Question Overclock 9700K - Degraded?
Question Overclock 9700K - Degraded?
Hi gurus.
Here are the details:
I’m using the following components:
i7 9700K, NZXT Kraken X72, Asus Rog Strix Z390-F GAMING, Asus Rog Strix RTX2070 O8G GAMING, Corsair Vengeance RGB 16GB (2x8GB DDR4 3000MHz), Samsung 970 EVO Plus MZ-V7S500BW 500GB NVMe M.2, Toshiba P300 3.5" 2TB SATA 7200rpm, Corsair RM750x, NZXT H series H700i Window Black.
I’ve noticed some instability when overclocking at 5GHz. I thought starting fresh would help, but updating the BIOS reset everything again. Now I’m having trouble reaching the same stable overclocks as before. I’m concerned my CPU might be failing.
The settings I was using were:
Multiplier: 50
XMP: Enabled
AVX: Auto
MCE: Auto
LLC: 5
Voltage: 1.385v
VCORE @ load: 1.350v
VCCIO: 1.250v
VCCSA: 1.150v
I successfully passed 8 hours on Realbench and 2 hours on Prime95 26.6, Intel Burn Test & OCCT Linpack NonAVX.
Now I’m struggling with 2 hours of Realbench and 15 minutes on P95.
My temperatures have stayed normal (50-65°C) during gaming, so I’m not sure this is the issue.
Could there be something wrong with my setup or process?
Thanks in advance.
Best regards.
New BIOS updates may influence overclocking adjustments. It's important to reset to the default core configuration and verify that the RAM is functioning correctly. After confirming the RAM stability, proceed with testing using aida64. Once the stock configuration appears normal, you can resume overclocking the CPU.
Updating BIOS can require a complete restart. This is why it's advisable not to update unless there are significant problems.
The problem with your initial stress test was that you didn't verify the RAM using memtest at its maximum speed. As frequencies rise, the IMC may become unstable. For instance, on my Ryzen system, I adjusted the RAM timings and overclocked it to 3800. This worked well at stock core speeds, but once I enabled PBO EDC bug, the entire RAM overclock became unreliable.
I reverted to the stock BIOS and ran memtest repeatedly. Each run produced inconsistent results—some passed, others failed. I discovered that I had left the Termination block and CAD_BUS block timings set automatically by the motherboard, matching the DRAM Calculator settings for Ryzen 1.6.2. Occasionally, the system would be stable when powered on, but other times the RAM would fail the test with errors.
Extensive testing was required to identify stable configurations, ensuring consistent performance every time I booted my PC. Eventually, I achieved tighter memory timings and even reached higher frequencies at lower vcore values. I initially believed the memory controller or cores were faulty, but it turned out the issue stemmed from incorrect settings in the DRAM Calculator for Ryzen 1.6.2. The latest DRAM Calculator for Ryzen now aligns with my adjustments.
Thank you for your response.
My system was already unstable, so I chose to update the BIOS, allowing me to begin with a brand-new BIOS version.
Indeed, my first overclock about ten months ago did not involve testing my memory, at least not in a way that would consider it stable.
But recently I encountered instability again, so I turned off the overclock and checked my RAM with and without XMP enabled—using tools like Memtest86, HCI Memtest, and GSAT. All tests showed no issues. I haven’t yet combined RAM testing with CPU overclocking.
I think PBO EDC is similar to XMP for AMD systems?
During my late-night tests I tried various settings:
Multiplier: 50
XMP: Enabled
AVX: Auto
MCE: Auto
LLC: 5
Manual voltage: 1.390V
VCORE @ load: 1.350V-1.359V
VCCIO: 1.250V
VCCSA: 1.150V
For thermal management: Prime95 v26.6 completed in 15 minutes – Passed
Max temperature: 97°C
Regarding stability: Realbench v2.56 ran for 8 hours – Passed
Max temperature: 92°C
If I increase the VCore by 0.005V, it should pass without any WHEA or BSOD.
What would you recommend moving forward?
Testing my RAM with CPU overclock + XMP?
Or adjusting timings myself, like you did?
UPDATE: I just ran an hour of GSAT using CPU overclock + XMP with no errors.
I’m currently using GSAT because I’ve heard it’s more strict and faster at detecting faulty RAM than Memtest.
Thanks.
PBO EDC serves mainly as a way to auto overclock AMD CPU models. With my 2500k unit I had to raise the core voltage after some time. Everything appears fine so far. The silicon lottery consumes too much power—about 1.387V on the core. [1] [2] If you're concerned about long-term effects, it's wise to reduce the voltage slightly. A 1.39 vcore seems acceptable now. [1] It might have been that your overclock was close to stability (stable enough for initial tests but still a bit shaky), and a small boost in vcore helped it settle. I hope this will continue to work smoothly from now on. For testing, I use Aida64, prime95, and memtest. I wouldn't raise VCCIO beyond 1.250v continuously. [3] [1] https://forums. [2] https://siliconlottery.com/pages/statistics [3] https://www.tweaktown.com/guides/8481/co...ndex5.html
Thank you for the update. As you can observe from my original maximum temperatures, my fan curve in the system was severely misaligned, with some fans operating only at 40% while the CPU reached 97°C. I recalibrated the fan curves and conducted further tests.
With a 5GHz speed of 1.390v (1.350v under load), I successfully completed two 8-hour realbench sessions. The maximum temperatures in both realbench and p95 small FTTs were 87°C. This represents a -10°C reduction under load, which was quite impressive.
I hadn’t realized that thermal gradients could also lead to instability. Initially, I assumed only thermal throttling would occur, not the broader instability issues. This led me to consider whether lower voltages might help with stability thanks to improved cooling.
Additionally, I adjusted VCCIO to 1.20V and VCCSA to 1.15V—no errors were reported in GSAT. However, it didn’t pass the IBT test when VCIO/Sa were lowered further. Not sure what to do about that situation.
Also, is there a free version of Aida64 available?
Thank you for the link. I've since fine-tuned my fan curve, which has significantly improved performance. It appears that excessive heat was affecting system stability even when operating at lower voltages. By managing the thermal output more effectively, I've been able to reduce my voltage while maintaining the same speed (5GHz). I've run tests with Prime95 small ffts (15 minutes), 12-12 in-place (1 hour), and Realbench (8 hours). The results show stable temperatures across various scenarios. In the p95 test, a voltage of 1.360v passed both tests, though it failed Realbench after 6.5 hours with an error message. I've since increased the voltage to 1.360v and will continue testing. Once successful, I plan to verify RAM stability using GSAT, TESTMEM5, and Aida64 Extreme. I'm surprised how much better my temperatures have dropped, which has greatly improved system reliability. I hadn't realized that heat could destabilize the system. My initial thought was to conduct thermal tests to ensure compliance with tjmax limits and enable longer stability assessments without risking hardware damage. It seems lowering the CPU temperature from 97°C to around 85°C has enhanced stability at reduced voltages.
heat influences not just the CPU and DRAM stability but also other parts such as VRMs, which further affect system reliability. I invested a lot of effort trying to pinpoint instability in my overclocked components by adjusting BIOS settings, only to discover that the real issue was heat management around the VRMs and DRAM, resolved by placing a fan directly on them. Even pushing the PC to its limits in a different room with less effective cooling can disrupt stability. RAM timing is particularly vulnerable to temperature changes.
I recently secured a typical 120mm case fan within the enclosure directly above the DIMMs, ensuring continuous airflow.