Thread about Ryzen 7000 undervolting and PBO curve tuning
Thread about Ryzen 7000 undervolting and PBO curve tuning
I kept tweaking the per-core offsets for my CPU. Yesterday everything seemed fine, but today the same settings caused problems during testing—errors popped up even though the system didn’t crash or freeze. During Cinebench runs, I encountered multiple errors at those offset levels. I opened Ryzen Master to check what it had assigned. It showed all cores were set to -30 offset. It looks like AMD might still have some issues with the optimizer profile, since even lower offsets (like -26 or -28) led to instability. I plan to keep adjusting the settings manually and re-testing until I find a stable configuration.
Good post, thanks! I'll check more details tonight. So far I've set PBO to -30 in BIOS and everything seems stable. My system has 6000MHz RAM with EXPO enabled—uncertain if it affects performance much. In Cinebench times dropped to the low 80s, wattage stayed high in the 80s, and the score is around 20,800. I plan to run tonight and refine the numbers. Any suggestions for a cooler or case? PBO should be fine with cooling, but it's worth mentioning.
I'm using a Noctua NH-D15 cooler with a Be Quiet Silent Base 802 case. During Cinebench runs with fans up and front panel off, I've seen temperatures reach 93-95°C. The 6000 MHz EXPO profile is active for my RAM. I'm curious if heat could be impacting the stability of the offset. It's possible that earlier stability came from lower values, but today I'm getting errors despite adjustments. Raising negative voltage curves should help with cooling and stability. In tests I've passed Cinebench with cores 1-3 at -30°C and others at -20°C, but not at -20. I plan to try higher offsets after confirming those settings work. Could I possibly increase the negative offset further for better stability, setting limits first before adjusting PBO?
Up to 95°C seems possible with continued testing. An AIO would perform better. I encountered a Cinebench error during testing of the 7th core set at -27, while 1, 2, 3 and 8 remained stable at -30, which appears consistent. The 7th core at -30 triggered a BSOD. I plan to test at -25 next. Once I determine the offsets, I'll configure temperature and power limits. After that, I expect the maximum load to stay under 85°C, possibly even lower. Then I'll attempt to adjust the offsets beyond -30.
Last test ran at 5.34GHz, 87°C, 85 watts. Didn’t run a serial test on the 5.45 bench, planning to recalibrate it now.
Cool. Your setup looks fine so far. I adjusted my core offsets to -30, -30, -30, -20 (still not fine-tuned, just moved them away), and -25, -30, -25, -30. Early results were promising. But then I encountered issues in Cinebench. During more testing, errors appeared every 3 to 5 minutes during longer runs. Even after trying different cores, occasional problems persisted. Despite this, I managed over an hour in Prime 95 and an hour in OCCT without trouble. It seems Cinebench is having trouble with these offsets. Now I plan to test one core at a time, running 80-minute Cinebench tests for each offset, hoping to pinpoint the problematic core. I wish Cinebench supported selecting a specific core for testing, so I wouldn’t need to tweak offsets or run full multi-core tests repeatedly. The same applies to OCCT (I haven’t found that option in the free version). In my last pass, I offset one core and waited 30–50 minutes. If it was stable, I offset another. I continued this until I reached seven cores. Then I started seeing occasional errors even after reducing offsets further. The first three cores have been tested more extensively and appear reliable at -30. Currently, I’m running those three together at -30 in a 90-minute Cinebench test—still no errors so far.
I executed it and all cores dropped to -30, even though it wasn’t reliable. Using that setup, Windows would freeze under light load. After entering my credentials and pressing enter, the system would crash immediately. Another problem is that Cinebench becomes unresponsive, especially when I disconnect the monitor from the 7700X. I usually unplug it to run another PC, but this stops testing altogether. The visual output halts briefly after removing the monitor from the DP port, though processing might keep going for a short while—sometimes finishing the test within minutes. Once the monitor is unplugged, CPU fans slow down and then restart after about 20 seconds when I reconnect it. The fourth core remains stable at -30, but the visual freeze happens twice after disconnecting, even though processing seems to finish. I’m retrying with the monitor back on now, since the test has lasted 45 minutes without issues so far. I suspect previous instability came from mismatched offsets, but now I’m testing with more cores (-30) and still facing problems. It’s possible individual cores handle -30 better when others are less offset, though they tend to lose stability as more are pushed to extreme negatives. The recent test where AMD Ryzen Master checked all cores at -30 and set them to that setting also failed in practice, which might explain the inconsistency.