Workstation AND Render Farm

[Guest dialog]

Hey guys, so I finally get a new workstation.

We have a deal with Dell, so we need to stick with this configuration and not go to BOXX.

 

Work is thinking about 2 Xeon processors (not sure exact specs), going with the maximus set up...either the Quadro K4000 or K5000 paired with a Tesla card. Quadro obviously for viewport and the Tesla to start pumping stuff out with VRay RT (which I hear will be developing fairly nicely here in the future...correct me if I am wrong). This set up I think will streamline the workflow at my workstation but will not solve our issue of having 5-10 renders a day, some hitting 6000px.

 

That being said, Is there a good render solution for a farm out there or a set up that people are using that they find works well? Is there a way to use 3DS Max with VRay and avoid using backburner? Should we just load up on CPU with a few rack mounted solutions?

 

Any feedback on this would be great guys! Not too familiar with the render farm side of things.

[Tommy L]

Hey guys, so I finally get a new workstation.

We have a deal with Dell, so we need to stick with this configuration and not go to BOXX.

 

Work is thinking about 2 Xeon processors (not sure exact specs), going with the maximus set up...either the Quadro K4000 or K5000 paired with a Tesla card. Quadro obviously for viewport and the Tesla to start pumping stuff out with VRay RT (which I hear will be developing fairly nicely here in the future...correct me if I am wrong). This set up I think will streamline the workflow at my workstation but will not solve our issue of having 5-10 renders a day, some hitting 6000px.

 

That being said, Is there a good render solution for a farm out there or a set up that people are using that they find works well? Is there a way to use 3DS Max with VRay and avoid using backburner? Should we just load up on CPU with a few rack mounted solutions?

 

Any feedback on this would be great guys! Not too familiar with the render farm side of things.

 

5-10 x 6k renders = render farm. The xeons will be a waste of money (unless the money is not yours) go for an i7 and spend the savings on offloading the render tasks to some dedicated machines.

[Guest dialog]

5-10 x 6k renders = render farm. The xeons will be a waste of money (unless the money is not yours) go for an i7 and spend the savings on offloading the render tasks to some dedicated machines.

 

Ya the money is not mine. The company has allocated a budget for my workstation and a separate budget to figure out our render farm solution, which our I.T and myself are not sure what the best approach to it will be. So for now the workstation will be used as a temporary solution for rendering until we figure out a farm solution.

 

Currently we run 2 machines with single xeon in each to do our production stuff and it is obviously inefficient for our needs.

 

If you have a redner farm solution you use or know of that works well, that would be great!

[Tommy L]

If you dont have any actual workflow issues (max operates ok, your post work is smooth) then dont upgrade the workstation at all. Get nodes. Every time you hit render, even for tests, your DR will kick in and pull power from the nodes. Either that or just stick a new GPU in your current machine (if you must use RT) and buy nodes also. Tying your workstation up with render tasks is like tying one hand behind your back for half the day, its stupid. Tempting to think a powerful machine will make your more productive, but it doesnt really. The two bottle necks are usually the user and the render time (no offense) so a big GPU and CPU under the hood doesnt help. Remote power at render time does.

[Guest dialog]

If you dont have any actual workflow issues (max operates ok, your post work is smooth) then dont upgrade the workstation at all. Get nodes. Every time you hit render, even for tests, your DR will kick in and pull power from the nodes. Either that or just stick a new GPU in your current machine (if you must use RT) and buy nodes also. Tying your workstation up with render tasks is like tying one hand behind your back for half the day, its stupid. Tempting to think a powerful machine will make your more productive, but it doesnt really. The two bottle necks are usually the user and the render time (no offense) so a big GPU and CPU under the hood doesnt help. Remote power at render time does.

 

Workflow issues are probably %50 of the time right now. Files are so big and we have lots of graphics issues.

 

So as I am not familiar at all with the nodes...what would we buy and how would that get set up to pull power from?

[Tommy L]

Workflow issues are probably %50 of the time right now. Files are so big and we have lots of graphics issues.

 

So as I am not familiar at all with the nodes...what would we buy and how would that get set up to pull power from?

 

If you set up Distributed Rendering correctly in Max/Vray then every time you hit render, any registered machines on the network will contribute their CPU to the rendering. You'll see one extra bucket on your frame for every physical core, or two extra per physical hyper-threaded core.

Sounds like you just do still images, not animations, so do some reading on DR, its a game changer.

I have 7 x i7 nodes and 2 x i7 workstations, so when everything pitches in on a frame I see 72 buckets, equivalent of around a 240gHz processor.

If I render locally its super fast so I dont waste much time. If I sumbit to the farm then I dont waste ANY time cos I can just keep on working and the farm node uses the DR to pull in the other machines and they all work together.

[Guest dialog]

If you set up Distributed Rendering correctly in Max/Vray then every time you hit render, any registered machines on the network will contribute their CPU to the rendering. You'll see one extra bucket on your frame for every physical core, or two extra per physical hyper-threaded core.

Sounds like you just do still images, not animations, so do some reading on DR, its a game changer.

I have 7 x i7 nodes and 2 x i7 workstations, so when everything pitches in on a frame I see 72 buckets, equivalent of around a 240gHz processor.

If I render locally its super fast so I dont waste much time. If I sumbit to the farm then I dont waste ANY time cos I can just keep on working and the farm node uses the DR to pull in the other machines and they all work together.

 

Ya we can't seem to get around our network issues with the DR. The computer can see the IP address, but will not connect to it. Not sure if it is an issue with the security on our network. We do animations as well yes.

 

So I assume the nodes are the way to go for a farm.

[Tommy L]

Ya we can't seem to get around our network issues with the DR. The computer can see the IP address, but will not connect to it. Not sure if it is an issue with the security on our network. We do animations as well yes.

 

So I assume the nodes are the way to go for a farm.

 

Farm is just slang for a bunch of nodes. A render node is a machine dedicated to rendering. Spending alot of moeny because you cant get DR to work seems a bit siily, just allocate a day to get it working its not that complicated.

[Guest dialog]

Farm is just slang for a bunch of nodes. A render node is a machine dedicated to rendering. Spending alot of moeny because you cant get DR to work seems a bit siily, just allocate a day to get it working its not that complicated.

 

Sounds good. For the farm solution our BIM manager has looked into Dell and some of their blade servers so we can add them as we start to build the farm.

[Jason Matthews]

I use DR for all my jobs. The way I made mine work was to put all machines on the same workgroup (Windows 7). Then make sure the central repository for all jobs, textures, etc. is accessible from all nodes. Then make sure the machine that carries the Vray license is seen by all nodes. Then Register VR Spawner as Service on all machines(including workstation). Make sure you right click on it and launch as administrator (Windows 7). Then launch VR Spawner on all machines (including workstation), again as administrator. Then go to the DR section of Vray and add your nodes. Hit render and see what happens. I advise starting with a scene that has enough geometry but has no maps assigned. If you can get that to work then assign a few textures to the scene and see if any of the nodes render without the textures. If they do then those are not seeing the texture maps.

 

I also want to stress the importance of paths. All paths should be relative paths not absolute. In other words, the paths should not be C:\folder\folder\textures. The path should say something like ..\\network\folder\folder\textures. You can do this in the asset browser of max for the entire project. To the right of the file it should say network path.

 

As for your computer setup, getting the existing computers in your network setup to render DR should be priority number one, IMHO. Once you can get them setup, then look to add nodes. If you are doing this via Xeons, then go ahead and start putting them on racks to reduce space.

[Guest dialog]

I use DR for all my jobs. The way I made mine work was to put all machines on the same workgroup (Windows 7). Then make sure the central repository for all jobs, textures, etc. is accessible from all nodes. Then make sure the machine that carries the Vray license is seen by all nodes. Then Register VR Spawner as Service on all machines(including workstation). Make sure you right click on it and launch as administrator (Windows 7). Then launch VR Spawner on all machines (including workstation), again as administrator. Then go to the DR section of Vray and add your nodes. Hit render and see what happens. I advise starting with a scene that has enough geometry but has no maps assigned. If you can get that to work then assign a few textures to the scene and see if any of the nodes render without the textures. If they do then those are not seeing the texture maps.

 

I also want to stress the importance of paths. All paths should be relative paths not absolute. In other words, the paths should not be C:\folder\folder\textures. The path should say something like ..\\network\folder\folder\textures. You can do this in the asset browser of max for the entire project. To the right of the file it should say network path.

 

As for your computer setup, getting the existing computers in your network setup to render DR should be priority number one, IMHO. Once you can get them setup, then look to add nodes. If you are doing this via Xeons, then go ahead and start putting them on racks to reduce space.

 

 

Thanks for the tips Jason! Once we get these things purchased, it will be interesting to get it all set up and going!