Quote from: BeeDeeEff on June 18, 2019, 12:48:28 PM

Step 2 - Divide by 0.8 so that if your OCs get reset and you don't notice you've sized to pull no more than 80% of the psu's rated power.

...

You really don't want to be maxing your psus out anyway for efficiency sake, you're burning more power by not running closer to the peak of the psu's efficiency curve.

It's unfortunate how tenacious this 'common wisdom' is. Maybe it was translated from home wiring circuits (where it's def important to maintain 20% headroom when talking about sustained loads,) or it's from older/cheaper/lower-end PSUs w/ more bell-shaped curves, but current high end PSUs have no issue running at capacity. In fact the graph you pasted shows measurements up to 12.5% above cap - though I wouldn't necessarily test that w/ a 24/7 load.

As to the efficiency and power use concern, peak eff is around 33% of cap - that would be insane to leave so much unused cap. Assuming 75-80% is the guideline as suggested, you're talking about maybe a 2% eff drop between that level and full cap. Which assuming a $0.1/kwh power rate translates into:

((1600 * .02) / 1000) * 0.1 * 24 * 365 = $28

That's $28 per year in extra energy cost by fully utilizing your PSU vs handicapping it at 80%. Even if you double that (splitting the load will get closer to peak on each PSU,) given that that the AX1600i runs about $470 new, and the avg lifespan of electronics is 3-5 yrs, it's clearly much more financially prudent to simply utilize the full potential of the PSU.

Also, if someone is really worried about a reset-situation kicking the GPUs back to stock, they should just use PPTs. Although even w/o that, the over-current protection of the PSU will handle what should be a rare occurrence anyway.

Quote from: BeeDeeEff on June 18, 2019, 12:48:28 PM

On another topic, I've found that for ethash the point at which the gpu core no longer bottlenecks the memory on this gpu to be when the gpu clock is about 1.8-1.83x the memory clock. Or 1800mhz core for 1000mhz mem, or 1620mhz for 900mhz mem. Is everyone seeing similar in their fiddling?

I found 1575 cclock was roughly the line for 935 mclock in my case. Though this was a while ago, on an early bios and older drivers.