Checking performance at 3800m/s, look for ways to enhance efficiency.
Checking performance at 3800m/s, look for ways to enhance efficiency.
The project involves a 2x8 GB RAM kit built for Vengeance 3000mt/s C16-20-20-38 (micron e die). Previously, I ran it at 3600 MHz with specific settings, but now I'm aiming for higher speeds. ----------------------------------------------------------------------------- Original configuration used 3800 MHz and maintained the same timings—dram 1.45V, increased SOC to 1.25V in 0.5V steps, and IDDT up to 1.2V in 0.5V steps (still at 1.05V). With the new target of 3800 MHz, I adjusted everything: IDDT remained at 1.45V, SOC stepped down to 1.25V in 0.5V increments, and IDDT rose to 1.2V in 0.5V steps (still at 1.05V). The voltage DPD increased to 0.95V (down from 0.9V at 3600 MHz). So far, these adjustments are helping me push performance further. I'm still trying to figure out the best approach and have a bit of confusion in the Discord community—any advice would be appreciated.
It's quite high for regular use; keeping the SOC below 1.2V is essential for CPU health. Many processors cap around 1.15V anyway. You might want to adjust the 1.8V rail instead. For consistent performance it didn't improve the Ryzen CPUs you tested, but it does boost scaling above 1900MHz (max FCLK on the boards I've used). On the X570 Master I run, it reaches up to 2.35V on that rail—even at room temperature—but for daily use I wouldn't aim beyond 2V. It seems you're missing out on a good fit with that chip. I don't think much of it, the 5800X3D isn't really focused on memory settings or FCLK values unless you need top performance. Spending a lot to hit 3800MT/s stability on it isn't worth it. Try lowering it to 3733MT/s or 3666MT/s and see if it works, but don't invest too much time chasing that result.
Ensure voltages stay below 1.2V—exceeding that is unnecessary and risky. Stable timing matters more than pushing higher levels. VDDG and VDDP should always be lower than SOC voltage. You lack enough details about your clock speeds and memory ICs. If you see a WHEA Blue Screen, it probably points to DRAM issues. When HWinfo reports errors, it suggests Infinity Fabric instability and DRAM problems.
I revisited the 1.15V SOC setting for my 3600 profile and think sticking with it is a good idea. The added voltage might slightly improve timing, but the benefit could be minimal.
I confirmed it's a micron E die. SOC is back to 1.15 v. 3600. 18-22-22-42 is stable at the moment, I'm trying to see how far I can go and also want to study more.
These settings seem a bit relaxed for E-die performance, but it shouldn’t be an issue from the DRAM side—possibly the CPU memory controller is struggling or the interface becomes unstable at 1900MHz when aiming for 3800MT/s. Try adjusting VTTDDR to half your DRAM voltage and set VPP_MEM to 2.52V. Turn off ECC in BIOS, which can help with memory training even without ECC support, though it skips the check at startup. Increase VDDP and VDDG; I’m currently using 0.925V VDDP and 0.975VDDG in BIOS for 3800/1900 speeds. These aren’t final numbers yet, but they work reliably for me—running four 16GB Micron E-dies with dual rank sticks. I’ve fixed primaries, tRC, tRRDS, tRRDL, tFAW, tRFC; the rest auto-adjust. You can set tRDWR to 10 from 18 for stability, but I haven’t tested much recently. I’ve maintained stability with lower VDDG and VDDP during DRAM tests, though WHEA errors appeared in HWinfo due to Infinity Fabric instability, so I raised them as above. If you use single-rank sticks, you should be able to set tRRDS, tRRDL to 4 and tFAW to 16. EDIT: if you have a RAM testing tool, it would be wise to use it—otherwise, avoid overclocking just in case.