XMP becomes unstable post-installation of 3080
XMP becomes unstable post-installation of 3080
Hello! It's taken me some time to pinpoint this problem, but I'm glad you reached out for assistance. Here are the details: OS: Windows 10 Pro 20H2, RAM: 19 GB, CPU: AMD 3700X, GPU: EVGA 3080 XC3 Black Ultra, Storage: 1 TB SSD, Power Supply: Corsair HX 750W (Platinum). I originally built this system in November 2019 with a GTX 1060 and waited for the 3000 series. It has been stable and great. In December 2020, I upgraded to the 3080, but recently some games have started crashing without any error messages—just a CTD. The reliability monitor would say "APPCRASH" or "Hardware error," but after stress testing everything, it seems the crashes only happen when using XMP profiles with the 3080. I've also tried disabling XMP and the issue disappeared. I ran benchmarks and stress tests on all components, including Memtest86 and OCCT, which passed without issues. It looks like my RAM is fine, but I'm unsure how to get it back up to spec. Any tips would be greatly appreciated!
Consider adjusting the DRAM and SoC voltages slightly. The GPU's current spikes might cause voltage instability. Have you possibly updated your BIOS for ResBAR?
Enable XMP and adjust the dram frequency to 3200mt/s. With 4x8GB storage, performance should improve noticeably.
I'll give it a shot! What does "a little bit" mean in this context? It's completely unfamiliar territory for me—never worked with RAM speed, timings, or voltage before. The problem started in December of last year, and I've kept updating my drivers, BIOS, and GPU firmware to the newest versions without success. Right now, ReBar is enabled. Thanks! I'll give it another try!
What is your SoC voltage? It should be around 1.1 for 3600Mhz memory, usually 1.35. Try increasing the RAM voltage to 1.36 if that doesn’t bring the SoC up to 1.11. A DRAM calculator can also help you find the right speeds. You might also set your XMP Infinity Fabric manually to match the 1800 Memory 3600 timings, then adjust the SoC and DRAM settings while keeping the rest automatic. RAM performance can vary between different chips—like your Patriot Viper running at full speed on B550 Pro4 but not on X570 Strix. Your Ballistix sticks work well on B550 but not on X570. Both single-rank timings are similar, but differences in die type (Micron vs. Hynx) can affect results. Be ready to reset your BIOS about 200 times and spend a few hours; you might achieve speeds above the standard.
It's likely the 3080 generates a significant amount of excess heat, which could raise temperatures of nearby parts. Your stable RAM at 45-50°C might start to behave unpredictably at around 60°C or higher given the 340+ watts it draws. Samsung chips tend to be more heat-sensitive than other types. If previous fixes don't help, consider introducing airflow over the RAM module.
Here’s a revised version of your update:
I attempted to lower the speed to 3200Mhz, which initially helped. However, the problem reappeared. Increasing the SOC and memory voltages seemed to resolve the issue (even at 3600Mhz). My main concern is that the RAM is set to operate at 1.36V, but hwinfo indicates it’s actually running between 1.38 and 1.39V. I understand that this range is within safe limits, but I’m curious about the difference. The RAM chips are manufactured by Hynix, which I only discovered after researching. It performs reliably at around 55–58°C under load.
It seems the main issue lies with the Samsung chips. They tend to behave poorly when temperatures rise above 50-60°C.