First thing is to get the drivers installed and see how low the undervolting can go on your new card. 
ive actually never touched undervolting i was just tonna max power slider and go ape shit
That is how not to do it on vega.
1 Like
Air cooled that is⊠Just sayinâŠ
its going in a custom loop
Youâll still want to undervolt, and then set the power limit to +50%.
but what ahoc said just pump the slider to 100 as the cap stops at 400v and at 100% ur only at 320. so ur not even maxing the card out. idk so much miss information
NoâŠ
The whole idea of undervolting is to get highest clocks available without thermal throttling. Custom loop, done correctly, will take thermal throttling out of the equation. So the card will want all you can give it power and voltage wise. All you have to consider is heat dissipation and power consumption vs performance gain.
Well my experience with my (admittedly Vega 56) card is to do what I said, even though itâs under water. It doesnât run fully stable on the 64 LC BIOS so I am running the 64 air BIOS (this one in fact is the best I have tried so far: https://www.techpowerup.com/vgabios/194442/amd-rxvega64-8176-170719-1 ) and undervolting while maximizing the power slider gets me the best performance period. When I raise the base voltage, the only thing that goes up is the temperature and⊠you guessed it⊠power draw. I donât get better framerates.
YMMV of course.
Small piece of advice if you want itâŠ
Do not be cheap with the thermal paste. Use âthin, easy spreadâ as opposed to thick type. So say MX-4 over Kryonaut. Vega has a âhot spotâ sensor that will thermal throttle your card if paste/block coverage sufficient. To my knowledge only GPU-Z monitors that sensor. It is very important if you want to get good stable clocks anywhere near the Vega Liquid version. On both my Vega 64s the âhot spotâ tends to be about 8c higher at 1670 core clock, so 55c running at 1690ish boost clock.
I could get better âhot spotâ temps with reseating blocks with more thermal paste (I used Kryonaut), but will not bother till maintenance time and with MX-4. I can run these cards at for 46 hours straight,100% load without temps going over 35c core, 41c HBM2 and 50c âhot spotâ at 1650 core boosting to 1672 constant - rendering. Fluid does not go above 29c in 21c ambient room. However they have a 560 and 360 rads exclusively for GPUs at 60% fan speed as I intend to add a third card.
im running 2 360mm ek 40mm thick rads and im using 1 gpu 1 cpu. so i should be fine. and im gunna run the card vertical so ill gain more airflow. also. i have mx4 i was wondering if that is better the what ek sends
HWiNFO64 is now reporting hot spot temps (third line under GPU):
Note the HBM bug.
1 Like
Did not know you watercooled, custom loop.
Run GPU-Z and watch your âhot spotâ sensor. Most likely the culprit for your issue and need to undervolt.
Compute loves the OCsâŠ
I run same clocks in games, but I rarely game.
See above, I do watch the hot spot 
Edit: it never goes over 55°C.
I also used Kryonaut, spread evenly with good coverage. I quite like it. This is my first custom loop but Iâm addicted. I did have a whole thread on it while I was working it out: Full Loop with XSPC?
GPU-Z does not have a HBM2 bug, try that. Nice to see HWinfo have the âhot spotâ. Now stress your card with that open. Letâs see what you get.
I only just noticed the bug actually, it wasnât bugged on the previous version so thatâs a little curious.
what kinda clocks did u get im hoping for 1700plus core 1100plus memory
That is what I thought. But you cannot run the 64 liquid BIOS. The âhot spotâ is preventing that. That BIOS drops temp limit to 70c and most likely starts to down clock at 60c.If done correctly, your âhot spotâ should not be higher than HBM2 temps.
1100 memory is easy but I do not suggest going higher for everyday use. Benching only.
Sustaining over 1700 clocks you are best to flash to a 64 liquid BIOS or edit to get the extra voltage allowed.
Your âhot spotâ temp will be very important to do that.
Note I am speaking gaming. Compute is easier to sustain clocks than gaming. Vega is a compute card that can be used for gaming.
Max I saw was 57°C, HWiNFO hot spot matched GPU-Z. Whereâs the evidence that hot spot should match HBM2? I hadnât heard this before, and have only really known that the hot spot needs to stay under 80°C before itâll throttle, same with HBM before timings loosen and performance drops.
I agree about compute, rendering doesnât heat things nearly as much and clocks are dead solid.
