Here's an interesting result. I have two render nodes chewing on this image. The blocks generated by one of my machines doesn't "match" the other two. Two are home built machines and the third one which is out of sync is a cheapy emachine. All are AMD machines but are of different speeds. Weird and disappointing. -Kix

-Kix

Hmmmmm, I haven't seen this one. I'd definatley pass the file and image along to Eovia.

I am sure that both machine did not have same photons (rays) information to both machine because it may have high numbers to calculate with from this blub. It is not easy method to split photons information to network machines.

Did you make sure you had whatever plugins you have installed with carrara installed on the render node(in carrara/extension) too.

mmoir, I don't know how to install extensions on a node. Good question. Regardless, I've not installed any extensions or plug-ins on any of the machines involved. Gweb, thought of that, still don't know. I'm off to send Eovia some stuff. If this doesn't work, I'm going to be seriously bummed! -Kix

-Kix

Hi Kix- Do all the machines have the same CPU? I wonder if a difference in floating point units might cause a difference in rounding. . . Markd

The AMD 2.4 and AMD 1.4 match results. The third AMD 2.4 does not. Machine 1 = AMD 2.4, Nvidia Video, home built, "server" Machine 2 = AMD 1.4, Nvidia Video, home built, "Node 1", matches server. Machine 3 = AMD 2.4, Video unknown, emachine off the shelf, "Node 2", Results do not match. -Kix

-Kix

Do you know to any more detail what CPU is in there? You might need to check at boot time to get the full description- ie, are they Athlons? Durons? Same generation? Tbirds? Bartons? Just guessing here. ;-) Markd

I am database server programmer expert. I know the network and calculation stuff. It is the photons trust me on that. It was not your machine, it was how the data was split into 2 machines. The photons is very difficult to share data with PCs. The Master PC is supposed to handle all that and give specific photons axis data to the node PCs and it is supposed to calculate with and send the photons axis data back to master PC. Other possible theory, the other node PC was busy with a checker and other node pc missed some of the photon datas when it completed one of checkers ahead of others. The master PC was supposed to pause until all the current checkers are completed to send the photons back to master PC to redistribute it. It won't matter if you bought PCs with same speed because there are chances that one of the pc will get delayed. So I think that all the PCs need to have photons precalculated before it may start rendering with true colors on surfaces. It probably will need 2nd renderer engine for color and photons separately.

This is very interesting. When I perform the render, each machine performs the GI calculations separately (sampled lighting etc.). You can see the little progress bars scanning across the tops of each computers screens. I do think each machine is performing all of the GI calculations separately already. The network rendering areas don't commence until each machine has completed its own separate GI analysis. -Kix

-Kix

Are they finishing the GI calculations at the same time? If so, they must be getting fed the GI data from the source computer. If not (i.e. some machines start rendering while others are still calculating), then each machine is fed the scene data only and renders at its own risk. It wouldn't make sense though if data truncation is happening between networked computers. What would be the point then of having a render farm?

Message edited on: 11/01/2004 20:05

www.youtube.com/user/ShawnDriscollCG

Render-Farm is good for per frame render work.

They don't finish the GI calculations at the same time. They're completely independant. I realized there is one more significant difference in my machines which is the OS. Machine 1 = XP Pro, AMD 2.4, Nvidia Video, home built, "server" Machine 2 = XP Pro, AMD 1.4, Nvidia Video, home built, "Node 1", matches server. Machine 3 = XP Home, AMD 2.4, Video unknown, emachine off the shelf, "Node 2", Results do not match. -Kix

-Kix

You didn't get my point. Every PC collected photons from what it hit on the object in renderer. I should have said that earlier

Did you send your file to our tech support (support@eovia.com) ? If not please send me a copy and I will take a look (charles@eovia.com) Thanks Charles

Wow someone from Eovia reads our threads!! :)

Attached Link: http://www.renderosity.com/messages.ez?ForumID=106&Form.ShowMessage=1990337

Hi Charles,

He reposted related issue at http://www.renderosity.com/messages.ez?ForumID=106&Form.ShowMessage=1990337

You may go to this link to read some more of his posts.

Message edited on: 11/05/2004 01:06

I just noticed that the problem you were having, Kixum, also happens when I'm rendering a single image on a single computer and cancel it. When I start the render again in the same window (instead of a new window) without making any changes to the scene, I see different diffusion has been used. The edges of each rendered area don't match the original edges made by the previous render. It looks like the seed used to start the random numbering to create diffusion for a scene is not the same value for each render. So, of course, each computer node will have its own seed value to start its rendering of the scene when using multi-computers. Carrara Pro 4 needs the host computer to send the same seed value it's using to each node so they have the same calculations for global illumation before they start their renders. It would be better if Carrara didn't use a different seed each time the render button got pressed, as well. It needs a user setting that chooses to auto set/change seed values for a scene. So far, the best way to remove the checkerboard effect in renders is to turn on Interpolation Precision to 100%.

www.youtube.com/user/ShawnDriscollCG