Hi,
I came across a weird issue regarding memory timings and wondering if this is normal.
It seems if I make changes to Memory timings (that then show signs of errors during testing) and then I revert back to my last good working "safe" timings, I need to fully shutdown my PC and power back on otherwise I still get errors during testing. A soft reboot doesn't seem to fix it (e.g. Windows Restart -> go into the Bios (set the safe timings) -> Exit Bios -> System Reboots back into Windows). Only a full Windows shutdown seems to clear out the issue. I was testing using Prime95 (26.6) custom blend test using 14GB memory to use because some other tests didn't seem to pick up on the issue e.g Intel Burn Test.
I am new to Memory overclocking/timings but thought this was not expected as haven't seen this mentioned before and thought maybe this could be a motherboard bug? What do you think?
This might be useful for anyone experience similar issue where reboot didn't fix the issue but maybe try full shutdown then power back on instead.
My system is:
Ryzen R7 1700 (overclocked to 3.6Ghz @ 1.200v)
Asus X370 Prime-Pro
16GB DDR4 RAM (Corsair CMK16GX4M2B3600C18)
Windows 10 64-bit
Long description (background):
- RAM is on QVL list rated to support 3200Mhz (DOCP profile).
- RAM is Samsung B die.
- In Bios picked DOCP profile and set for 3200Mhz. The Bios automatically sets timing to 18-19-19-39 (it doesnt set timing for tRC, it leaves it on Auto) and DRAM voltage at 1.35v. Tested successfully PASS of 1 hour Google Memory Stress Test (93% coverage) via Bash on Ubuntu on Windows.
- At this point I am also overclocking the CPU @3.6Ghz @1.187v.
- Then manually set tighter timings 16-17-17-35-52 for 3200Mhz calculated based on XMP profile (3600Mhz 18-19-19-39-58) , it passes Memtest86 v7.4 (7 hours, 4/4 passes) and 1 hour Google Memory Stress Test (93% coverage), and 1 hour of Prime95 (26.6) blend test.
- System is pretty stable but at some point 1 or 2 days later I get the blackscreen issue - PC doesnt respond but FANs still spinning, the power or reset button doesnt work so need to turn off PC using the switch on back of the PSU.
- So I Reverted RAM timings to my last "safe" values 18-19-19-39 (and set tRC back to Auto like before). Test PASS 1 hour of Prime95 (26.6) blend test. No errors.
- Next day I find out that I can test system better if I use Prime95 custom blend test and set more memory to use, so I set blend test to use 14GB memory.
- After about 15 minutes 1 worker thread shows as failed so I decide to increase CPU vcore voltage as this is most likely the issue as Im using the same RAM settings stated in the motherboard QVL list. I set CPU 3.6Ghz @1.200v.
- Tests PASS 2 hours Prime95 custom blend test (14GB) and Intel Burn Test (Very High).
NOW HERE COME THE ISSUE:
- I want to try some different RAM settings in Bios e.g. RAM @3333 and 3466 Mhz; using slacker timings 18-19-19-39 (timings written on the RAM sticks). It boots to windows ok but fails Prime95 custom blend test 14GB - various workers fail.
- Try RAM speed at 3600Mhz (the rating of my Memory) but windows does not boot.
- So I reverted Bios settings to 3200Mhz and "safe timings" 18-19-19-39 (and set tRC back to Auto like before). Exit Bios and boot into Windows but I still got Prime95 errors using the custom blend test @14GB. I also tested using Intel Burn Test (Very Hard test) and this says it passed but I think this test only uses 4GB.
- Shutdown PC and power back on. Now it passes Prime95 custom blend test (14GB) 4 hours so far and no errors, all worker threads working full utilisation and 15/15.9GB RAM usage.
This issue was difficult to catch because it didnt show up in Intel Burn Test and on some occasions took 15 mins to show up the error in Prime95 (custom blend test @14GB), I got some hints when benchmark tests went down instead up when using faster RAM speeds and tighter timings, but cant always be sure with benchmarks so it was useful to get confirmation with the stress tests.
I came across a weird issue regarding memory timings and wondering if this is normal.
It seems if I make changes to Memory timings (that then show signs of errors during testing) and then I revert back to my last good working "safe" timings, I need to fully shutdown my PC and power back on otherwise I still get errors during testing. A soft reboot doesn't seem to fix it (e.g. Windows Restart -> go into the Bios (set the safe timings) -> Exit Bios -> System Reboots back into Windows). Only a full Windows shutdown seems to clear out the issue. I was testing using Prime95 (26.6) custom blend test using 14GB memory to use because some other tests didn't seem to pick up on the issue e.g Intel Burn Test.
I am new to Memory overclocking/timings but thought this was not expected as haven't seen this mentioned before and thought maybe this could be a motherboard bug? What do you think?
This might be useful for anyone experience similar issue where reboot didn't fix the issue but maybe try full shutdown then power back on instead.
My system is:
Ryzen R7 1700 (overclocked to 3.6Ghz @ 1.200v)
Asus X370 Prime-Pro
16GB DDR4 RAM (Corsair CMK16GX4M2B3600C18)
Windows 10 64-bit
Long description (background):
- RAM is on QVL list rated to support 3200Mhz (DOCP profile).
- RAM is Samsung B die.
- In Bios picked DOCP profile and set for 3200Mhz. The Bios automatically sets timing to 18-19-19-39 (it doesnt set timing for tRC, it leaves it on Auto) and DRAM voltage at 1.35v. Tested successfully PASS of 1 hour Google Memory Stress Test (93% coverage) via Bash on Ubuntu on Windows.
- At this point I am also overclocking the CPU @3.6Ghz @1.187v.
- Then manually set tighter timings 16-17-17-35-52 for 3200Mhz calculated based on XMP profile (3600Mhz 18-19-19-39-58) , it passes Memtest86 v7.4 (7 hours, 4/4 passes) and 1 hour Google Memory Stress Test (93% coverage), and 1 hour of Prime95 (26.6) blend test.
- System is pretty stable but at some point 1 or 2 days later I get the blackscreen issue - PC doesnt respond but FANs still spinning, the power or reset button doesnt work so need to turn off PC using the switch on back of the PSU.
- So I Reverted RAM timings to my last "safe" values 18-19-19-39 (and set tRC back to Auto like before). Test PASS 1 hour of Prime95 (26.6) blend test. No errors.
- Next day I find out that I can test system better if I use Prime95 custom blend test and set more memory to use, so I set blend test to use 14GB memory.
- After about 15 minutes 1 worker thread shows as failed so I decide to increase CPU vcore voltage as this is most likely the issue as Im using the same RAM settings stated in the motherboard QVL list. I set CPU 3.6Ghz @1.200v.
- Tests PASS 2 hours Prime95 custom blend test (14GB) and Intel Burn Test (Very High).
NOW HERE COME THE ISSUE:
- I want to try some different RAM settings in Bios e.g. RAM @3333 and 3466 Mhz; using slacker timings 18-19-19-39 (timings written on the RAM sticks). It boots to windows ok but fails Prime95 custom blend test 14GB - various workers fail.
- Try RAM speed at 3600Mhz (the rating of my Memory) but windows does not boot.
- So I reverted Bios settings to 3200Mhz and "safe timings" 18-19-19-39 (and set tRC back to Auto like before). Exit Bios and boot into Windows but I still got Prime95 errors using the custom blend test @14GB. I also tested using Intel Burn Test (Very Hard test) and this says it passed but I think this test only uses 4GB.
- Shutdown PC and power back on. Now it passes Prime95 custom blend test (14GB) 4 hours so far and no errors, all worker threads working full utilisation and 15/15.9GB RAM usage.
This issue was difficult to catch because it didnt show up in Intel Burn Test and on some occasions took 15 mins to show up the error in Prime95 (custom blend test @14GB), I got some hints when benchmark tests went down instead up when using faster RAM speeds and tighter timings, but cant always be sure with benchmarks so it was useful to get confirmation with the stress tests.