Parsec,
I never no when to leave well enough alone. My Taichi X370/1700X/BIOS v3.20 system was running without any issues for months with the EVGA 3000-15 memory. The system would setup at 2933-16 and would run overclocked at 3900 all day and night!

I went and bought the wonderful G.Skill Flare X 3200-14 memory thinking I could get a little more out of the system. Unfortunately, nothing seems to work correctly. As long as I leave all the system settings at UEFI defaults the system boots (10 Seconds to post screen) and runs ok. But as soon a I turn on Profile 1 the system goes nuts and after 3 reboot cycles and ~60 seconds it usually turns the memory settings back to 2400-16. Before I take the G.Skill memory back out and put the EVGA back in and call it a $180 mistake is there something I'm missing?

I've tried clearing CMOS and starting over several times but always end up with at the same place - 60 seconds of the BIOS screwing around and then going into windows with the default memory settings. If I don't do a cold boot it will use the 3200-14 memory settings but goes nuts on the next cold boot.

Any ideas on what I have screwed up to make this thing refuse to use the G.Skill memory at the XMP Profile 1 settings would be appreciated!!

Edited by Ken429 - 20 Nov 2017 at 2:06am

calm down dear

right easy things first

make sure am4 advanced training is set to auto

and secondly

did you reset the cmos before installing the ram ?

I hate to admit it but I must have Disabled Advance Training one to many times. I went back and cleared CMOS, loaded UEFI Defaults, switched to Profile 1 and made the Pstate changes to get to the 3900 OC. All is well! The only minor issue is the "Cold Boot to the first BIOS Post screen" went from ~10 to ~20 seconds as soon as I set the Memory to Profile 1. That same thing happened with the EVGA memory but for some reason went back to ~10 seconds when I used Pstates to OC. Not so lucky this time. But then who know this thing might have a change of heart in a few days.

Anyway thanks all for making me go back and start over with the Advance Training "locked" on Auto. This thing sure is touchy when playing with memory.

Edited by Ken429 - 20 Nov 2017 at 3:23am

Can some one explain how all the synthetic tests indicate the faster memory turns in slightly better scores and yet when I run Handbrake and convert my favorite Blu-ray test case to an MP4 file the conversion takes ~2 minutes longer with the faster memory? I have run it several times and each time the conversion takes ~2 minutes or ~3% longer with the faster memory!?

Ken429 wrote: I hate to admit it but I must have Disabled Advance Training one to many times. I went back and cleared CMOS, loaded UEFI Defaults, switched to Profile 1 and made the Pstate changes to get to the 3900 OC. All is well! The only minor issue is the "Cold Boot to the first BIOS Post screen" went from ~10 to ~20 seconds as soon as I set the Memory to Profile 1. That same thing happened with the EVGA memory but for some reason went back to ~10 seconds when I used Pstates to OC. Not so lucky this time. But then who know this thing might have a change of heart in a few days. Anyway thanks all for making me go back and start over with the Advance Training "locked" on Auto. This thing sure is touchy when playing with memory.

Not that I am telling you to do the following, but I never use AM4 Advance Boot Training on my different ASRock X370 board. I set it to disabled, which shortens the startup time by quite a bit.

I use FlareX memory as you seem to know, but I don't use the XMP profile, it won't allow the memory to run at 3200. I use timings derived from a program written by a Ryzen user called Ryzen DRAM Timing Calculator. Which means I set ALL of the advanced memory timings manually. If your memory is working fine with the XMP profile, just use that with Boot Training enabled (Auto.)

Why does the seemingly faster memory take longer in a file conversion? Many possible reasons for that, one of them posted above.

So you ran synthetic tests on the 3000 and 3200 memory, which showed the 3200 to be somewhat faster? We accept benchmark test results as being perfect and correct and real, but when we see real world differences like this, who knows what is correct? Who verifies the benchmark results?

Better timing settings on the 3000 memory than the 3200 memory could result in better performance for the 3000 memory. For example, if the 3000 memory could work at that speed with a Command Rate of 1T, but the 3200 needs 2T to be stable, that can be a big difference in performance.

The test would only be valid if your CPU OC and CPU and Windows performance and power saving options are identical. Also the drive(s) being used for the IO must be identical, and any difference in that or the load on the drives at the time of the test could easily account for a 3% difference. But since you ran the test several times, that would not seen to be the situation.

If you read any memory reviews, you'll find that high memory over clocks do not always provide real world performance improvements. I've seen tests where the stock/default memory speed provided better performance than the over clocked speed.

Also, how long have you used the 1709 build of Windows 10? Did you use that with the 3000 memory?

DRAM memory can be weird, what should be better in performance with speed and improved timings does not turn out that way.