concerning a 3900X. It was an older post, but it's still relevant. These situations do occur.
It's similar to the experience when the Ryzen 5000 series launched, where many hardware reviewers were testing their chips at 1.4-1,425v all-core. Some of those units failed within a week...
Be careful and cross-check with other sources.
It started as EDC=0; fully turning it off. However, pushing it that low (or setting it to 1) didn't help my 3700X—it would just run at a base frequency of 3600Mhz without boosting. Adjusting it precisely is challenging because even a slight increase can make it seem functional for multi-core tasks but not for single-threaded ones. Its performance shifted with newer BIOS versions, AGESA updates, and SMU releases; what worked best was originally around EDC=10, now the optimal range is closer to EDC=5. You also need to disable Advanced C States to prevent it from entering deep sleep at C6. Although the cores consume a bit more average power (possibly 2 or 3W according to HWInfo), I don't see it as an issue. For AMD, the large number of Zen 3 CPUs consuming 2 watts globally can be significant when they're not needed. Ryzen, on the other hand, operates so efficiently even without C6. Ultimately: it's not ideal for everyone. But for those who like fine-tuning, it offers unique enjoyment or, at best, a waste of time.
I've been adjusting CPU settings since the year 2000 while keeping GPU unchanged. Over the years, CPU and GPU tuning has become incredibly quick and straightforward, often taking just a couple of days. Still, it usually takes around a week to confirm stability. My interest shifted to RAM overclocking. I extensively overclocked all three sets of my DDR4 RAM, using various chips. It took me more than a year—possibly two years—to master the process. A significant amount of learning was involved, with many factors affecting performance, some dependent on the motherboard.
Extreme overclockers are valuable sources of information. I continue to gain insights about DDR4 by reading the OC subreddit.
I've also used two 4-pin AM4 motherboards; the older B350 motherboard failed due to excessive flashing. Initially, first-generation Ryzens RAM support was limited, then improved, but after a year, my old RAM overclocking became unstable and eventually required at least twenty BIOS flashes before it would brick.
Currently, I own a 5600X on an X470 platform. While I could attempt per-core overclocks to extract every last megahertz from the CPU, it feels more like a chore than a challenge, offering minimal benefit. I use tools like Corecycler and OCCT, testing each core carefully. My goal is to ensure the CPU remains stable even when idle. I've spent considerable time experimenting with Fclock, searching for a setting that maintains stability above 1900 MHz. Unfortunately, no matter the voltage or configuration, I consistently encounter WHEA errors. The only variable was the frequency per time unit. I've tried between 100 to 1000 different combinations, observing what others have successfully achieved on identical motherboards, RAM types, and chipsets.