Prime95 large FFT instability
Prime95 large FFT instability
Hello, another P95 issue has arisen. I’m running a fresh build with an 8700k at stock configurations (Enhanced Turbo off) on an MSI Meg Ace MB, 32 GB (2x16) NVMe 3200 with XMP enabled, paired with a Noctua D15 cooler. The power supply uses Seasonic Focus Plus Platinum 750. The GPU is an older MSI 1060. All other components are new except the graphics card. The system passes all tests except P95 blends or large FFTs. The same outcome appears when using XMP or default speed settings. I’ve even adjusted the CPU clock and disabled turbo completely, yet it still works.
I frequently run memtest86 with sequential CPU enabled, and it always passes without errors—only two failures every so often, both at the same addresses, which consistently reappear. When P95 fails, it’s a rounding error, and only one of the 12 cores stops working, while the rest continue normally. Temperatures typically stay in the high 50s, never exceeding mid-70s during any p95 tests. The failures happen randomly across cores and aren’t tied to specific conditions.
A large FFT passes when I restrict memory usage to 25000 instead of the default. That test has been stable for a few hours now.
Before attempting an RMA, is this more related to RAM, CPU, or MB? Are there other steps I can take to pinpoint the cause?
If it’s important, I’ve performed some overclocking experiments on this machine without major issues, except for the P95 failures.
Here’s the full part list:
https://pcpartpicker.com/list/JhvxD2
The only positive result is IA: Max Turbo Limit.
There seems to be no throttling either, as the ratio stays steady at 43 throughout the entire test.
Last night I tried adjusting the custom large FFT sizes (from 146 to 8192) with varying RAM amounts, aiming to find the point of failure. It never failed, even when I increased RAM back up to its maximum (29698).
I performed a blend test lasting six hours without any errors—always failing within about 90 minutes.
The only possible fix I considered was reloading the Default profile from Intel Extreme Tuning. After rebooting and running another blend test, it still failed. I then tried reapplying the Default profile, but it refused to apply because it assumed no change had occurred. So I changed it and reloaded Default to try again. The test failed once more after an hour.
I’m puzzled as to why all those tests passed until I rebooted.
Reduce memory to a single stick of RAM and verify performance by alternating each one.
I’d also look into recommended or approved timing specifications for Vengeance LPX at various RAM voltages, since not all '3200 MHz' settings are the same—some could actually benefit from lower timings like 2933 or 3000 MHz.
On Friday night, I conducted realbench stress tests for several hours and all passed. However, in HWiNFO64, an extra "performance limit reason" appears (IA: Turbo Attenuation (MCT)). I don’t know what that means. Turning off the turbo in the BIOS resolves it, but restarting it to "Yes" doesn’t. Power cycling didn’t clear this issue either. Apart from that, the machine’s behavior remains unchanged.
There would be no need to turn off the standard turbo options... since you invested in a 4.7 GHz processor, I’d definitely prefer it.
I’m not certain if the 'Enhanced Turbo' you mention is just a rebranding of the MCE function or if it’s the regular Turbo setting.
If all cores reach 4.7 GHz when under load, the MCE feature will be active, regardless of the term your motherboard manufacturer uses for it.
In any case, Prime 95 blended mode is more RAM-demanding...
Try testing with a single stick, checking each one individually.
I’d set your BIOS to default so that standard turbo remains enabled, which should let a single core reach 4.7 GHz—unless I suspect that running on normal speeds is already causing Prime95 problems.
Re-test stability for Prime95 at the default CPU configuration, using lower RAM speeds (such as 2666 MHz).
If your motherboard supports MCE, you can enable it, but expect a VCORE upgrade and the additional 300-400 MHz across all six cores will likely increase temperatures.
I assumed we were troubleshooting P95 issues...?
If you’re only focused on gaming stability or Realbench performance and are content with the current setup, that’s your decision.
Thank you for your assistance. My motherboard (MSI) labels it as Enhanced Turbo, but it is actually turned off. Typically, after booting, most cores remain around 4.3Ghz and then fluctuate based on system demand—ranging from about 800Mhz to roughly 4.5Ghz. I seldom reach 4.7Ghz. I am certain MCE is also disabled. Something unusual I observe is that I rarely see it hit 4.7Ghz either.
I plan to adjust this over the next few days. I performed a week ago and believe each test run passed both overnight memtest86 and Prime95 without failure, though I will recheck to confirm stability.
As mentioned in your post, P95 fails even at the standard non-XMP frequency of 2133MHz, using the latest BIOS and default settings—except for "Enhanced Turbo," which is disabled and defaults to "Auto."
I was running p95 to verify that all my new parts are working properly, so I could experiment with overclocking. For clarity, I used realbench to understand why p95 is failing, but the results were confusing. It wasn’t until after running realbench that the "Turbo Attenuation (MCT)" flag appeared in HWiNFO86—before that, the only positive entry was for "Max Turbo Limit."
Thanks!
Check if you're relying on Intel XTU for profile tweaks. The reason behind it remains unclear.