Jump to content
rohandalvi

GPU and Mantra?

Recommended Posts

Guest tar

Have you tried the GL ROP in H15? The more serious issues with it have been resolved (transparent objects, material issues, BG image support).

 

Definitely appreciate the work done there, it was the second thing to check after the FEM improvements, motionblur and dof are as critical too to bring it up to the flipbook level.

Share this post


Link to post
Share on other sites

Definitely appreciate the work done there, it was the second thing to check after the FEM improvements, motionblur and dof are as critical too to bring it up to the flipbook level.

 

The current flipbook dof & moblur are built into the flipbooking system, not the viewport. They also predate any sort of GL shader work, so there are much better ways to do both (quality and performance-wise). It's on the todo, it lost out to Onion skinning this time around.

  • Like 1

Share this post


Link to post
Share on other sites

The current flipbook dof & moblur are built into the flipbooking system, not the viewport. They also predate any sort of GL shader work, so there are much better ways to do both (quality and performance-wise). It's on the todo, it lost out to Onion skinning this time around.

Well if at some point would be possible to get a decent approximation of the mantra render in the viewport this would be great to block the lighting and use it as a tool to approve lighting before fully rendering the shot in Mantra.

This would be so useful

  • Like 1

Share this post


Link to post
Share on other sites
Guest tar

Well if at some point would be possible to get a decent approximation of the mantra render in the viewport this would be great to block the lighting and use it as a tool to approve lighting before fully rendering the shot in Mantra.

This would be so useful

 

This is why a master class in the viewport/glsl would be wicked - glsl can do so much and a kickstart would be great to get the community going.

Share this post


Link to post
Share on other sites

Well if at some point would be possible to get a decent approximation of the mantra render in the viewport this would be great to block the lighting and use it as a tool to approve lighting before fully rendering the shot in Mantra.

This would be so useful

 

when you see what sketchfab guys can display in a webgl viewport

https://labs.sketchfab.com/siggraph2014/viewer.html?model=devastator

 

even for transparency you have cheats :

http://madebyevan.com/webgl-water/

 

i guess that you can definitly get some kickass accuracy beetween a Mantra PBR render and a OpenGL 3.2 PBR viewport.

considering that the set of possibility offer by OpenGL 3.2 beats hands downs the very limited WebGL sets of function.

 

And i agree that realtime previz would beat hands down any GPU previz stuff for efficiency ...

Edited by sebkaine

Share this post


Link to post
Share on other sites

 

i guess that you can definitly get some kickass accuracy beetween a Mantra PBR render and a OpenGL 3.2 PBR viewport.

considering that the set of possibility offer by OpenGL 3.2 beats hands downs the very limited WebGL sets of function.

 

I hope H16 will have materials with some decent representation in the viewport, compared to the final render in Mantra.

If at least the "speculars BSDF shapes" are similar then it will be very useful.

Share this post


Link to post
Share on other sites

If we had mantragpu and if it had the functionality of Redshift it would be more than enough for a massive number of users. And I think it would bring many more users to the camp.

Share this post


Link to post
Share on other sites

So four Phi cards that cost around $10K (not including the rest of the computer) are on par with really old Xeon processors. I don't get why anyone would want to do this.

Share this post


Link to post
Share on other sites
On 6.1.2017 at 5:20 PM, lukeiamyourfather said:

So four Phi cards that cost around $10K (not including the rest of the computer) are on par with really old Xeon processors. I don't get why anyone would want to do this.

in the cinebench the render wasn't optimized(compiled) for the xeon phi, there are some optimizations options open, so it didn't show the full power.   

on page 19 there is a compairing of two systems with the embree renderer:

https://embree.github.io/data/embree-siggraph-2016-final.pdf

if the render is optimized it will be in the same speed range of a 2x E5-2699(2.3Ghz / 18core) system

 

Edited by Mandrake0

Share this post


Link to post
Share on other sites

That's the thing with these massively parallel hardware architectures - you pretty much have to re-express your algorithm on its terms, or you just don't see much improvement. Rewriting C++ as OpenCL is a good start.

  • Like 1

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×