x570 - Why is my latency so 'high'

What DRAM voltage are you using for 3800 CL16?

You can easily try 1.45v and try for CL15 then 14 if you haven't done so already.

Even 1.5v for Samsung B die, but people tend to have an irrational phobia of that for some unknown reason.

See what it can do on 1.45v in any case, no point to having B die and not trying at least that much.
 
A lot easier on 5900x

x570_memorylatency5.jpg

Weird, my run time is half yours...

Your timings are a bit off as well, trc should be 50, trfc try 300 and twr is way too high.
 
Hello guys, can I jump in this thread for some help with my Ryzen DRAM calculator tuning?

I read my Crucial Ballistix Micron E-die 2x 16Gb 3600Mhz dual rank sticks with Thaiphoon burner, exported in nanoseconds the complete HTML report then imported this into the DRAM calculator. I can then click SAFE to get these settings which after a long time I managed to input into my MSI Tomahawk X570 motherboard BIOS. The settings seem stable on memtest with 1.4v memory and the recommended voltage settings on the others.

https://1drv.ms/u/s!AsMDJqyHDc8RguETIJibrQaDB8p-yQ

This seemed to only gain me 1% in 3d Mark timespy vs XMP which seemed less that I was expecting for dropping CL16 to CL14.

When I try to choose the Fast option it tells me it is not possible with Manual mode (which is what it switches to when importing the data from Thaiphoon burner). If I change to A0 memory quality it deletes all the left side readings but does give me new values with the FAST button. Is this normal?

My memory quality is 114% and apparently could go up to 4723Mhz at CL16.
 
3600/16 is already well into the realm of diminishing returns so you won't gain much more real world performance by going tighter.
 
https://1drv.ms/u/s!AsMDJqyHDc8RguEW3J4j2RJyB7u0HA

Here is my MEMbench run with the SAFE timings which include CL14. My run time is a bit slower than the ones posted above but I think I can try to tighten one or two from what the calculator suggested (tRFC and tRC based on calculations suggested elsewhere). Also mine being a 5800X with 16 threads rather than 24 might make a difference. I had one time that Chrome froze for a few seconds despite finishing 4 runs of memtest without errors which I wonder could be a sign of overly tight timings. Lastly I am going to run Aida64 with XMP vs Tuned timings to see if there is a latency difference. The <60 result here seems quite good though.
 
Back
Top Bottom