The typical FCLK inquiry 🤔
The typical FCLK inquiry 🤔
Looking for credible evidence on performance drops when memory clock exceeds fabric clock? I found several comparisons between Cinebench results for different configurations. Running the same multicore test three times on each setup confirmed the trends. The 3666/1800 achieved a higher score, supporting the idea that pushing memory clocks too high can impact overall performance. While I haven’t overclocked fabric clocks beyond 1800 yet, this aligns with current findings. It seems some tests even reached 1900MHz on FCLK, though that was done in Linus’s videos. For fabric clock limits, stick to what Cinebench shows as realistic.
This result is quite close to what we expect, and Cinebench doesn’t focus much on available RAM.
No evidence confirms the clocks are fully independent, which raises concerns about increased latency and performance drops. If Cinebench isn't reliable, consider using other benchmark tools like AIDA64 or 3DMark for more accurate results.
Your theory holds true. Decoupled introduces greater latency compared to Coupled. However, it's possible that pushing the Memory clock even further than the FCLK could reduce latency. Linus mentioned this too; I plan to explore it and share more insights soon. The gap between MCLK and FCLK seems substantial.