V16 really is faster

[Glob Studios]

I just purchased V16. In the forums there were several accounts about how it is significantly faster, and I have to admit I was a bit skeptical that it would be faster, even on something simple and straightforward. But it turns out, it is.

 

I rendered one frame of a project I am working on using the Final preset in the 1080 HD format on a snazzy machine with an Intel i7 quad core 3.4GHz processor. I rendered on 15.0j+, 16.0a 32-bit and 16.0a 64-bit. Rendering a single frame does not do any overt parallel processing, but with V16 you can use NetRender to do 2 different frames at once, so in a real project you would likely get something like double the improvement I'm describing here.

 

My scene is a planet with a simple spaceman on it. The ground is a couple of simple colored sheets with roughness and a small amount of reflectivity. There is one sun, and a sky backdrop image showing Saturn. Here's what the numbers look like:

 

Version				Seconds					Improvement
15.0+j					168						   -
16.0 32-bit			   129						  23%
16.0 64-bit			   121						  28%

 

The results were almost identical. I did notice one small difference. I have a small amount of reflectivity on the planet's surface, and a dark area reflects the backdrop image of Saturn, which is yellow colored. (The Saturn image is out of the frame.) The V16 images, which appeared to be identical, show a slightly higher amount of yellow reflection than the V15. Here's the V15 version:

 

 

And here's the V16 64-bit version:

 

 

Other than that they look the same. By dialing back the reflectivity, I'm sure I can get the same look, but to get it identical, you may have to make an adjustment.

 

Rendering two frames at once using V16 NetRender, I can render a project in 36% (a little over 1/3) of the time it takes with V15. If I can do three at once (two with NetRender and one with the A:M program), I can render the project in 24% (less than 1/4) of the time. This is a big win.

 

-Vance

[robcat2075]

Yay!

 

 

Rendering a single frame does not do any overt parallel processing, but with V16 you can use NetRender to do 2 different frames at once, so in a real project you would likely get something like double the improvement I'm describing here.

 

You should be able to render up to four at once. What does it say when you do "About" in Netrender?

[Gerry]

Interesting test, Vance! Thanks for taking the time to explore this.

[John Bigboote]

Yeah, great testimonial! Come back often!

[robcat2075]

You can also try the semi-informal-not-quite-official Three teapots benchmark scene...

 

http://www.hash.com/forums/index.php?s=&am...st&p=316705

[Glob Studios]

Yay!

 

 

Rendering a single frame does not do any overt parallel processing, but with V16 you can use NetRender to do 2 different frames at once, so in a real project you would likely get something like double the improvement I'm describing here.

 

You should be able to render up to four at once. What does it say when you do "About" in Netrender?

 

It does say 4, which is great because I have 4 cores! I had read somewhere (I think it was an old forum post, but it might have been the Hash site) that they just let you do 2, and had assumed it was correct (they may have changed their minds since then). Assuming the numbers stay the same for simultaneous running of 4 NetRender processes, V16 should render a project in 18% of the time, or 5.5 times faster than a single process V15.0j+ system. For me, this is huge!

 

-Vance

[robcat2075]

It does say 4, which is great because I have 4 cores! I had read somewhere (I think it was an old forum post, but it might have been the Hash site) that they just let you do 2, and had assumed it was correct (they may have changed their minds since then). Assuming the numbers stay the same for simultaneous running of 4 NetRender processes, V16 should render a project in 18% of the time, or 5.5 times faster than a single process V15.0j+ system. For me, this is huge!

 

-Vance

 

When using all four cores, use the Windows Task Manager to set the "priority" of one NetRender node to "below normal" so the windows interface can still grab enough CPU time when you need to do something like move or open a window.

[Glob Studios]

It does say 4, which is great because I have 4 cores! I had read somewhere (I think it was an old forum post, but it might have been the Hash site) that they just let you do 2, and had assumed it was correct (they may have changed their minds since then). Assuming the numbers stay the same for simultaneous running of 4 NetRender processes, V16 should render a project in 18% of the time, or 5.5 times faster than a single process V15.0j+ system. For me, this is huge!

 

-Vance

 

When using all four cores, use the Windows Task Manager to set the "priority" of one NetRender node to "below normal" so the windows interface can still grab enough CPU time when you need to do something like move or open a window.

 

That's a good hint to know. Thanks!

 

-Vance

[Glob Studios]

When using all four cores, use the Windows Task Manager to set the "priority" of one NetRender node to "below normal" so the windows interface can still grab enough CPU time when you need to do something like move or open a window.

I've been playing with this, and it appears that downgrading the priority is entirely unnecessary on my machine, at least for the projects I've been rendering (the most serious takes about 60 seconds per frame). As I type this, I have four NetRender slaves hard at work on the machine, and I cannot detect any change in the responsiveness of the system.

 

Windows, unlike Unix-based systems (like the Mac), has traditionally given a priority boost to the program whose window has focus. Perhaps that's what's responsible for the lack of impact.

 

I'm experimenting with seeing whether and how much having this boost (or giving it to a different program) affects render times, and whether it is better to give the render server or a render slave focus. No definitive results yet.

 

-Vance

[Glob Studios]

OK, I've got some benchmarks. The Intel i7 processor @ 3.4 GHz doing the Three Teapots benchmark (v005) in V16a NetRender takes 3:16 to do the render, while V15j+ takes 5:07, 56% longer. V16 really is significantly faster.

 

But the Three Teapot benchmark is just set up to render one frame. I ran some tests using NetRender with four NetRender messengers running on the system, with the number of frames for Three Teapot extended to 16, so each process needed to render 4 frames.

 

I first ran it keeping the Render Server window having focus. This took 15:31. Next I did the same test, but put the focus on the render slave window that was going to finish last. It took 15:34, very little difference. Finally, I minimized the NetRender windows and gave focus to a browser window. This time it took 15:51, which is about 2% longer.

 

-Vance

[mouseman]

I wonder if this setting makes any difference. (as seen on Win7 Home Premium 64bit)

[Glob Studios]

Yes. this turns the normal Windows "give a boost to the window with focus" behavior off and on. (You get to it through Control Panel's System tool -> Advanced System Settings -> Advanced tab -> Performance Settings -> Advanced tab.) I tried changing it to background, and ran the test again, with a browser window having focus, and being actively used. I noticed no reduction in responsiveness at all, as I tried a variety of browser operations. It did the 16 frames in 15:28, slightly faster than any of my previous times. I think I'll leave it that way.

 

-Vance

[mouseman]

Interesting. Maybe it's a case of once your CPU is fast enough or has enough cores, the "Programs" vs "Background services" differentiation isn't noticeable to a human.