Is the NB frequency identical to the FCLK?
Is the NB frequency identical to the FCLK?
In BIOS you adjusted the FCLK clock to 1867, while CPU-Z shows the NB clock as 1599.6, matching your RAM speed. This discrepancy is being considered to better understand the cause of your OC instability by increasing the FCLK above the RAM speed and then checking if that resolves the issue.
Pushing FCLK above ram frequency is a poor method to assess ram stability. Whether FCLK remains steady has no connection to ram speed. Typically, the maximum achievable FCLK depends on the CPU's memory controller quality, which is largely determined by manufacturing variation. This has nothing to do with how far ram can be increased, and most chips reach around 1900Mhz, while some support up to 2500Mhz. Simply apply higher ram speeds right away—if it fails, stability isn't the issue.
I'm trying to check my CPU's binning by increasing the FCLK while keeping memory OC off. Sorry if I was confusing myself earlier. I understand my system is unstable at 1867 FCLK with 3773 RAM, but I'm looking for the cause. It works at those frequencies but fails during stress tests.
It's unlikely it's FCLK at that stage. If it's unstable, adjust voltages and timings first—they're more likely to cause instability than FCLK. For regular use, 3600Mhz with tighter settings should work better on any Ryzen chip. NB UCLK matters, not FCLK; UCLK matches memory frequency.
To check if the overclock will work, keep FCLK at its default setting and increase RAM first. Then adjust FCLK to match once stability is confirmed.
Also what are the chances of ram overheating causing a system crash in memtest86? I'm curious about pushing my ram to CAS 16 3733 at 1.45V to test its limits if higher voltage could help, and if a crash occurs, how would you tell it wasn't just a temperature issue?
Boost FCLK until it matches MCLK consistently. Maintain this value without deviation unless you're operating with very high-end RAM above 4.2GHz and your FCLK isn't stable at that frequency. RAM typically doesn't overheat much. Under normal conditions, heat generation comes mainly from the CPU and some GPU components, which can cause minor temperature rises. 1.45V is adequate for regular use, but avoid exceeding it unless you're targeting performance beyond 2100Mhz. Monitoring temperatures is essential for stability verification. Perform all 11 tests in memtest86 across four passes. While rare, temperature-related crashes in memtest86 aren't uncommon, though they can occur. Even if it passes, it might still be unstable since stress testing raises temperatures. After passing, apply a CPU stress test using Prime95 to further validate stability. I follow a specific configuration for these checks.
I'm running version 286 because it supports Ryzen well and provides suitable FFT sizes. First, turn off AVX by editing local.txt and adding "CpuSupportsAVX=0". Enabling AVX can cause unrealistic temperatures. For CPU stress testing, use p95 with a blend profile, custom settings, min FFT size 512K, max 4096K, memory usage around 75%, and run each FFT for 15 minutes. Execute this for 4 to 8 hours. If all threads remain functional, your setup is stable—provided memory issues aren't present. This process is quite time-consuming. For routine checks, I recommend relying solely on memtest86 until you finalize your daily testing parameters, then proceed with the full suite.
So I have a stable overclock, with my CPU running at 89°C during heavy Prime 95 load and my RAM at around 40°C. Is this okay? The max temperature for the 3600 is 95°C, so I'm under that limit, though it still feels quite high.
It's quite high, and lower temperatures would be better, though it shouldn't cause immediate harm. Over time, it might speed up the chip's wear, but it should still remain functional for more than five years. Are you adjusting the CPU temperature? The RAM temperature is likely acceptable, but stability is important. If you aim to reduce both, consider increasing the fan speeds in the case or reducing the CPU temperature if you've already optimized it, as this can also affect RAM temps. Running a benchmark such as unigine superposition or 3DMark Timespy can help confirm that the GPU isn't overheating the RAM.