In Windows the numbers will be slightly lower than Linux.
However, I can give you a few pointers to optimize the overclocking settings,
1- Fold with the card stone cold (<55C), no OC.
2- Read the GPU frequency before the card hits 55C. This will be your aim to overclock to.
3- Set fanspeed to 100%.
4- Reduce the power limit. In Linux you can have a very precise power tuning. In MSI Afterburner for Windows, there's only a slider. But drag the slider all the way down (lowering your card's performance,while reducing it's heat as well). You can pair the information with what GPUZ tells you your card is consuming.
5- Increase overclock. Most of these cards do 100Mhz overclock easily, but some can only do 60Mhz. Others can do 135Mhz (and GTX cards can often do even more than that). It's just the luck of the draw. Start increasing overclock ~20Mhz every minute or so. When the WU fails, dial that number down by 30. This is your baseline overclock value. Then run further tests:
6- Dial down the fan speed, to where the temperature either stays below 55C or below 60c (or below 65C); whatever you can bear in terms of fan noise.
7- Optimize airflow in your case. Try to figure out why a card isn't cooling well. Sometimes taping off certain areas, and placing fans in the right location/moving air in the right direction, or removing some PC case panels, can help keep the GPU cooler.
8- If the WU runs stable, and no failure happens increase by +5Mhz increments every few hours or so. If the log displays an overclocking error, dial down by 5Mhz. The readout's accuracy is only 15Mhz, but internally the GPU can be precisely adjusted to a single Mhz or two.
9- Once you have found your maximum stable overclock (say +125Mhz at minimum power levels), dial down the overclock by 5 or 10 Mhz, and increase power levels and fan speed to where the GPU frequency is close to the stock boost frequency found when you tested the card cold, and where the temperatures stay below half a decimal value (50, 55, 60, 65, or 70C).
When done right, you should have ~95% of the stock performance, for only ~60-80% of the stock power consumption; depending on which RTX/GTX card you own.
From there on, you can play with the power levels, to see how much PPDs you gain for every +10W (or +10% of) extra power.
Most cards perform poorly when set to minimum power levels, and have a peak efficiency somewhere between minimum power draw, and stock power.
Once you found where on the power scale the card runs most stable, highest PPD per watt, keep the card there.
Some people prefer to run the card with low power settings, saving $$$ on electricity.
Some people prefer to fold more efficiently, at slightly higher electric cost.
Some people prefer to fold closer to the maximum the card can fold to, at even slightly more $$$ on electrical power.
The higher the power levels, the faster the card will fold, but the lower the efficiency; to the point where the last 10-30% of power added to the card (ending in stock power levels), results in higher electric cost, at no noticeable performance gains.
post edited by ProDigit - 2019/07/26 22:39:33