1:1 or separate FCLK/Memory?
1:1 or separate FCLK/Memory?
I've noticed this subject swings in both directions regarding performance improvements and setbacks. There aren't clear rules on what to do or avoid when exploring it. From what I understand, decoupling might carry some kind of cost, though the exact impact is unclear. I'm curious about how penalties could be balanced with gains, whether there are specific voltage settings for stability, and if testing it is worthwhile.
On Ryzen 3000/5000 models there was a significant delay when using memory without synchronization, and unless you could achieve very high memory speeds—over 4400MT/s—the performance drop was hard to ignore. For regular use it's usually safer to aim around 3800MT/s to maintain near-optimal latency. On Threadripper the issue was even more pronounced, so memory decoupling offers little benefit, though you'll likely run at lower speeds where a 1:1 setting makes sense. With Ryzen 7000 there’s no 1:1 mode since MCLK and FCLK are fully independent, so pushing both clocks is ideal. On Ryzen 7000 and 3000 the main concern is SOC voltage; it usually settles around 1.1–1.2V, with 1.15V being a solid choice for most chips. VDDG voltages assist in memory/FCLK tuning, but from my experience the motherboard handles them well without needing adjustment. The 1.8V rail aids FCLK overclocking, though it’s mainly useful for advanced users trying to squeeze extra performance rather than everyday use.
The 7000s don't have a direct 1:1 match. That suggests a penalty was added in the earlier version.
I concur, my system uses an AMD /900X with 12 cores Zen4. It appears "synced FCLK" has no impact on performance in games or Cinebench—results remain consistent, though power usage drops slightly. My SOC operates at 1.1V with two 16GB 5600MHz CL36 DDR5 modules.