OTOY Forums

So just to clarify then, when 2.0 is out we will be able to use both distributed (still frame) and sequence (animated) over a number of different boxes directly from inside Cinema 4D without transferring to standalone? Will the accessable GPUs show up under the c4d plugin devices tab?

Thanks

I think it's about GPUDirect that kelpler is able connect by LAN
so, I don't think it has to export whole scene to all of cards
it wouldn't make sense otherwise

ah ok, thanks for clarifying aoktar,
i thought this is a 1.5x feature already, but also for 2.0 it is VERY good news !
is 2.0 close? weeks? months? a little hint?

enigmasi wrote:I think it's about GPUDirect that kelpler is able connect by LAN
so, I don't think it has to export whole scene to all of cards
it wouldn't make sense otherwise

It's not GPUdirect,but our own implementation. We distribute render data to other computers via LAN. We try to do it as efficiently as possible (using compression, caching, differential updates etc.), but it's still quite a bit of data, so 1GBit/s Ethernet (or higher) is highly recommended. Fortunately 1GBit/s is quite cheap these days.

indexofrefraction wrote:ah ok, thanks for clarifying aoktar,
i thought this is a 1.5x feature already, but also for 2.0 it is VERY good news !
is 2.0 close? weeks? months? a little hint?

The beta is becoming more and more stable, so I think it's not months anymore, but weeks.

Excellent news - just in time for my additional GPUs!

abstrax wrote:It's not GPUdirect, but our own implementation. We distribute render data to other computers via LAN. We try to do it as efficiently as possible (using compression, caching, differential updates etc.), but it's still quite a bit of data, so 1GBit/s Ethernet (or higher) is highly recommended. Fortunately 1GBit/s is quite cheap these days.

Awesome, do you guys have any numbers for that, in terms of percentage decrease over direct PCI-E cards doing the work?

brasco wrote:

abstrax wrote:It's not GPUdirect, but our own implementation. We distribute render data to other computers via LAN. We try to do it as efficiently as possible (using compression, caching, differential updates etc.), but it's still quite a bit of data, so 1GBit/s Ethernet (or higher) is highly recommended. Fortunately 1GBit/s is quite cheap these days.

Awesome, do you guys have any numbers for that, in terms of percentage decrease over direct PCI-E cards doing the work?

After all data is transferred, the numbers add up without loss. How long the data transfer takes, depends on the scene, the number of slaves and the network. In other words: If you make big changes in geometry and render only a few samples per pixel, it doesn't work too well, but if you render hundreds or thousands of samples per pixel, it scales nicely.

abstrax wrote:After all data is transferred, the numbers add up without loss. How long the data transfer takes, depends on the scene, the number of slaves and the network. In other words: If you make big changes in geometry and render only a few samples per pixel, it doesn't work too well, but if you render hundreds or thousands of samples per pixel, it scales nicely.

Man, that is awesome to hear.
Going to have to rethink my rendernodes strategy

cheers
brasc

brasco wrote:

abstrax wrote:After all data is transferred, the numbers add up without loss. How long the data transfer takes, depends on the scene, the number of slaves and the network. In other words: If you make big changes in geometry and render only a few samples per pixel, it doesn't work too well, but if you render hundreds or thousands of samples per pixel, it scales nicely.

Man, that is awesome to hear.
Going to have to rethink my rendernodes strategy

cheers
brasc

i have to say, Network rendering is unmatched feature on Live Viewer. Also it's good for animations, changes accordingly scene content. Very good job