Low bandwidth issue at 3900MHz
Low bandwidth issue at 3900MHz
Hey everyone, I recently upgraded my setup: CPU is an i9 10900K, GPU is a Gigabyte RTX 3080 Gaming OC MOBO, motherboard is Gigabyte Aorus Z490 Ultra rev 1.1 with BIOS F7a, RAM includes Viper Steel Patriot DIMM 16 GB DDR4-4400 (2x8GB), and the power supply is Thermaltake Toughpower PF1 850W. I also added a 2 TB SSD and an AIO RAID controller.
I faced some stability issues with the RAM sticks—none of the default XMP profiles worked well. The XMP1 settings were set to 4400MHz at 19-19-19-39 at 1.45V, while XMP2 was at 4266MHz with the same specs. After testing for hours, I found two reliable configurations: one running at 3600MHz with 15-15-15-35 and another at 3900MHz with 15-15-15-35.
The 3900MHz version gave me a noticeable boost in read/write speeds, especially for bandwidth-intensive tasks. However, I’m still struggling to get consistent results. If I switch to 4000MHz, I’d expect even faster performance (around 57K/58K/52K), which aligns with what the community recommends. But right now, stability remains a challenge.
Have anyone noticed why my 3900MHz setup is underperforming? I haven’t changed any secondary or tertiary components yet because I’m still getting the hang of things. The big question is: how much of an impact does a 100MHz difference really have on these metrics?
Increased voltage paired with active cooling. (Fans positioned directly on the memory). Performance depends on CPU and memory speed together. Delay stems from these factors, especially when tighter timings are used. For testing only, I benchmarked at 4000mhz 14-14-14 at 1.60V. Daily operation at 16-16-16 at 1.510V should keep memory at 40°C or below if possible.
Hi @ShrimpBrime, thanks for your reply. Unfortunately, active cooling isn’t available for me—it’s part of my regular setup. I’m not aiming for top-tier benchmarks; I just need reliable, stable RAM so my CPU and GPU don’t get overwhelmed. The issue is that I can run at 4000 MHz while keeping voltages the same as 3900 MHz, which feels a bit shaky. The bandwidth difference in the AIDA benchmark seems huge—from 40K values to 57K values in a single screenshot. I’m puzzled by this. Ideally, staying around 3900 MHz with acceptable bandwidth would be perfect. I’ve spoken with colleagues who report around 45K at 3200 MHz out of the box, which is five kilohertz higher than my 3900 MHz setup. I don’t understand why that’s happening.
I dislike encountering identical issues in forums that seem to lack any resolution. I realized the problem lies in two extra settings automatically configured by my motherboard at low frequencies—tRDRD_dg and tWRWR_dg. At below 4000mhz, my board was assigning them a value of 6 instead of the correct 4. Changing these manually fixed the matter. To discover this, I had to study extensively and adjusted other parameters as well, resulting in a 59GB/s bandwidth at 41.5ns, which I’m very pleased with.
Yes, sorry if my response wasn't timely or clear. This isn't part of my role—I'm here to assist independently and not for payment. Often, automatically set board timings can be incorrect, but usually it's not a major problem. I'm facing the same situation on an older AMD platform, though the principles remain similar. It's interesting how sensitive PiFast is to memory timing. Thank you for getting it working! Good show!
mine runs smoothly at 50 ns, 4000 MHz with settings 19-19-19-39. Overlocked version is 45 ns at 4000 MHz using 16-16-16-32 at 1.45V. Everything seems fine. Please adjust secondary and tertiary timings if needed—missing them can cause stuttering in games.