The CPU GTL Voltage depends on the CPU GTL Voltage Reference and the VTT. I dont think it changes by itself.
One example how it works:
9x450MHz
After i managed to get Large in Prime95 stable (here the VTT is important, CPU-GTLīs doesnt matter (Auto or 0.63) as long the vCPU is high enough), i lowered the vCPU. 1.275v (Bios) nearly passed 8k small in Prime 95 (1.84v load), but know matter which CPU GTL Reference, one core fails nearly at the end. With seperate GTLīs for every core this would pass.
Ok next try 1.281v in Bios (1.192v full load):
first try with CPU-GTL Reference: 0.63
1.12v x 0.63=0.7056v fail
next try 0.62:
1.12v x 0.62=0.6944v fail even earlier
0.65:
1.12v x 0.65=0,7280v fail, but after 12 minutes
0.67:
1.12v x 0.67=0.7504v pass!

So i need about 0.7504v to be stable with that Vcore.
When 0.67 would have failed, i would have to raise the vCPU by one nodge and would have to start the tests again.

With Bios 0001 (modded) like always, when a new Bios arrives, i made the long run with Prime95. With the same settings which were stable before, Prime95 Large failed. The solution was, to raise the VTT to 1.14v. And now Prime small failed...

1.14v x 0.67=0.7638v
so this seems to high for to pass with 1.192v Load
1.14v x 0.65=0.741v passes again!

To be stable with Linx i have to raise the voltage to 1.287v or 1.29xv in Bios. But the Load-Voltage stays the same 1.192v.
I made a run on vCPU Auto yesterday, looks like 1.275v in Bios, 1.84v load (prime95 small), I knew this wouldnt be enough. Linx gave me a BSOD after 2 minutes, but funny the Load-Voltage was about 1.192v-1.200v!? No way to use vCPU on Auto...