Well I spent all day yesterday tweaking the system, and I can say system is error free. The biggest culprit was my BCLK clock (even though it was only 100, but more on that later).
This should apply to all Gigabyte boards, even though I was not crashing by buddy with the same board (his is the WiFi version) and CPU was.
Like I mentioned above, I always set my BCLK clock manually to 100. But whenever you set it manually the option for Spread Spectrum disappears, according to Gigabyte it should automatically disable. Well turns out it does not, if I leave BCLK on auto and then disable spread spectrum this resolved 40 percent of my issues. This was not found out until the very end when I just started messing with BIOS settings to see what worked (thank god for save profiles). I think modern platforms have spoiled us, because everything just runs smoother. It's not like the good ole 775 days where you had memory holes and needed certain straps to be stable. Or my old P5n-SLI board being quirky as hell with my Q6600 that I had to resort to a special BIOS just to run stable at the 333 strap (1333FSB/667DDR2). So in essence we have become lazy because boom we either just adjust the multi and voltage or enable PBO. I think that's why I enjoy RAM overclocking and tweaking as well.
Since I did not figure out the above I was pulling my hairs out to get zero errors. First issue I tackled was the SOC volts, this board with XMP enabled defaults to .995. It was never an issue because it never crashed. With my original settings (listed in the original post), I would get 230 errors with the following tests. Start up, play doom eternal for 10min, loop CB20 for 10mins, and then 10 more minutes of general usage. 230 errors and I stopped it at that, if I would have kept it open it would be higher because it would error in idling. With 1.15 SOC volts my errors dropped to 130 errors with the test mentioned above. I upped the SOC voltage pass 1.2 but gave me the same errors so I worked back down to 1.145 and that resulted in the same errors as a voltage of 1.25. I adjusted VDDP to .920 and VDDIO and VDDG to .945 (this setting voltage moved automatically). With final voltages tuned I got 90 errors.
90 errors, this is where I was pulling my hair. Started tweaking memory at this points but this resulted in more errors, with me tuned DJR running 3800CL16 I got less errors than with running stock XMP at 3600. What I find weird is 3800CL16 (IF ran 1:1 in all test) ran better and less errors than 3533-3733 and gave me the same errors running 3266-3466. At 3200 all errors disappeared, but I should have paid better attention, at 3200 even though I had BCLK at 100 it was 100. Every other strap was 99.6-99.9, this should have been a dead giveaway spread spectrum was doing its thing. Memory speeds 2400-2933 gave me same amount of errors. 2133-2400 error free and again bclk stuck 100. It was here that I finally decided to look into the BCLK jumping around even though it was set to 100 and spread spectrum should have been disabled.
With BCLK left on auto and spread spectrum disabled left me with a BCLK at 100. With this little change and 3800/1900, 10 WHEA errors. Now 10 errors I was excited, so know all I did was disable PBO and my lovely +150mhz and boom zero errors!!! Even with PBO disabled I still get my 4600 all cores and 4.85 single core boosts, so I am still happy. According to AMD and Gigabyte AGESA 1.1.8.0 and beyond should fix that (hopefully 4000/2000IF).
Just want to remind everybody, even though I had a crap ton of WHEA errors I was still apparently stable.