[jfrancis]

I have a bunch of frames that are all pretty similar, just a slow move.

My times per frame are ranging from 50 minutes to 3+ hours and my benchmarks are in the 20s and 30s, the 60s, the 90s, and sometimes 120+

The machine is really not doing much but rendering. I'm using about 6 out of 16 gigs of RAM.

I know that's vague, but any ideas?

[jfrancis]

The frames all look pretty much like this.

There is an animated texture map that masks a lambert-like material over an emitter, and the same texture serves as a displacement map, so the emitter is, I guess, pretty high poly. I don't think it interpenetrates itself or any other surface since a displacement shouldn't be able to create self-intersection.

[jfrancis]

I woke up to a 3 hour render so I rebooted and now things are going twice as fast on the same frame.

[jfrancis]

I had what was shaping up to be a 6 hour frame. Benchmark 13.

I rebooted the machine. Now it's flying along at benchmark 197.

It really seems Maxwell doesn't free up some sort of resources from frame to frame.

[jfrancis]

Actually it seems like after a night of inactivity the benchmark is lowest, and if I putter around on the machine surfing the web, etc, the rendering stays fairly fast.

The opposite of what I would imagine.

I'm trying setting priority from low to high.

[rusteberg]

your benchmark can change dramatically just by changing the layer opacity value of a material layer by 1 value point on an object that's instanced 1^X times in a scene vs. not changing that one little parameter. given you've got a changing material parameter that's also affecting displacement (which is inherently sensitive) it sounds like you've gone and made yourself a troubling nightmare of hell trying to debug this one.......

[jfrancis]

It's always slow if I walk away for a while and it always speeds up again if I am busy using the mouse and keyboard.

I'm experimenting with power utilization settings and I'm running a windows program called insomniac that prevents the computer from sleeping.

[rusteberg]

why wouldn't you just change your windows settings instead of adding more bloat?

2015-05-13_1121.png

[jfrancis]

[quote="rusteberg"]why wouldn't you just change your windows settings instead of adding more bloat?



Because it didn't solve my problem. Nor, so far, have the additional settings indicated in my image above helped. But I'm continuing to test it later today by not touching it again for a while.

[rusteberg]

that's weird....

one solution: https://www.youtube.com/watch?v=sN37Fn3Hw8E

[jfrancis]

Seems like the hard disk was the problem. It stays fast now.

[nachob]

Also a CPU not properly cooled can produce the CPU to reduce frequency to keep the temperature in acceptable values. Check the cpu frequency is as it should during rendering with some CPU frequency monitor like CPU-Z or so.

nachob

[jfrancis]

Also a CPU not properly cooled can produce the CPU to reduce frequency to keep the temperature in acceptable values. Check the cpu frequency is as it should during rendering with some CPU frequency monitor like CPU-Z or so.

nachob

I'll look into that.

I haven't stopped my render since applying those latest power settings but actually what I've noticed is that now the idle computer rendering speed is about 2/3 of what it is when I am interacting with the computer (sending emails, etc). So an overnight frame in my current render might be 90 minutes and a daytime speed might be 60 minutes. The frames are all very similar otherwise. Night time benchmark in the low 50s, and daytime in the low 80s.

2/3 speed is a big improvement over what I was getting before (I was getting 1/10 speed before), and I'm happy with that, but It's kind of interesting and still a little mysterious so I thought I would mention it.

[numerobis]

I'm not sure if single thread processing is involved here, but you could try to set the power plan to "High Performance" and the minimal processing power of this profile to 5%.
There seems to be a problem with Sandy Bridge-E and Ivy Bridge-E concerning the single thread performance under certain conditions when you select the balanced scheme.

[jfrancis]

I'm not sure if single thread processing is involved here, but you could try to set the power plan to "High Performance" and the minimal processing power of this profile to 5%.
There seems to be a problem with Sandy Bridge-E and Ivy Bridge-E concerning the single thread performance under certain conditions when you select the balanced scheme.

I'll check it out. Right now I'm running a modified version of "Balanced" with the changes indicated in the image above plus I think I have both min and max processing power at 100%