More RAM typically doesn<|pad|>, can lead to higher memory latency.
More RAM typically doesn<|pad|>, can lead to higher memory latency.
I applied AIDA during my Windows setup (sorry, I don’t know if there’s an AIDA64 for Linux) and noticed the latency stayed almost the same even after adjusting sub-timings back to the original values before adding more RAM. So unless @Pandur or @mariushm shares further details, I’m puzzled about why memory latency rose from 2 to 4 sticks after restoring the timing settings.
The options for 2 sticks are meant for two sticks. If latency has increased, it might indicate instability in your settings. What steps have you taken to verify this?
I didn't verify the long-term stability of the adjusted sub-timings, particularly after noticing minimal impact on memory latency once I restored the original RAM values before adding the additional sticks. As mentioned earlier, after fine-tuning the timings, I saw a 0.4 nanosecond latency increase according to Intel's MLC, while AIDA64 reported a 0.5 nanosecond drop. Given the small difference between these measurements, I attributed the variation to measurement variance. I also tested CSGO benchmarks to check if adding more RAM from 2 to 4 sticks caused any performance drops. Surprisingly, running four single-rank modules in dual channels now matches the performance of two dual-rank modules in dual channels. My average FPS in CSGO has risen by 6% to 9% compared to before, even with the overclocked 3400 MHz CL17 modules. So, it seems I'm probably overreacting, or are there still tasks that will suffer from such a tiny latency increase?
I'm also doing 4x8, which is great. I wish I could assist you with your tuning journey, but I'm new to modern Intel configurations. There are people here who know how to handle that platform—hopefully they'll find this thread useful. You won't really feel the extra latency unless you constantly check benchmarks. I prefer TM5 with the Anta777 Absolute or Extreme profile; I've been using Absolute more recently. I also like Linpack Xtreme, picking option 5 for the 10GB load, plus a few other settings. I wouldn't overthink it too much—just aim for stability.
It's definitely a challenge when working with Intel software to evaluate a Ryzen system, especially when memory latency is fluctuating. I hope those people you mentioned have a better grasp of how the memory latency checker functions, so they can help explain why your latency rose from 2 to 4 sticks despite adjusting sub-timings. I was just as puzzled, but I'm more eager to understand the cause and see if there are simpler fixes without going back to my 2x8 GB configuration. I ran three rounds of memtest86 (about 50 minutes each) and spent 35 minutes on prime95 Large FFTs during the initial RAM upgrades and 3266 MHz overclock.
I've been away over the weekend, so I'm a bit lagging behind. Memory timings are per populated side of each RAM module (one rank). When you add more modules or switch to double-sided ones, your memory controller will take longer to handle routine tasks. This often results in a performance hit, even with the same settings. I'm saying this generally, since other factors can boost speed at the same time. This gives you a quick overview of what's happening.
Thanks to @Pandur, @freeagent, and @mariushm for clarifying my memory latency issue. After reviewing all the details in this thread, I think I’ve got a solid understanding of what’s happening. I’m finalizing this post as resolved. Unfortunately, I can’t tag multiple answers as solutions since the forum doesn’t support that. Also, labeling this thread as the answer wouldn’t be appropriate because I wasn’t the one who resolved my problem. @Pandur’s latest update helped clarify everything, so I’m choosing to highlight his response. Appreciate all the support!