baker
About shared GI and animation
What is the correct way to setup shared GI if it is to be distrubuted to render on say 5 machines? (for cases where only the camera moves)
Here are a list of methods that i'm going to try but just need some advice before doing so.
1.
If it is set to "Both" mode, which means each machines after rendering an alternate frames, will save the additional FG data onto the GI file, so here is the question, what if both or more machines are writing the same GI file (which is shared by all) at the same time? will the GI file be messed?
2.
If set to "Load", then which mean i'll need a fairly good GI file with most of the FG already cache say every 25 frames or render a large bakemap so most of the visible FG will be stored, in order not have to recalculate the FG for everyframes?
3.
What about duplicating the GI file for each machine having their own, and mode set to 'both'. Which means each machines will have their own shared GI within the set frame steps. But then there will be redundent calculation of FG from the neighbouring frames that will increase the render time more that neccesary right?
pc 1 render frames : 1 6 11 16 21 26 ... sharing it's own GI file
pc 2 render frames : 2 7 12 17 22 27 ... sharing it's own GI file
pc 3 render frames : 3 8 13 18 23 28... sharing it's own GI file
pc 4 render frames : 4 9 14 19 24 29 ... sharing it's own GI file
pc 5 render frames : 5 10 15 20 25 30... sharing it's own GI file
Is this an effiecient setup?
or is it better to have each machines render in a continuous time chunk, say pc1 do frame 1-50, pc 2 : 51-100, pc 3 : 101-150... etc etc?
need some advice here. Thanks
and Eddy, I'm very happy with 1.7
Here are a list of methods that i'm going to try but just need some advice before doing so.
1.
If it is set to "Both" mode, which means each machines after rendering an alternate frames, will save the additional FG data onto the GI file, so here is the question, what if both or more machines are writing the same GI file (which is shared by all) at the same time? will the GI file be messed?
2.
If set to "Load", then which mean i'll need a fairly good GI file with most of the FG already cache say every 25 frames or render a large bakemap so most of the visible FG will be stored, in order not have to recalculate the FG for everyframes?
3.
What about duplicating the GI file for each machine having their own, and mode set to 'both'. Which means each machines will have their own shared GI within the set frame steps. But then there will be redundent calculation of FG from the neighbouring frames that will increase the render time more that neccesary right?
pc 1 render frames : 1 6 11 16 21 26 ... sharing it's own GI file
pc 2 render frames : 2 7 12 17 22 27 ... sharing it's own GI file
pc 3 render frames : 3 8 13 18 23 28... sharing it's own GI file
pc 4 render frames : 4 9 14 19 24 29 ... sharing it's own GI file
pc 5 render frames : 5 10 15 20 25 30... sharing it's own GI file
Is this an effiecient setup?
or is it better to have each machines render in a continuous time chunk, say pc1 do frame 1-50, pc 2 : 51-100, pc 3 : 101-150... etc etc?
need some advice here. Thanks
and Eddy, I'm very happy with 1.7
Re: About shared GI and animation
Well kray seems to keep GI file in memory untill you stop rendering and then it dumps it to a file. This means that if you're computers are rendering consequtive frames they won't realy share GI data untill you stop and restart render.
So IMHO I would proceed this way:
First render a frame or two with "Both" on single computer. This way you'll get GI file with irradiance data and a little bit of FG data. Then use that GI file on every computer and set each computer to a different frame range (pc1: 0-100, pc2 101-200...)
This will cause a little bit of extra time for first frame on each comp but then it will render just about the same speed as it would on one comp. It is neccesary to share the first GI file so you don't get sudden light changes because of different irradiance computation.
So IMHO I would proceed this way:
First render a frame or two with "Both" on single computer. This way you'll get GI file with irradiance data and a little bit of FG data. Then use that GI file on every computer and set each computer to a different frame range (pc1: 0-100, pc2 101-200...)
This will cause a little bit of extra time for first frame on each comp but then it will render just about the same speed as it would on one comp. It is neccesary to share the first GI file so you don't get sudden light changes because of different irradiance computation.
- Jure
I was just checking this Vray fly-through tutorial and I got another idea about setting up Animation on render farm.
There I read that the guy rendered irradiance map on single computer for only every 5th step because camera was moving slow. Then he used this saved GI file for all other computers but used only "From file" mode.
This could be translated to Kray by calculating every 5th frame with "Both" mode selected on single computer. This will create GI file with enough GI data which can be used on other computers then. You send this GI file to every computer on the network and use "Load" mode. This way network computers will render remaining of the animation very fast because all the GI data is allready stored.
Well that's the theory at least.
Let me know if someone tries that...
There I read that the guy rendered irradiance map on single computer for only every 5th step because camera was moving slow. Then he used this saved GI file for all other computers but used only "From file" mode.
This could be translated to Kray by calculating every 5th frame with "Both" mode selected on single computer. This will create GI file with enough GI data which can be used on other computers then. You send this GI file to every computer on the network and use "Load" mode. This way network computers will render remaining of the animation very fast because all the GI data is allready stored.
Well that's the theory at least.
Let me know if someone tries that...
- Jure