My speed test of 3600MHz CL16 vs 3800MHz CL18 (need input)
My speed test of 3600MHz CL16 vs 3800MHz CL18 (need input)
The dual ranking system doesn't compensate for the limitations of 16GB single-ranked configurations. This becomes more apparent with competitive benchmarks that often achieve a 2:1 performance ratio but struggle with memory speeds around 4000MHz CL12-12-12-12 and 12-11-11-11 on B-die boards at high voltage. It seems SK Hynix solutions with AMD or Intel chips aren't the optimal choice, regardless of ranking method. The performance tends to drop with increasing voltage. If I'm using an 8700K processor, I adjust the memory frequency and fine-tune between 1.50v and 1.60v, experimenting with slightly higher IMC voltage. It appears the system starts to behave noticeably around 1.30v, giving a bit of margin if you're below 1.20v.
Well Buildzoid suggests 3800MHz in Dual Rank works best with a Ryzen 3000, comparing RAM and FCLK settings to performance on that processor.
I noticed that. Maintaining a 1:1 ratio is definitely important. Your timing and latency aren't terrible, though they're not outstanding. Much of Intel's speed comes from the 40ns latency and high frequency. I'm running at 1933mhz (effective) with a clock of 14-14-14-24-444, TRFC 2TCR, 1.55v. Would you like screen captures?
It varies by where you search for bandwidth. Raising the CPU frequency at 100MHz can boost memory bandwidth, and you don't always need to overclock the memory to achieve it.
I think I'll stick with 3600MHz for now. It seems the IF clocks might influence temperatures more. If that's true, faster RAM with extra PBO headroom could boost gaming FPS even more. The default XMP setting is 17-19-19-39, but after GEAR DOWN it settled at 18. Since it was already 18, I wondered how much higher it could go—3800MHz works fine for Ryzen. I also checked how far timing can be reduced on stock chips; the results showed 16-18-19-37, with the last setting likely extendable further down but stability remains solid.
They might go 16-16-16-16-36. It sounds like a well-rounded pace.
At 1.35v the results were fine, possibly better at 1.4v. Buildzoid mentioned timing isn<|pad|>, so I didn’t test it. Temperature concerns are valid under full load conditions. :3
You mentioned a screenshot with 14-16-16-32 1T. It looks promising, perhaps a command rate of 2T would work better. Nothing will be lost from this approach.