Check if the Ryzen Curve Optimizer delivers value for your system.
Check if the Ryzen Curve Optimizer delivers value for your system.
Has anyone tried adjusting the Per Core feature for the Curve Optimizer to fine-tune precise undervolts per core? Are there any methods to assess stability or decide which cores to modify? I lowered my negative offset from 28 to 26 after two crashes showed my 5900X struggled at low loads. This seems like a typical issue for Ryzen during undervolting, though it's tough to test directly—only by running the system for days and watching for crashes. After a week without problems, I believe -26 offers solid stability for my setup. Although I thought this was a reasonable outcome, I considered the Per Core option. Even if only one core works at -26, others might handle lower values better, potentially restoring performance. The challenge is the sheer number of settings—12 cores with 5 offsets (26 to 30) creates nearly 250 million possible combinations. If verifying each takes a week, it could take almost five million years to find the best one. I’m starting to think that testing all options might not be practical. But if anyone has useful advice, it would help me optimize before things get serious.
Begin with core 0 at -30 steps. Launch Windows and perform a CPU benchmark in OCCT, focusing on large AVX2 usage for just core 0 (advanced mode, then disable others). Run the test for 15 minutes. If errors appear, lower the step count by 5 and retest. Keep adjusting until errors resume. Proceed to increase steps gradually, checking after each change, until you reach the last working value. Repeat this process for the remaining cores one at a time. The process may take time, but it should fit into an afternoon, not weeks.
Lol... I don't worry about that since Windows doesn't use the CPU that way. I just adjusted my limits and everything works except for core cycler stuff. It always fails when things go wrong. Boost varies, but the worst is 4500 with Linpack Extreme. F@H and WCG run at 4650, OCCT large, extreme runs hit 4750, TM5 at 4850, and 3DMark usually between 5000-5150. Superpi 32M hits 5150. If you use an Asus board and enable APE, it helps a bit. The biggest issue is your own limits, not the motherboard's restrictions.
This test checks how well the system handles minimal stress. It resets the computer completely under different conditions—while idle and while you were browsing—to see if it remains stable.
Uncertain about what you mean. There’s no such thing as "low load" or "high load" stability. If you had an all-core undervolt setting, likely some cores weren’t steady. When one of those cores dropped out, it could either work fine or cause problems—this is instability. For instance, I started with -30 for all cores but experienced random issues. It turned out a few cores needed to be brought down to -28 or -27. Most handled the full -30 fine. After getting everything stable, performance has been excellent.
Various software options can lead to instability at specific stages. For instance, adjusting GPU voltage or pushing overclock settings often coincides with running Heaven simultaneously. If the application crashes, it usually indicates I pushed too far. Even if tools like Heaven, Kombustor, and other benchmarks function correctly, stability remains uncertain. My main benchmark for GPU reliability is Control, as it tends to fail when others succeed. With Ryzen, the problem appears linked to its idle handling—though I’m unsure of the exact term. I opted for “low load” since only a few cores are active, which seems to trigger the most voltage reduction from Curve Optimizer compared to default settings. My CPU performed well on Cinebench, Intel Burn Test, gaming, and OCCT across all cores. However, it crashed even when I wasn’t present. Testing one core at a time might offer a more accurate low-load simulation.
Recognized, but that's exactly what I meant by testing. In OCCT you'll notice each choice targets the option most likely to reveal issues (large, variable, AVX2). You're also evaluating performance on a single core with both threads active, eliminating random timing variations seen in full-core tests. Essentially, it's designed to catch even minor instability. Based on my experience, if you can complete this test for 15 minutes on each core without errors, you'll confirm full stability.
It seems there were issues early on, so even an eight-day uptime under different conditions didn’t fully validate the results. I encountered errors in Core 0, 1, and 2 within just a few minutes. This is surprising given that Core 1 and 2 were highlighted as top performers in Ryzen Master, yet Core 3 appears to be holding up well so far. My CPU might not be performing as expected for the 5900X, especially since it’s already running at -26°C and showing faults. It does suggest that CCD1 could be the culprit—it tends to run slightly higher than ideal. I’m confident most cores on CCD2 will work fine and should reach around 30°C.
A useful suggestion is to run everything in Windows Safe Mode—it helps with the lighter performance conditions we discussed. Testing one core at a time is essential for achieving full stability with Curve Optimizer; I used a 5700X that stayed solid even at -30°C, but it would freeze during games. This happens because under heavy loads, the boost frequency drops, making weaker cores more vulnerable.
We evaluate each core at its peak boosting frequency while applying maximum load. This ensures conditions similar to high-frequency operation with minimal load. If the core supports AVX2 at its highest frequency, it also enables efficient web browsing at full speed.