How long before GPU RT?

[tricky]

Probably been asked 1000 times before, but having seen the Siggraph presentation, we are thinking of how RT could be inserted into our production pipeline.

For us, it will make sense only when it can make use of the GPUs within a local workstation.

Do you have a date in mind?

[Thox]

This is something I would like to know also. I could persuade my boss to buy VRay RT now, but if I have to go back again in a months time asking to buy another version of the software he won't be impressed.

Any indication when VRay RT GPU will be released? next month, this year, next year?

[tricky]

I have a feeling that they will be part of the same product (with the multi-kernal version of RT being a seperate product).

[vlado]

We will start a beta of V-Ray RT for GPU in a few weeks.

Best regards,
Vlado

[tricky]

We will start a beta of V-Ray RT for GPU in a few weeks.

Best regards,
Vlado

Thanks Vlado. Will this be a beta for anybody to use, or just people who have purchased RT so far?

[vlado]

It is not yet decided; I would like it to be for everyone, but I don't have the final say on this.

Best regards,
Vlado

[GatesB]

I hope they side with an open beta. There are some players in the GPU market that have a commercial headstart (Octane), althougth Vray is a superior product, you still gotta get people to try it. And if they already have a working pipeline centered around a competing product, that can be difficult.

[posterus]

We're designing a render farm for a customer and they know they want to take advantage of V-Ray RT for GPUs. I know you can't give firm dates but the timing of this greatly affects what hardware we recommend. If I suggested to my customer that they might be able to implement RT for GPUs in say 3 months and recommend hardware accordingly, would that be way off? Any guidance you can give would help out a lot.

Also they are interested in our 8 GPU systems. I just want to confirm a few things. The product brief says that RT supports distributed rendering, is this true for the GPU version? So if they have render nodes with fx cards installed will a user be able to access those GPUs from their production workstation. Likewise if we want to use our 8 GPU systems as GPU render servers.

thanks,
Joe

[vlado]

We're designing a render farm for a customer and they know they want to take advantage of V-Ray RT for GPUs. I know you can't give firm dates but the timing of this greatly affects what hardware we recommend. If I suggested to my customer that they might be able to implement RT for GPUs in say 3 months and recommend hardware accordingly, would that be way off? Any guidance you can give would help out a lot.

V-Ray RT GPU beta is already available for all users of V-Ray RT from our web site.

Also they are interested in our 8 GPU systems. I just want to confirm a few things. The product brief says that RT supports distributed rendering, is this true for the GPU version?

Yes, this is correct. Each license of V-Ray RT allows you to use up to 10 machines for distributed rendering, no matter how many CPUs or GPUs they have.

So if they have render nodes with fx cards installed will a user be able to access those GPUs from their production workstation. Likewise if we want to use our 8 GPU systems as GPU render servers.

Yes, you can do this. So if you have 10 machines with 8 GPUs, you can use all 80 GPUs with one license of V-Ray RT GPU.

Best regards,
Vlado

[ErikJDurwoodII]

So, in theory, I could buy a server (like one from Tyan) that holds six NVIDIA Tesla C2050 cards for around $15,000 USD and have essentially 2,688 Fermi CUDA/OpenCL cores and 18GB of GPU memory amongst them and that would only constitute as one of ten possible licensed machines?

[plastic_]

So, in theory, I could buy a server (like one from Tyan) that holds six NVIDIA Tesla C2050 cards for around $15,000 USD and have essentially 2,688 Fermi CUDA/OpenCL cores and 18GB of GPU memory amongst them and that would only constitute as one of ten possible licensed machines?

GPU memory doesn't stack.

[vlado]

Yep, each GPU is completely independent from the others and must keep its own copy of the scene.

Best regards,
Vlado

[ErikJDurwoodII]

Ahhhh. I see. So ideally each individual GPU should have as much memory as possible. Maybe the pricier Tesla C2070's with 6GB of memory would be better... Of course I am fantasizing. ^__^

[Dariusz Makowski (Dadal)]

Or w8 for ati and buy shapphire 4 gb dual gpu for bargain 800£ ! ^^