Stock configurations cause instability for 7900X.
Stock configurations cause instability for 7900X.
Just to confirm, were you testing at 1.15V SOC or 1.3V SOC? You should use 1.3V (the safe maximum) because memory clock adjusts based on SOC voltage. It’s not a huge difference, but it could still matter depending on the situation. From what I know, the Ryzen 7000 series doesn’t actually support running tRAS at 28; it might show 28 but is likely running at 30. That’s not a big deal since tRAS isn’t a major performance factor. The rest of your primary timings seem fine, you might even lower them slightly with higher voltage, though the gap between CL28 and CL30 isn’t significant. The only areas with some flexibility are tWTR_L, tRTP, tRRD_L, and possibly tRC. You’ll likely get only a couple ticks from tWTR_L (negligible), tRTP can vary by kit, tRRD_L has minimal impact especially below 8, and tRC timing can cause errors late in testing, so it’s probably not worth optimizing. Personally, I’d stop there—especially since ASRock’s latest BIOS enables 2:1 mode, making DDR5 8000 viable on their 4 DIMM boards. You might want to adjust settings once MSI updates the BIOS, which could happen soon. With such low SOC voltages, you should be able to reach around 2167MHz or even higher, though 2133 MHz might surprise you.
It was at 1.15v, as you mentioned it doesn’t really matter much—perhaps we can go back once the BIOS update from MSI is confirmed stable. Should we aim to increase the SOC to 1.3 for higher FCLK clocks, or is a SOC of 1.15v sufficient? Also, running y-Cruncher VST for an hour plus memtestpro at 2000% seems to guarantee memory stability before proceeding to CPU curve optimizer and PPT limits.
At around 1.15V you can push the FCLK quite high. It seems FCLK adjusts based on SOC voltage in reverse. With 1.3V SOC you probably reach about 2100, but at 1.15V it might hit 2167 or more. From what I've observed with that BIOS update (which was released recently), the actual SOC voltage has little impact, so you should be able to achieve around 8000 with just 1.2V SOC. More testing will clarify, but a performance of 7200 with very low SOC and fast FCLK could be ideal. The stock 2GHz FCLK in 1.0.0.7b AGESA seems to work well, especially in 1:1 mode for best results. It's still uncertain, but it looks promising. Keep an eye on the 1.0.0.7b AGESA update, as it fixed the 2:1 mode issue on ASRock boards. Definitely worth checking. A 1000% boost is usually sufficient, though 2000% would also be fine.
I managed to achieve 2167 with about 1.5 hours of y-cruncher VST stable work. The 2200 won't be viable. Edit: Talked too early—just got a random blue screen now. Still unsure what triggered it. Could be the FCLK=2167 setting or the memory profile. All memory tests passed with MemTest86 1500% and y-cruncher VST 1 hour twice. Moving on to the curve optimizer tuning. I could hit -40 on some cores and it would pass CoreCycler Prime95 easily, but the system would randomly crash while idle, even when the screen is off. Sometimes it won't boot in Windows. I've tried -25 across all cores without any crashes yet. Though I suspect it might happen eventually. Any advice on tests or durations that would reveal errors early during curve optimizer testing? What would count as a stable run and when should I expect it to fail?
I've experienced unusual instability issues with the FCLK, so reducing it to 2133 might help. After an hour of using VST for FCLK, it should be stable enough to confirm. Based on your stress tests, other factors like the curve optimizer or faulty software could be responsible for the BSOD. I'm not very experienced with the curve optimizer, so I'm focusing on standard tests such as CoreCycler and Prime95. While idle it can crash, I usually wait for a blue screen before adjusting the undervolt by 5 to try again.
I believed outdated graphics drivers or auto-installed NVIDIA versions could be the cause. I updated them and lowered FCLK to 2133, achieving stability after two days even with the -15 all-core curve optimizer. I rechecked 2167 and it crashed during the first minute of y-cruncher. It appears y-cruncher struggles to detect FCLK issues due to insufficient clarity. Once it passes an hour of VST testing, it crashes again within the first minute. I have two more points: the AMD 7000 series seems to consume more power during idle compared to my old Intel CPU. Is there a method to lower idle power usage? Also, do you run an AIO? If so, could you share a screenshot of the fluid temperature during Cinebench startup from idle? My reason for asking is that I've refurbished my 10-12 year-old Corsair H100i with fresh fluid, cleaned it, and lightly sanded the CPU copper block. I don't recall much about new AIOs yet. The fluid temperature stays around +5 to +6°C above ambient during idle. Once Cinebench starts, the CPU temperature rises to 95°C instantly—no questions about its thermal mass. The fluid temperature gradually matches, reaching about +13°C above ambient while the CPU runs hot at 95°C. Temperature1 is the fluid temperature. When Cinebench stops, the temperature drops slowly.
FCLK instability is hard to detect in general, but Y-Cruncher stands out as the most reliable. It’s the top choice among less-than-perfect options. HCI Memtest also catches it, though usually after a significant load—around 400-500%, which means several hours of stress testing before an error occurs. Personally, when I was tuning the chip, I would stress test the FCLK and then lower it to verify stability. That’s not really how it works technically; AMD processors tend to wake up quickly from idle, even with a tiny voltage change (compared to Intel, which has stricter idle thresholds). If anything is running, it won’t stay idle properly. As far as I understand, there’s no setting to adjust this, it’s just a characteristic of AMD CPUs. You might try tweaking Windows power plans, but I haven’t found any success. I’ve used AIO coolers, but none of them work in my main system, and the ones I own don’t function either. My setup uses custom water cooling with no software-readable temperature control. The problem with the Ryzen 7000 isn’t its cooling capacity, but its inability to dissipate heat effectively—regardless of whether you’re using a custom loop or a Wraith Prism, it will still reach 95°C during testing, which affects performance scores only slightly.