Practical Memory Settings

Performance Delta

  • 2%

    Votes: 0 0.0%
  • 5%

    Votes: 0 0.0%
  • 8%

    Votes: 0 0.0%

  • Total voters


Slightly less n00b
Jul 17, 2019
So we have a thing like this, looks great and all - dual-rank 8 gbit Sammy B running at a chilly 1.63v
Bit of a pain to set up, requires a fan, questionable long-term (12+ hour high-intensity workload) stability.

Thus I raise a question to the folks that peruse the place:
How much performance do you expect to see me drop, if I was to lower these values to something much more reasonable?

2x16 GB 3200 CL14-14-14 kits weren't particularly expensive, back when I last looked, and they're more than capable of doing "easy" 3800 16-16-16 at 1.46v with less pushed subtimings.

what's it gonna be??

I'll go run some Tests(tm) after some votes pour in, even if I already have a vague idea of what it's going to be.
I can't speak for the 5000 series of CPUs, but for the 3000 series running RAM at more than 3600 didn't yield any significant performance differences. This was my take-way from:
Well, you gave no specifics on what you plan to drop to or what work loads you're talking about, so really this could be anything. Just posting to so I don't forget to check back on results 😉.
I haven't tested the 5000 series, but the 3000 series didn't benefit from memory speeds over 3733MHz according to AMD's own documentation. Some boards you could push it a bit higher if you could get the Infinity Fabric to behave. But clocking over 3800MHz always required a divider, which tanked performance and wasn't worth while. Your FCLK is higher, and you are over 3800MHz, but I can't speak to how the Ryzen 5000 series will handle that. But, that RAM voltage is probably not sustainable. Benchmark it at reasonable settings and see what it does.
I usually don't mess with RAM too much. I buy RAM with official specs for the highest non-OC clock the platform I am using supports (or at least only a mild overclock considered practically guaranteed), with the lowest available CL timings, and then just use XMP settings.

In my current case this has been a mild OC of DDR4-3600, and CL16. It seems to work well.

Hasn't failed me yet. It just doesn't seem like it makes enough of a performance difference to worry about doing anything beyond this.
Last edited:
I've run DDR4 4000MHz+ RAM on the test bench and 3866MHz, etc. on multiple platforms since DDR4 came out. It rarely makes much of a difference.
Become a Patron!