The results of extended stress tests don't indicate any issues.
The results of extended stress tests don't indicate any issues.
Hello, hope everything is fine. I own a 7800X3D built with an ASRock X870 LiveMixer WiFi and Corsair 2x48GB 6400MHz CL32 1.4V. I set the Performance Preset to PBO, Tjmax at 85°C, and Curve Optimizer at -40mV. After 30 minutes of stress testing Google's stressapptest with -W and 1800 seconds, everything seemed okay. Then I ran the full MemTest86 suite (which takes around 9 hours) and noticed some issues: first a single error around the 6.5-hour mark, then no errors for a while, followed by several hundred errors near the end. I turned off Curve Optimizer and re-running the test after 9 hours succeeded. When I reactivated it at -20mV (PBO, Tjmax 85°C, Curve Optimizer -20mV), the test passed again after another 9 hours. I’m glad the CPU never exceeded 73°C most of the time—well below the 85°C limit. In my view, the CPU stayed well within safe limits.
Regarding your questions:
1. Should I worry about that error at 6.5 hours with -40mW if I don’t plan heavy stress for long periods? Data integrity is important, yes.
2. Is it safe to assume the 9-hour test would pass at -20mW? It’s hard to say—maybe it could fail after a few hours.
3. Does disabling Curve Optimizer really make a difference? Yes, and enabling it at different levels affects results.
4. Do you think Curve Optimizer only saves energy or not performance when disabled? It helps both.
5. You’re right to question why the temperature never hit 85°C in those tests. If it did, the throttling would have been obvious.
I’d appreciate your advice on these points. Thanks!
Are you really starting with a weird stress test when your CPU hits ~80/85/95°C? That’s not what a stress test should be about. Prime95 is fine, but it doesn’t actually simulate real-world stress. If errors pop up at certain points, it could mean crashes or data loss (like that SSD failure you mentioned). I’d recommend doing just plain jdec speeds without PBO for now—until you can upgrade to RAM that won’t show issues. These tests aren’t very trustworthy, but they might still point out problems. No errors don’t mean everything’s fine; they could appear later (after 24h, 48h, or even weeks). It makes sense to run a full stock test for a few weeks and watch for any problems like crashes or weird behavior. Yeah, 73 is your safe limit even with default settings (reset CMOS).
I own the 7800X3D at +200 and -30. Mostly CineBench’d it and probably won’t dig deeper. It’s remained stable during all my work and gaming sessions over the past two years. Right now I’m at 74c with 67% usage in StarRupture while sharing this. I believe heavy stress testing and benchmarking can be harmful instead of reassuring users. Like your comment. I’d just let it sit and enjoy it.
I successfully adjusted the settings to -30 on my 5800x3d. No major issues beyond occasional crashes in one game... I decided to run it stock now. Phew, that was a relief! I also tested -25, but it still caused problems. On my laptop, I set it to -20 and haven’t seen any problems so far—thanks!
The results are quite different from what I anticipated. (As the original poster noticed, I also ran tests at -20mW and without any adjustments.) It was unexpected that the device remained stable under heavy stress for six and a half hours with just -40mW, not failing within the first two and a half hours. I had expected -40mW to be unsuitable right away and was surprised by the prolonged performance. This prompted my curiosity about whether such stability at -20mW or even without optimization would be reliable—or if continuous stress testing over weeks wouldn’t provide certainty.
You definitely can't... The best way to be *truly* confident is to drive the machine for a few weeks/months without any regular settings. Reset the CMOS, apply the recommended defaults, and consider setting XMP if needed. I haven’t had a crash in over three years—not once (that’s coincidentally because I use my Samsung B-dies). *Except* when I tried PBO, which was fine. ** Also, one crash happened right after I overclocked the b-dies, but that shouldn’t be a big deal... It felt odd at the time.** What I mean is that none of what you’re asking about really matters or is essential... Just give it a try with overclocking and PBO, but if you run into stability problems, take a step back and think about your goals. It won’t bring huge gains compared to the original specs.
It's a voltage curve that creates scenarios where a single frequency becomes unstable due to insufficient voltage, which won't appear in stress tests focused on maximum clocks. You can't truly guarantee perfect stability of voltage offset, which is why the default curve starts higher. My 5950X performed well under load but would randomly lock up when idle.