CPU+GPU hybrid mode

[TheLooopCGI]

Is there a plan for future Vray GPU development to make a hybrid CPU + GPU rendering on one machine? And can it be expected sometime soon (within a year) or is it planned for some day ahead of us when all goes well and the world doesn't explode?

Right now one GPU card is quite slow compared to a modern CPU so if you really want to enjoy the speed boost you have to buy few of the cards together making it quite an expensive venture. And then the nice CPU power is just idle which makes this strange feeling in the stomach...
It would be great to just get the GPU to tag along and help the CPU with the render...

Also, AMD seems to have their APU processor ready (CPU merged with GPU in one piece). And it seems the GPU cores can add their calculation speed for any application. Would it make the CPU render considerably faster or is there still need for some special instructions or compilers? And if yes, is there a plan for development?

Thanks for answering.

[Dariusz Makowski (Dadal)]

It is possible right now. I am using it all the time in fact...

Search on in older topics, I cant really remember what topic is it but its there... U have to edit some settings in windows...

GL

[vlado]

Also, AMD seems to have their APU processor ready (CPU merged with GPU in one piece). And it seems the GPU cores can add their calculation speed for any application.

This is not true. The GPU must still be accessed in a dedicated way (OpenCL, DirectX or whatever).

Best regards,
Vlado

[ghiom3d]

It is possible right now. I am using it all the time in fact...

Search on in older topics, I cant really remember what topic is it but its there... U have to edit some settings in windows...

GL

I didn't know ! I can't find the topic .
can someone give this info here ? the topic title is really explicit for this tips .

[TheLooopCGI]

It is possible right now. I am using it all the time in fact...

Thanks for the tip. I searched the forums and found it. I'm happy to see it's developping.

This is not true. The GPU must still be accessed in a dedicated way (OpenCL, DirectX or whatever).

Hmm but then it's strange they make such a fuss about it. I mean, what's the deal? Making a new word like APU and big press announcments. If you still need to access it through OpenCL or DirectX then it's just a GPU sitting next to a CPU in one package. The same like Sandy Bridge CPUs....

[TheLooopCGI]

I didn't know ! I can't find the topic .
can someone give this info here ? the topic title is really explicit for this tips .

Yea it also gave me a bit trouble to find it. I didn't find the original topic where Vlado explains how to do that but I found a topic which already quotes Vlado so I just copy it here...

Go to Control Panel, System, Advanced, click on the "Environment Variables" button. Then click the "New..." button for the system variables; for the "Variable name" type "VRAY_OPENCL_MIXED_MODE" without the quotes; for the "Variable value" type "1" without the quotes. Click OK to close all the dialogs and then close the V-Ray render server. Next time it starts, it will attempt to run the CPU and the GPU engine in parallel. But, like I said, this is very experimental.

Best regards,
Vlado

[ghiom3d]

WIth Arion, when you use CPU + GPU , some core are recomended to not use for computing but to make the card working.
For example, with a system based on a Corei7 ( 8 core where 4 are Hyperthreaded cores ) and 2 GPU cards, it's recomended to use 6 cores intead of 8 to keep the stabilty of the system and staying away of freezes.

Is VRAY RT hybrid mode is working the same way ?

[TheLooopCGI]

As far as I know people are reporting frozen GUI of the whole OS when rendering with VRay only on GPU which doesn't take much of CPU power. I don't believe that leaving two CPU cores out of the render when going CPU + GPU solves anything as it seems more of a GPU issue. Maybe it's just an Arion issue with priorities somehow.

What I would be also interested in is if there is a way to reduce the GPU load in order to leave some power also for GUI rendering. The solution to have two cards, one for the GUI and the other one for calculations is a bit unfortunate. One doesn't want to have the card for GUI crappy because the viewport performance sucks. And if the GUI card is good then it's a pitty to not include it in the render...

[ghiom3d]

keeping some core out of rendering is to haven't any bottleneck with data traveling i think.
To make all your cards transfering data in/out and working, i thing it's use some little CPU ressources .

But i'm not an expert in anyway, so i can be wrong.

vlado ?

[ghiom3d]

But, anyway , my experience with Arion show me that an 8 core is helping nicely if you only have 1 GPU.
But with 2 GPU, the Cpu is helping less .
With 3 GPU, the CPU power is nearly invisible , because 3 GPU are very fast.
With 4 GPU, i prefere to keep my CPU to read my mails and listen music

[vlado]

An option that we are also looking into is support for OpenCL implementations on the CPU, like the Intel OpenCL driver, or the AMD one. With these, it is not a problem to use both the CPU and the GPU in a more consistent way.

Best regards,
Vlado

[ghiom3d]

Thanks Vlado for your answer , that's Sound great , and even CPU aren't competitive when using many GPU, theyre stay usefull when we already have a renderfarm .

[Alan Iglesias]

An option that we are also looking into is support for OpenCL implementations on the CPU, like the Intel OpenCL driver, or the AMD one. With these, it is not a problem to use both the CPU and the GPU in a more consistent way.

This sounds really promising!

-Alan

[ghiom3d]

After some test, i think is really not recommanded to use mixed mode at this time if we have many GPU.
When using many GPU without mixed mode, the CPU is already use a lot.
With Mixed mode, CPU is too much solicited and some bugs and crash happend more often after some hours of rendering .

On my 6 580GTX, i have seen that to manage rendering on 6 GPU, it's using many RAM and more than 60% of the CPU.

This can be different with only 1 or 2 GPU.