GTX 1080 and vray RT

[glorybound]

This is very confusing to me. Some say that GPU's stack and others say that it doesn't. All your scene has to fit in one GPU. I can have ten Titan's, but it'll only actually use only one. I guess using DR might allow you to use more GPU's.

[mitviz]

This is very confusing to me. Some say that GPU's stack and others say that it doesn't. All your scene has to fit in one GPU. I can have ten Titan's, but it'll only actually use only one. I guess using DR might allow you to use more GPU's.

no, no matter how many gpus you have if its say 5 titans and each has 12 gigs of vram, the scene will only use 12 gigs of vram but each gpu has a certain speed, so that is added, so two gpus is twice as fast as one gpu, if you have 5 then you have 5 times the speed but your scene can only use 12 gigs of vram, thats your textures and geometry that gets loaded into vram, like regular ram, you can run out if the scene used more than the gpu offers-in this case 12 gigs but i wish they found a way to add gpu vram, that would be amazing, as of now it doesnt exist, if it did, then i think most people would abandon cpu rendering all together

[Sbrusse]

I finally understand your confusion Bobby.
Look at it this way :
Each GPU can be considered as a computer on it own, it has a CPU and Ram.
Your computer that contains the GPU act like a DR master, it feeds each GPU the data that it needs to render.
Exactly the same way you would have your computer with 5 render nodes, your computer tells each of the node (containing CPU(s) and Ram) what to render.
So the same way as normal CPU DR, the scene that will render needs to stick inside the ram of the nodes but the more nodes, the faster (nearly linear).
GPU acts the exact same way, the scene needs to live inside each GPU that has it's own ram, and calculate on the processor if the GPU. Your computer only tells the GPU what to render, the same way you would DR with normal cpu render.
If you have 5 nodes with 32Gb of ram, you don'r consider you can render scenes up to 160Gb, no, you can render scenes up to 32, but on 5 different computers so it's 5x times faster.
With GPUs it's the same thing, I have 5x Titan X 12Gb, so I can't render scenes up to 60Gb but only to 12Gb, only 5x times faster because I have 5 cards (or computers) inside my workstation.

Ram doesn't stack, but speed is multiplied by the number of cards

Does this make more sense?

[glorybound]

Okay, that makes sense. What happens if you have a mixed batch of cards? Is your smallest card the most you can render on the GPU? I have a GTX980ti and a GTX 1080 coming today. The GTX980ti has 6GB and the 1080 has 8GB. My GTX980ti will also power my 4K, so I am not even sure how much ram remains. If there is only 4GB left on the GTX980ti than I wasted the money on the 1080 and another GTX980ti, for 1/2 the price, would have made more sense. Heck, I could have gotten two more GTX980ti's. Actually, my motherboard only supports two cards, I think. Maybe, in a few months, I can take the GTX980ti out and replace with a second 1080.

[Sbrusse]

See it the same way as I described above, you have a card with 4Gb, one with 6Gb and one with 8Gb.
What would you do if you had a CPU render node with 8Gb, one with 16 and one with 64Gb, you would be able to render scenes of 8Gb on the 3 computers and a scene who takes 50 Gb only with the 64Gb node.

Same here, depending on your scenes, you check or uncheck the cards that you will use to render, if it's a small scenes that takes less than 4 Gb, use all the cards, if it's more, then you only select the cards that can contain the scene inside it's ram

It's easier to have all the same amount of ram so you're not limiting yourself to either the scene size or the number of cards you can use, but the exact same way you would manage CPU nodes.

I have 3x 970 of 3.5Gb and 1 980Ti on my workstation at home, depending on the scene size, I can either use 4 cards or 1.
At the office, I have 5 Titan X 12Gb so it makes it easier, I can either use all of them or none.
But as always, the only limited factor is if the scene fits inside the ram of each specific card, if it does, you can use it, otherwise, you can't

[glorybound]

That was my biggest headache with DR on CPU. I had several machines with 32GB of RAM, which back then was massive, but the others had 12GB. The 12GB machines took so long on their buckets, it made no sense to use them.

[glorybound]

Is anyone using one of these GPU boxes?

https://developer.nvidia.com/devbox

How nice would it be to buy a box that can house 10 cards and you just keep adding when you have the funds? 10 - 24GB cards would be the cats peow.

[Sbrusse]

They are good, but way over prices compared to what you could build by hand.
Nowadays it's piece if cake to build nodes with up to 7 cards for bottom low budgets compared to the NVidia VCA or devbox.

Just add more nodes and done

I would be able to build a 7x 1080 rig for 7500 euro

[glorybound]

How about those GPU cloud services?

[Sbrusse]

Great a one time thing when you need ponctual burst and when you can spend time on IT as you need to config everything yourself usually but not great for every days wip

[oren]

im so sad.... no one in the world can tell me how the gtx 1080 perform with vray rt .... even on active shade... even on simple interior.... something.... helllooo

[Sbrusse]

https://www.facebook.com/groups/VRay...9004246585485/

[glorybound]

It has arrived!

[mitviz]

It has arrived!

[ATTACH=CONFIG]33098[/ATTACH]

nice! what version of the 1080 is this?

[oren]

we gonna see vray rt tests soon?