MAJOR issue: Zen 3 Motherboard BIOS issues (UPDATE: FIX FOUND! 1933MHz IF Stable)

[terminalinfinity]

On 11/11/2020 at 2:52 AM, Eajimoba said:

Manual overclocks and XMP profiles past 3200Mhz resulted into no boot, I couldn't even get into windows to check on the WHEA errors  

 

At 3200Mhz and below there were no issues, had looping 3DMark Timespy tests back to back for like 2 hours.

 

Now at my 3600Mhz CL14 [Manually tuned off of the XMP profile] I'm not seeing any Whea errors, it's weird though how some were getting them and some weren't

Looking into early reasons why MSI boards were the most harshly affected now that I've really ascertained what's going on, they seem to have the most aggressive voltages for the SOC and IF in XMP mode over other boards.  Until I passed 3866, my I'd get WHEA errors if my SOC voltage was 1.1, which is MSI's XMP auto setting.

 

Once I needed the voltage for IF stability, it was the opposite.

 

From preliminary findings I'd start with 1 volt SOC, 0.900 VDDG voltages and bump as needed when going up the ladder.  It seems what causes errors at one point, might not at another.  Entering these manually should also work for XMP profiles on Zen 3, not just manual OCing.  Should be a fix in lieu of a BIOS update

[terminalinfinity]

It's all coming down to SOC voltage.  I have tons of leeway on VDDG voltage, Im back down to stock 0.900 voltage, but I have to be within .02 volts of a certain point in SOC voltage once I get above 3666. 

Too high: instability.  To low: instability.  While this was somewhat the case with Zen 2, the sensitivity of IF to SOC voltage is on a whole other level with Zen 3.

I got it to 1933 before I hit a wall @ 1967 that both going lower and higher on SOC voltage results in more instability.  But I was still able to post unlike the wall I hit with Zen 2 at 1900.

 

Also, for the first time, I'm seeing scaling in IF with voltage that is directly proportional.  Didn't see any type of linear scaling with my 3600 (Zen 2) and 1700 (Zen).  The returns were disproportional to any voltage applied.  The only time applying more voltage was useful for IF was if you were on the threshold of stability.

[terminalinfinity]

Alright, had time to do more testing, moving to dialing in ram.  This is my procedure for dialing in your Zen 3 IF.  Posted to reddit, getting feedback of success on other MBs as well. 

BETA GUIDE TO DIALING IN ZEN 3 INFINITY FABRIC

1.  Set SOC voltage to 1.0, both VDDG voltages to 0.900, and ram to 2133 16-20-20-20-40 to eliminate any chance of a fluke RAM error

2.  Set FCLK to 1600.

3.  Test with AIDA64, OCCT 30 minutes each.  Confirm stability.

4.  If unstable, increasing and decreasing SOC voltages in .025 steps.  Max SOC voltage is 1.2

5.  If you hit an SOC wall where you reach 1.2 volts or changing voltage either way increases the WHEA error rate, trying increasing VDDG voltages in 0.025 increments  Remember that VDDG voltages can only be .05 volts below SOC voltage at the most (they can go as low as stable) as infinity fabric voltage is directly derived from the SOC voltage.

6.  Dial up FCLK one notch at a time.  Repeat step 3.

7.  If instability, repeat steps 4-5

8.  For final stability test at desired FCLK, run both AIDA64 and OCCT for an hour each.

9.  Bring your RAM frequency up to match and begin dialing in timings as normal.

 

[Eajimoba]

On 11/14/2020 at 3:57 PM, terminalinfinity said:

Alright, had time to do more testing, moving to dialing in ram.  This is my procedure for dialing in your Zen 3 IF.  Posted to reddit, getting feedback of success on other MBs as well. 

BETA GUIDE TO DIALING IN ZEN 3 INFINITY FABRIC

1.  Set SOC voltage to 1.0, both VDDG voltages to 0.900, and ram to 2133 16-20-20-20-40 to eliminate any chance of a fluke RAM error

2.  Set FCLK to 1600.

3.  Test with AIDA64, OCCT 30 minutes each.  Confirm stability.

4.  If unstable, increasing and decreasing SOC voltages in .025 steps.  Max SOC voltage is 1.2

5.  If you hit an SOC wall where you reach 1.2 volts or changing voltage either way increases the WHEA error rate, trying increasing VDDG voltages in 0.025 increments  Remember that VDDG voltages can only be .05 volts below SOC voltage at the most (they can go as low as stable) as infinity fabric voltage is directly derived from the SOC voltage.

6.  Dial up FCLK one notch at a time.  Repeat step 3.

7.  If instability, repeat steps 4-5

8.  For final stability test at desired FCLK, run both AIDA64 and OCCT for an hour each.

9.  Bring your RAM frequency up to match and begin dialing in timings as normal.

 

 

Man, many thanks for your hard work.

 

I spent whole day yesterday using dram calculator for zen2 on my 5800x [zen 3] and played around to no avail. Ended up going back to 3600 14,14,14,30,70 for my stable

 

Will use your method when i get home later and see where I land