Issue with 4 RAM sticks on a Ryzen 5 3600 system
Issue with 4 RAM sticks on a Ryzen 5 3600 system
I possess a PC running Proxmox with the following specs: CPU - AMD Ryzen 5 3600, RAM - 128GB (32x4) A-DATA 32GB at 3200MHz, SSD - Asus ROG Strix B450-F Gaming 500GB NVMe WD Black. Previously, I used a single stick of 32GB at 3200MHz without issues, following the A2 slot recommendation. Recently, I upgraded to three additional sticks of identical part numbers. When testing each stick individually, everything worked fine. After combining all four, I encountered SIGSEGV errors in some VMs/LXC containers. I ran MemTest86 on all four sticks and still faced problems. I eliminated possibilities by testing each stick separately, then combined them. After rebooting and booting Proxmox, kernel panic occurred again only when using all four sticks together. Some VMs experienced a full kernel panic, while others crashed immediately. I became concerned about the RAM capacity. I tested the newly purchased RAM on my main system (Ryzen 9 3950X, Asus ROG Crosshair VIII) and confirmed it functions properly. There were no Windows Blue Screens, even when loading large AI models like Facebook Llama into memory. I reached my server with two sticks and had no issues, but kernel panic appeared only when using four sticks simultaneously. What does this indicate about the 32GB configuration for a Ryzen 5 3600?
128gb works with the 3600, but you might need to reduce performance slightly to lessen pressure on the memory controller. Older Ryzen models often faced issues with 4 sticks.
Adjusted RAM speed from 3200MHz to 2166MHz. Currently, I'm using memtest86 on all four sticks to determine the best RAM clock setting.
I've checked the setup and found the 4 sticks at 2400MHz with 1.35V to be the most stable. At 2666MHz and using memtest86, no errors occurred, though the system would crash during boot despite passing tests. It seems the issue lies with the RAM rather than the hardware itself. You might try raising the +1.0V on the CPU VCCIO/MC pin.
Consider this adjustment unnecessary since they're designed for 1.2V at 3200MHz.
I've attempted to boost VCCIO/MC to 1.25 volts at 3200MHz with specific timing settings for T1 and T2, GearDown enabled or disabled. Despite those adjustments, memtest86 still reports an error: the four stick RAM glitch triggers BIOS corruption. This issue appears harmless because UEFI is using faulty 4x RAM, leading to maximum stable performance at 2400MHz with certain configurations. The T1 operates best at auto DRAM voltage (1.35V), suggesting the RAM stick you purchased might be a different model—possibly a SpecTek unit versus a Samsung one—causing stability problems even though the part numbers match.
Yes, did you get them the same place? If yes then return all and buy a 128gb kit instead