uh ahh just curious… since the size of render doesn’t matter I just wonder how will the vray GPU use the GPU memory…
Basically how much we need for 10 mil scene ?
How much for 4x4/8x8/10x10 textures…
Does jpg/png/tiff etc etc matter?
Any plans about some kind of virtual memory for gpus? If it get full, gpu will just flush and repack memory to continue render?
I’ve not much idea how the technology works, but I would imagine a huge part of the mental speed of RT GPU is the fact that it can all be loaded onto the graphics board and handled in one place. Flushing it backwards and forwards to RAM/hard disk will slow the process down a lot.
The output image size does not matter, but the number of triangles and textures do. If the geometry is too heavy, or the textures do not fit on the GPU, the scene will not render. As far as number of triangles go, it is about 100 MB for 1 million triangles, but this is a very rough approximation.
Thanx Vlado ! That souds like gtx480 should do the job for me… what about textures? Any idea how much they will take the memory… ? Also does tiling textures matter? And whats happenin to noise/gradients etc etc they need space in gpu too ?
Only bitmaps work in this version, and there is a setting to specify their resolution when uploaded on the GPU. Tiling does not matter (it does not take additional memory).
Reading these posts, I’m curios to see if a realtime solution like VrayRT can handle and run properly a scene already optimized for a standard realtime engine like UDK or Unity with the plus of raytracing reflections/refractions and shadows.
Did anyone make a test??
Vlado, do you think it’s long to come the days for VrayRT used for a videogame(or another realtime application)??
I have feeling u should ask that question to Game engine developers than renderers I guess at some point it will change to game… Look at Crytek/Crysis their engines are hardly games… theyre pushing more towards actual rendering than games hehe
Maybe you should check out the “Sapphire Radeon HD 5970 Toxic 4GB”, as supposed to the 1,5 gig of RAM on the Nvidia GTX 480 cards. Maybe a good option if GPU RAM is a issue?!
Oh god :o that GPU should work well in that case I guess… but isnt it 2x2 gb as it is dual GPU? Wondering if vray would use it as 2x2 gb or entire 4 gb humhhh
If you want to make use of GPU render I’d stick with nVidia as most developers are using CUDA and not OpenCL since I hear it is still quite in it’s infancy? Maybe I hear wrong, but that would be my suggestion. nVidia cards would support both correct?
I read the specs on the ATI SDK 2.2 and it supports OpenCL 1.1. I really hope these OpenCL drivers for ATI are mature, because they offer better GPU card options than nVidia.
About GPU rendering. it uses GPU mhz to render… so we should look for gpu that got more cores&more mghz? If yes then Ati with their 3200(right?) cores > nvidia… or how should I look at CGI cards ?
Also Any luck vlado with new 2.2 SDK ?
And also doest moddificators take more memory in GPU ? Noise/twist/bend etc etc?
How does VRay RT GPU work with multiple cards in SLI mode? Does it see them as one…like you get the cumulative amount of memory and processing? Will the memory hit be spread across the cards? I saw the demo videos but i’m still a bit unsure about how this works…and how i should proceed with future hardware builds. Would love to build a couple machines using this motherboard:
SLI dont work. U actually have to disable SLI to make it work and no u dont share memory either… each GPU have its own copy of scene inside it so its a bit fail…
And this motherboard is a bit pointless for you to be honest… all graphics card are 2/3 slot big… better go with cheaper mobo but with more space between each slot… / its dual CPU.. I changed my mind Go for it where can I buy it in UK ?
For cheap render nodes ill go with Shapphire 4 gb ATI… its dual gpu on 1 card which I believe should give u 4 gb(biggest so far for cheap price) of shared memory for 2 gpus… (3x6 gpus in 3 pcie mobo)
This would depend on the drivers really. Currently, the OpenCL driver sees both cards as separate devices, so sharing the memory is not possible. This means that the entire scene must fit on each GPU.