Thread about Ryzen 7000 undervolting and PBO curve tuning
Thread about Ryzen 7000 undervolting and PBO curve tuning
Consider testing a second cable connection to the monitor and adjusting settings directly. I’ve done this with both DP and HDMI cables when switching between laptops. Better cooling might indeed make a difference now that we talked about it. It’s not necessary to buy an AIO, but improved airflow could help. What type of motherboard are you using? Have you shared the full specifications before?
I assume cooler temperatures from an AIO would be beneficial. I’m fine with slightly lowering the offset to work with my air cooler, avoiding the need to replace it or fear future damage. However, the repeated errors after setting offsets for more than six cores are making this goal harder. I’m considering that achieving stable higher offsets might become simpler once temperature and power limits are included (such as 85°C and 90 watts), and since those limits are already part of the process, it could save me a lot of trouble. Right now, I just need to find out what offsets the CPU uses at its default configuration before proceeding. The system details: Corsair RMx 1000, Asus ROG Strix X670E-A Gaming, Wi-Fi, G.Skill 64GB, 6000Mhz CL32 DDR5 7700X with Noctua NH-D15 cooler, two Hynix P41 2TB M.2 SSDs. I’ll add the RTX 3080 later once I finish up with my current setup.
Do you think the onboard GPU is adding extra strain that impacts the -PBO? Yes, it would help if temperature and power limits were followed. I’d approach it similarly, enjoying the challenge of discovering your boundaries.
This was a promising suggestion! I transferred the HDMI cable from my monitor to my SNESC and connected it to my new PC, while the DP cable went into my older system. It’s still unclear, but this seems like a solid approach. Great news regarding my worries about unstable performance—testing of the 4th core at -30 was successful, while the 5th core caused a BSOD at -30 and then crashed at -27, but the 5th core finished its 80-minute Cinebench run without failure. This suggests the 4th-core outcome is dependable as well. I’m now checking the 6th core at -30, which has remained stable for about 20 minutes so far. In earlier tests, the 6th, 7th, and 8th cores were all tested at -30, -25, and -30 respectively. If the 6th core stays steady at -30 until the end of its testing period, I expect to see a different result in the 7th or 8th core compared to my previous findings—helping me identify any patterns behind the recurring error messages during my last batch of tests.
Additional checks showed the 7th core was shaky at -25 degrees. An offset of -26 triggered a crash, then moving it down to -24 cleared it for 80 minutes of Cinebench. However, during another 80-minute run it failed in the final five minutes. Going to -23 caused a crash too. Recently I finished two hours of Cinebench at -22. It seems my earlier tests likely included errors for the 5th and 6th cores (adjusted after setting the 7th) because the 7th was unstable only in the last five minutes of the session. Since I noticed the 7th core remained problematic just before the end, I’ll focus more on the 5th core with its -25 offset for better stability. The 7th core remains the weakest CPU component. With a -30 offset it almost never crashed before I entered my Windows login, but handling the subsequent BSODs was difficult.
I faced a challenge: tested each core separately for at least two hours, all passing at specific offsets. Then I ran the full CPU package at those points but encountered a BSOD after about one and a half hours. After further testing over three more hours, everything passed again. This suggests instability isn't fully apparent in isolated tests. A video at 4:47 might help pinpoint unstable cores. However, when I tried to set a custom view profile for WHEA errors, no results appeared even after multiple reboots. I plan to use the CoreCycler program he recommended.
It's interesting. I'm in agreement with your reasoning and testing. Stay updated on the Core Cycler. I'll also look into it.
CoreCycler executed for five cycles before my PC stopped working unexpectedly. There was no BSOD or warning screen—just a black screen that shifted between dark and gray tones, and the mouse RGB turned off. After waiting a few minutes, I pressed the reset button but it didn’t help, so I powered down at the PSU switch. Once back on, Windows booted up. I checked Event Viewer for WHEA errors, but none appeared. A quick search revealed others had similar issues without clear explanations. The logs mentioned a crash during testing of core 5, with offsets ranging from -30 to -29 and a test duration of about six minutes. After restarting, the system logged activity showing the selected core was 5, and the last entries indicated successful completion. I adjusted the offset for core 5 by one and retried, but the process remained unresolved.
In reality, the central 5 in CoreCycler should correspond to the 6th CPU core, since it starts counting from 0 to 7. However, Event Viewer indicates a shutdown at 10:05:55, suggesting CoreCycler was testing core 7 at that moment—likely the 8th core, which is marked as -29. Despite this, a failure occurred, yet CC reported core 7 (8) passed its test and continued. Windows Event Viewer noted the system shutdown, but CoreCycler proceeded with the next core, possibly bypassing the log entry for that failure. At 10:05:48, new log entries appeared, starting at line 743, confirming successful self-test. By 10:06:00, CPU usage dropped to 6.28%, frequency stabilized around 5492 MHz, and the stress test resumed without further issues. The latest logs showed no size changes and a final check at 10:06:30 confirmed stable performance with available cores listed as 5, 4, 0, 1, 3.