CPU output compare test

[handige_harrie]

Forgot the example again :D

[LucentDreams]

Originally posted by handige_harrie
Forgot the example again :D

the test won't be accurate at all though, have you ever rendered a radiosity image twice without saving ther solution (recalculate always) then take the two images into PS and blen one on top of the other with difference, man thres a difference and thats using the exact same system seconds after each other.

[handige_harrie]

Originally posted by Kaiskai
the test won't be accurate at all though, have you ever rendered a radiosity image twice without saving ther solution (recalculate always) then take the two images into PS and blen one on top of the other with difference, man thres a difference and thats using the exact same system seconds after each other.

Yes I know, I tried :). Of course there are differences, just not as rigorous that it's disturbing in an animation.

Maybe it is disturbing if you rendered it with a different cpu, and that's what I want to find out.

[bcbarnes]

Would be glad to help. Can you provide the scene as a .ZIP file?

[handige_harrie]

the test won't be accurate at all though, have you ever rendered a radiosity image twice without saving ther solution (recalculate always) then take the two images into PS and blen one on top of the other with difference, man thres a difference and thats using the exact same system seconds after each other.

I tried again using the blending option set to difference as you propose, the image is perfect black. At least I cannot find artifacts or such, even when heavily zoomed in.

Would be glad to help. Can you provide the scene as a .ZIP file?

I'll replace the .rar with the .zip in a few minutes. Although I like winrar better, most people use winzip, I should have thought of that.

[bcbarnes]

Here's one for you. This was done on an AMD Athlon XP 1.20 Ghz (Thunderbird, I think). Took 14:36 to render.

Looks to me like the light is a bit "blown out" in my image - it doesn't have the nicer shadowing around the glass object on the right.

AthlonXP Thunderbird Image (http://www.briancbarnes.homestead.com/files/AthlonXP.zip)

[AdamT]

http://bellsouthpwp.net/A/d/AdamTrachtenberg/comparo.zip

On dual 2.4Ghz Xeons, about 6 1/2 minutes.

BTW, never use an omni light for caustics. The attached image was done with the original scene, but changing it to a spot light reduced render time by about 40%, and with a better caustic effect.

[handige_harrie]

This is strange, AdamT's and bcbarnes' image look almost identical. Mine however looks a lot different. I must have done something wrong. I even clocked down to 2000mHz and rerendered a couple times, still the same. I downloaded my own file - go figure - and rendered it. Still the same.

I am confused and don't really know what to do right now :shrug:

Maybe you should make a new file Adam, you know much more what should and what shouldn't be done to keep the rendertime down.

[AdamT]

Have you upgraded to 8.1? That might make a difference.

To reduce render time all you have to do is convert the light to a spot and make sure it covers the whole scene. The thing is, an omni casts photons (and shadows) in 6 directions, so when you use it for caustics in this case, 5 out of those six calculations aren't affecting the rendered image, but they slow down the render. Also, you're getting 1/6th the photons where they count. To speed things up even more, use two spots: one to light the scene and another in the same place to generate caustics. The caustic light should be tightly focused on the causitc-producing object. That way *all* the photons are going where they're needed and you can get away with much fewer samples.

[dann_stubbs]

actually procedural textures are where you will probably see the biggest rounding difference errors.

you may also want to throw in some area lights with falloff and soft shadows.

and then of course there is the mathmatically oriented plugins like pyro or anything else math intensive. dynamics, etc...

just more variables to explore - but procedurals should be tested for sure.

dann

[handige_harrie]

I'm still on 8.0 that'll be the problem.

I have to upgrade one day soon, I know. Especially if I want this 'distributed rendering' thing to be a succes.

Adam, I immediately searched the manual and found what you are saying about the omni that emmits light (and photons) in all directions :).

So dann, you're saying I should toss out the radiosity and caustics, or just leave them in and add the things you suggest?

[CosmicBear]

well, here's your mac-test

rendered on a dual g4 2x1GHz, 1 GRam, Cinema 8.1

it took 15.15 minutes

hope it helps

zip-file (http://www.cosmicbear.de/download/rendertest.zip)

[AdamT]

You know the 8.1 update is free, right?

[handige_harrie]

Originally posted by AdamT
You know the 8.1 update is free, right?

Yes I do :). It's just that I have had very bad experiences with upgrading software. I read there were problems upgrading to 8.1 so I say to myself: "If it ain't broke, don't fix it". As long as I don't need the extra features I don't really want to upgrade.

But that's just me :rolleyes:.

I compared the amd, intel and mac pictures and could only see differences if I cranked the contrast up to > 90% in PS (after the difference blend on the second layer, and flattening the image).

You can check it yourself if you want.

[LucentDreams]

threy released a new updater that fixed all the problems. The OGL and edge melt are enough reason to upgrade.

[dann_stubbs]

Originally posted by handige_harrie
I'm still on 8.0 that'll be the problem.

I have to upgrade one day soon, I know. Especially if I want this 'distributed rendering' thing to be a succes.

Adam, I immediately searched the manual and found what you are saying about the omni that emmits light (and photons) in all directions :).

So dann, you're saying I should toss out the radiosity and caustics, or just leave them in and add the things you suggest?

for raydiosity i would leave the omni in (esp if you have more then 1 bounce) it should help add some light not all from one point as a spot is.

i would add those things in addition - after all you want a "worst case' scene to test issue with. then you can see the variations at once and know how to avoid or work around them.

maybe even put something behind the transparent shape and up the refraction to 1.6-ish (glass like) and see if the different processors produce variation in refraction calculation. just add a tag to ignore GI on the glass object that should help render times a little too.

i'd even put the scene in a box so you see more shading in the background walls and sides (good with the omni light raydiosity too) and you might see variations in the math from the drop off on the light creating differences in shading/shadows on the walls.

if you are looking for errors - look for them all - then you know how to avoid some or all of them.

dann

[handige_harrie]

I uploaded a new file (check first post for download) which includes:

3 different SLA materials
soft shadow omni w/ falloff
soft shadow spot w/ caustics, without falloff
soft shadow area light w/ fallof
separate glassy thing mesh for radiosity.
glassy thing itself does not produce radiosity.
radiosity
caustics

No dynamics or such.

[Cinema1954]

Here's my render of the new file. Athlon xp1800+, 512 RAM. Render time 17:47. http://www.camerasinternational.com/skinner/AthlonXP1800.zip

[AdamT]

http://bellsouthpwp.net/A/d/AdamTrachtenberg/cpu_test.jpg

6:35 on dual 2.4 Xeons.

[CGTalk Moderation]

This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.