Need some help understanding LLC and offset OCing - Page 2

I could always play with it more.

Offset will adjust voltage within set range depending on the load, right?

With 1155 it offsets the voltage a certain amount depending on the clock speed, which depends on the load. Then there's Vdroop and LLC.

I would imagine this is similar to 1366. So your idle voltage and load voltage and every voltage for every speed in between will all get +/- whatever your offset is set at. Then factor in LLC and Vdroop at load.

@OCScrub
No, I am fairly sure that the offset is pretty much a fixed value. If the clock vcore is on auto, then the offsett just adds or subtracts from that.

So the offset volt value is not depending on or adjusting with the load, afaik.

Here's how I think of it:
Vidle = VID(1.6 GHz) + Voffset
Vload = VID(4._ GHz) + Voffset - Vdroop(LLC)
where
- VID(x) is a look-up table (or curve) unique to each chip, found through experiments.
- Vdroop is varied via LLC, but with limited control (only 5 settings, which gives limited precision and maybe not enough range).

But if using Offset mode, be prepared for more work, confusion, and a good chance of no overall improvement vs Manual mode. With Offset, you have to get it stable at both idle and load, you don't know the VID curve upfront, you'll need to play with LLC and Voffset simultaneously to get Vidle and Vload to both be appropriate, and you'll have limited control on optimizing Vload.

With some chips it may indeed offer a good reduction of Vidle with an optimum Vload. With other chips (I think most), Manual mode will give the lowest stable Vload, and idle is guaranteed to be stable.

If a stable OC is achieved using Offset, you might try switching to Manual to see how low a Vcpu you can run the high speed stable. That Vcpu is likely to be lower than the Vload using Offset, since you can optimize it independent of idle issues. For chip life, I'd would think that a relatively small drop on the high end is better than a big drop on the low end.

For power efficiency, my kill-a-watt meter showed only a 2 watt difference for wall power between a 0.9 Vidle and a 1.25 Vidle. (At idle, there is so little current flowing that it doesn't much matter what the V is.) There may indeed be a larger difference at a mild load, but I'm not sure how to rig a constant mild load that doesn't raise the Ratio--any suggestions?

If Manual mode tuning resulted in a Vcpu that was so high it may shorten longevity (1.35? 1.4?), then I'd look pretty closely at Offset mode. But I'd keep it only if the resulting Vload was the same or just a little higher.

I really wish the BIOS had a more intelligent method for controlling Vidle and Vload, like setting them independently and interpolating, or having 2 offsets: 1 for Vidle, 1 for Vload, and interpolating on top of the VID curve. That would be sweet! I wonder why they haven't done it.

I have been wishing this was a thing too

Yes. LLC will raise your Vcore at load. The highest level of LLC usually overcorrects for Vdroop and ends up raising the Vcore past where you intend it to be. If you have a small positive offset with low LLC, your idle voltage will will be slightly higher than VID, but your load voltage will probably be a little lower than VID due to Vdroop.

LLC shouldn't affect anything at idle.

You raised your LLC and lowered your offset. I would say that your Vcore at load is probably going to be about the same as it was before. If this run of prime fails, try putting your offset at -.02 again and keeping the current LLC.

Yeah I wouldn't worry about the idle too much. As long as you're stable at idle, you're still gonna be getting lower temps and power usage than if you were just doing a manual overclock. Not to mention the fact that you're potentially prolonging the life of your cpu.

Like I poster earlier, I idle at 1.08, which makes my idle temps like 27-33-31-24, so it's really not a huge concern.