Optix 2.6 GPU paging

[Dariusz Makowski (Dadal)]

Heya

Looks like nvidia has released a while back this tool kit that can allow go over GPU memory limits and render big scenes on small GPU boards.

Will Vray get it ? when ?

[vlado]

Will Vray get it ? when ?

Yes; soon.

Best regards,
Vlado

[nikand]

this sounds amazing...will it only work for tesla and quadro's?

[saudade]

I think that the next GTX generation will have a different K20 processor and will perform a little better than a GK104.

This means that the new Geforce will be useless for Cuda raytracing.

We will never see a new GTX580.

We have to spend $3,199 for a single K20 Tesla and $2,249 for drive 3ds Max/Maya graphics.

Total: $5,448.

Nvidia is driving in the wrong direction.

[Alan Iglesias]

...Nvidia is driving in the wrong direction.

Which is what I tried to get across to nVidia's Greg Estes at Siggraph. Unfortunately he didn't have much positive (from our point of view) to say.

Basically he told me that the GTX team couldn't care less about our GPU rendering, the fermi platform being a highly efficient renderer for RT/GPU was basically an accident, and we woiuld have to buy Teslas and Quadros for that in the future. I just asked him to price them somewhere reasonable so I can reccomend them to clients based on price/performance... And told him how hard it is to (honestly) sell very expensive boards to folks when the relative performance is not there.

He promised they would do their best, but I'm not very hopeful.

In my experience, the only thing that could really change this is competition, so I spoke to the AMD director of sales as well. He seemed to think that AMD should really be a big part of all of this, and promised to give the development side a push in this direction. Once again, no big hopes but I did my best to give them a view from "trenches", so-to-speak.

-Alan

[sandykoufax]

Cheer up, AMD!

[Dariusz Makowski (Dadal)]

Which is what I tried to get across to nVidia's Greg Estes at Siggraph. Unfortunately he didn't have much positive (from our point of view) to say.

Basically he told me that the GTX team couldn't care less about our GPU rendering, the fermi platform being a highly efficient renderer for RT/GPU was basically an accident, and we woiuld have to buy Teslas and Quadros for that in the future. I just asked him to price them somewhere reasonable so I can reccomend them to clients based on price/performance... And told him how hard it is to (honestly) sell very expensive boards to folks when the relative performance is not there.

He promised they would do their best, but I'm not very hopeful.

In my experience, the only thing that could really change this is competition, so I spoke to the AMD director of sales as well. He seemed to think that AMD should really be a big part of all of this, and promised to give the development side a push in this direction. Once again, no big hopes but I did my best to give them a view from "trenches", so-to-speak.

-Alan

Heya

Thanks for info and your hard work !

Also thanks Vlado for letting us know on this one. I also saw Optix 3.0 coming out and so on so it seems that we will crack on big scenes with small GPU's soon.

In any case Alan what about making survey ? Getting some people sign up petition to Nvidia and so on ? It looks like you has connectionzzzzz ! :]

[vlado]

Actually the geometry is not the biggest problem; the GPU RAM can handle millions of triangles even now (and of course much much more with instancing); but the textures are a real problem - production scenes often have gigabytes and gigabytes of textures.

Best regards,
Vlado

[Dariusz Makowski (Dadal)]

awww thats bad. We cant page textures?

In any case for the automotive work we do currently and even my personal projects I could use GPU. There is not much textures... Unless I start making backgrounds. But thats another story

[instinct]

awww thats bad. We cant page textures?

In any case for the automotive work we do currently and even my personal projects I could use GPU. There is not much textures... Unless I start making backgrounds. But thats another story

In case for the automotive work we do that's sadly not true

Seriously tho. You can page textures of course, but as transferring between the GPU and System is very slow, it defeats the whole purpose of rendering faster.

Regards,
Thorsten

[vlado]

awww thats bad. We cant page textures?

I said that we are working on it; it is certainly possible. What I mean is that currently scenes most often fail because the textures do not fit on the GPU.

Best regards,
Vlado

[Alan Iglesias]

Thanks for info and your hard work !

And thanks for the kind words, Dadal. It really was my main reason for going to Siggraph (other than the Vray party of course!) and although I did speak to people that could make a difference, I left with the feeling that my concerns wer falling mostly on deaf ears, at least as far as nVidia is concerned.

They seem to be comfortable with their business model that markets expensive "CAD-based" video cards to corporations at a much higher price than the gaming stuff. And it's working for them, primarily because the people responsible for purchasing these cards are often corporate IT people, who typically have very little personal experience with 3D and just go by what is being sold to them by the reseller, who is looking to profit themselves.

What really kills me is that they insist on basing their price points on something cultural rather than performance! If they would only understand that making stuff that works really well will always perform great in an honest market, then things would be so much better.

Take The Chaos Group's business model, for instance. Anyone can see how well it is working and there is absolutely nothing there except great development, great performance, and honest marketing.

In any case Alan what about making survey ? Getting some people sign up petition to Nvidia and so on ? It looks like you has connectionzzzzz ! :]

A survey is a good idea, but then Estes would hate me even more than he does now! Ha! I think the best survey is really people voting with thier wallets, and for that to haoppen we need competition. It's time for AMD to step up and slow the nVidia train down before it gets out of control and our dream of all-GPU rendering is only for folks with the deepest pockets!

And sadly, unlike the "old days" I have no more real connections in the business. Everyone I knew from 3D Labs, Integraph, and so on are long gone. nVidia pretty much rules the roost and I'm just a old guy from the trenches.

But we can keep trying!

-Alan

PS - I do have the Director of Marketing's card from AMD so I will definitely email him soon...-A

[duke2]

Couldn't the textures be paged in the same way as virtual textures, whereby the resolution is based on the screen resolution of the texture (with a multiplier). This wouldn't work so well got textures that are only refracted/reflected (say, a painting on the wall behind the camera but shown in a window reflection), but maybe this could be worked out via a prepass?

There are also numerous ways of cramming more textures in, like cramming 4 greyscale maps in 1 RGBA texture.

[vlado]

Couldn't the textures be paged in the same way as virtual textures, whereby the resolution is based on the screen resolution of the texture (with a multiplier).

Of course. That's exactly what we intend to do.

Best regards,
Vlado

[Morbid Angel]

Vlado what about ptex, I'm not sure how it works but does it impact the ram the same way the regular texture does?