Render engines at 100%

I love working with Corona. One week ago I had to revert back to Mental Ray for one loooooong animation and an old problem came along.
As my computer renders a mental ray-image, everything is so slow. Cant do much anything with my computer when CPU-meter is at ~100%.
How come I can do much everything when I render with Corona and the CPU-meter is at ~100%?
Dont get me wrong, this is a good thing. Keymaster seems to be one hell of a coder :smiley:

Is there same things with Vray? ..I have never used that.
No going back to mental ray - that would be just mental.

Yeah, it is the same thing with V-Ray, I used to play some games while rendering or doing some designs, normally without hiccups. I guess that Corona and V-Ray just manage their resources better. Of course it won’t render as fast as when you would let it use 100% of CPU power, but as long as you can live with some lost time it is no problem :smiley:

Nowadays, I am just using another node for rendering so my main rig is free :smiley:

I find that as well as doing this, changing the processor affinity to leave a couple of free cores really helps too.

There are people with exactly opposite experience. In theory there should not be any problems, but it seems there are some fuckups in windows and/or 3dsmax.

or change max afinity to all bar one core?

I think that as of Max2014 Mental is also set to low priority by default. And if i’m not mistaken setting 3dsmax.exe to low priority wouldn’t help, there is some mental_some_or_other_process or service that needs to be taken down to low. This is what people did before Max2014, i never bothered to remember this because when i render i want 101% of processor power to the renderer :smiley:

Regarding rendering at 100% it’s interesting how Corona works. My processor is OCed (i7 3770k @ 4.5GHz), when i render with Mental, fan works at 80% or 100% (1900RPMs) to keep it at 75°C.
But with Corona fan is at 1200RPMs and processor is around 60°C. One more point for Corona :smiley:

This is weird… maybe because of all the computing optimizations Corona is more memory-limited, which causes pipeline stalls and idle cycles to be inserted. But I would have expected the exact opposite…

I don’t know what pipeline stalls and idle cycles are, but i know that all cores are at 100% all the time. When rendering, off course.

EDIT: i’m rendering my vegetation test right now, today is a bit colder so fan is at its lowest (1100 RPMs) with all cores at 60-62°C.

Even CPU showing 100% load can actually spend most of its time just idling and waiting for data from RAM. There are functions in Corona that spend 60% of time waiting for RAM. Fortunately it is a minority :wink:

Guys!
What about put the script “Low priority.ms” in c:\Program Files\Autodesk\3ds Max 20xx\scripts\Startup\ ?

the script: sysinfo.maxpriority = #low

or in attach
Low.ms (26 Bytes)

{There are functions in Corona that spend 60% of time waiting for RAM. Fortunately it is a minority } i didn’t get this point key master can you elaborate what is minority…60% is not minor
if we could deal with this problem it will be HUGE improvement…in speed …i think???

minority (10%) of functions used spend majority of their time (60%) waiting for RAM. So in the end it is 60% from 10% which is only 6%. Give or take. The problem is more complex, but basically modern CPUs are not limited by the processing power, because it is extremely easy to just add extra adder or multiplicator to the CPU. What is limiting is waiting for data from RAM/caches, because memory just cannot keep up with 4GHz+ CPUs.

so any solution for this keymaster???

Is DDR4 a solution keymaster???

No, but if you would find one, I would be very happy, as would the rest of the world, since the CPU performance would rise about 4 times.

And crush those GPU renderers to dust like they deserve? Keymaster, better start working on it :smiley:

So if i use on Corona a ddr3 memory 1666mhz and the a memory of 2800mhz will i see A BIG change?

no, since they both have comparable latency. - the 2800MHz one will have higher timing numbers. 2800MHz is useless, if you still have to wait 50 cycles for addressing. Plus the entire path from the cache miss to interrupt to bus to ram to bus to cache takes much, much more.

No Ondra. It will change
Maybe it affect while the geometry is loading in memory to render, but even benchmark run faster with more fast memory

sure, but not by 60%