V-Ray supporting custom render hardware

[Metin_7]

Hi guys,

I am planning to buy a new computer system next year and am wondering if there is already something like custom 3D render hardware that can take over 3D render calculations from the main CPU, maybe by using some kind of plug-in or library that redirects render calculations to a custom GPU card or so? And if that's available, can such a card be used to speed up V-Ray rendertimes?

Thanks and cheers,

Metin

[joconnell]

Nope and I wouldn't think that Vlado would go down that route either - the art vps cards do acceleration of their own renderer and when they realized they weren't selling enough they started supporting mental ray instead. If you use hardware acceleration you're limiting yourself to the software that supports it instead of just spending the money on a cpu that your entire machine benefits from. Gelato gets a big kick from nvidia quadro cards but again you only get the benefit in gelato - one of the Art pure rendering cards cost around €1000 which you'd build a cheap core duo machine for.

I could be wrong but I think that linking vray to hardware would make it more difficult to port over to different platforms and also by the times it was done and tested properly, cpus would have caught up and removed the benefit anyway.

[Metin_7]

Thanks very much for your reply, very interesting. You've got a good point that custom rendering hardware would indeed require separate rendering software testing.

[Clifton Santiago]

I second what joconnel said. When I started at my current job a couple years ago, I was given two old workstations with a couple art VPS pure cards in them. While it was a fun exercise to learn their proprietary version of render man, I decided they were crap because you could only use their software, which obviously means no Vray. I even had the art VPS sales rep give me a call and tell me their software and Pure cards were superior to Vray. Ha! I gave him an earful... And as joconnell mentioned, they have switched platforms to mental ray for their new render thing, and have discontinued their Pure cards completely. Pretty poor move if they thought they were so superior.

I am curious about Gelato though because I have an Nvida Quadro on my new workstation, but their free trial version doesn't work properly.

[Metin_7]

Still, the specialized rendering GPU revolution is slowly dawning, check this out for example:

http://ati.amd.com/products/streamprocessor/specs.html

I wonder what Vlado's opinion is about the possibility of utilizing this.

[Clifton Santiago]

Oh, I guess I misunderstood your premise. You mean GPU hardware renderers, not just proprietary hardware renderers like the Pure cards. GPU accelerated renderers like Gelato or whatever are definitely the way forward... Certainly isn't real-time photo-real renderings the holy grail of our industry? But I always assumed that a properly evolving software like Vray would cross into this area when the time was appropriate.

[Metin_7]

I guess I've become a little bit confused about the different hardware solutions, but what I'm looking forward to is one general standard in specialized render calculation hardware, so you can buy a relatively average PC and turn it into a mean render machine by adding a super GPU and a pleasant amount of RAM. But V-Ray and other renderers should support such hardware first, before it can become a mainstream 3D rendering solution.

[mike.edel]

the problem with this approach is the amount of data that's shoveled over the bus of the graphics card

the internal data rates of the card are around 86gigabytes per second (the gpu has a special bus connection to the video memory)
so if your scene fits into the video memory it is verrrrry fast
thats's what both nvidia and ati are planing to use for calculations on graphics cards:

you upload your data to the card, let the fast gpu (it is way faster at sequentially processing data than the cpu) do the job and download the data through the pcie bus

as soon as you have to constantly pipe data throught the pcie bus (about 16gb per second max theoretical transfer limit) it gets slow

and of course the application will have to be customized for the gpu since its processor design is radically different from cpu's
interesting nontheless, for example for physics simulation (havok fx is supposed to run on gpu)
both companies already announced ideas for general-use addon cards that aim at these markets
(see http://developer.nvidia.com/object/cuda.html which is supposed to be able to run c code)

btw: i saw a live demo of gelato 1 and wasn't impressed with its speed - it was by any means not fully running on gpu - they were using gpu calculations for parts of the pipeline
vray is more flexible and faster *g*

[Paul Oblomov]

Cuda is very interesting thing. I think there'll be smth similar from ATI.