cancel
Showing results for 
Search instead for 
Did you mean: 

Vega 56 2019 software allowing all p-state adjustments

feedmeink
Level 12
When I originally got my vega56 I put it under water and did the basic overclock on state 6 and 7. Been focusing on memory fine tuning and cpu tuning. Found out in the 2019 version of wattman it allows you to fine tune each state. I would like opinions and suggestions from others who have dove into stability checking each state. I would take say state 2 and adjust voltage and frequency. Then set state 3 4 5 6 7 to the same frequency and voltage. Through hwinfo I pull graphs for voltage for frequency and memory, and gpu core frequency and memory. I notice the voltage read out is about 20 lower. If I set say 950 for state 2, it will read 920-930. My mentality is to run firestrike ultra stress test and check for frequency and voltage fluctuation. Find the sweet spot for each p state to allow basicly a flat line on my graphs. I noticed when I alter state 7, no matter what I do, it runs at 5,950 and voltage tops out at 1120. Even if I set voltage to 1200. Am I going this properly? Memory voltage stays steady no matter what I set in wattman. So I just leave at auto voltage for memory and climb as high as I can to maintain that flat line in my graph. Any suggestions? (Its been 10 years since i pushed a card)
1,512 Views
2 REPLIES 2

feedmeink
Level 12
For anyone that has this question, I finished fine tuning each P state on the vega 56. This would apply for vega 64 as well. First I left all states at stock. Overclocked the memory with hwinfo open and graphs for memory frequency and voltage. Ran fire strike extreme STRESS TEST. Not benchmark. Monitored frequency and adjusted to get a flat line on the graph. To make sure it wouldn't overshoot and undershoot at random. Once I found memory frequency I moved on to p states. Don't bother messing with memory voltage. It is locked at 1.25 for vega56. Vega 64 is 1.35. No matter what you set this to in wattman it will stay the same in hwinfo. Then I started with state 1 frequency. Set state 1-7 at the same frequency and voltage. Run stress test and monitor frequency and voltage fluctuations. Adjust voltage and frequency as necessary to produce a flat line on graphs within hwinfo. Then move on to state 2. Keep state 1 values and set 2-7 same frequency and voltage. Continue this process through state 7. Previously I was able to get state 7 to 7,007mhz frequency. I tested that and saw it only reading 5,950mhz. Now my state 7 is 6,020. Same frequency with less voltage. Doing this process only changed benchmark score by 6 points, using 100 watts less power. It runs super clean on voltage and frequency graphs. Performance is noticeable. Any questions let me know. Hope this helps someone out there.

Just a follow up, do not change original voltage for state 1. For some reason was getting black screens. Changed from 930 to 900 and runs fine now. I think it complains when first state is different than stock voltage