Forum Groups
  All forums
    Help & Feedback
      Work in progress
      Finished Art
      Non-Max related

Featured Threads
  inspiration alert!!!
(36 replies)
  Indespensible MaxScripts, Plugins and 3rd Party Tools
(37 replies)
  The allmighty FREE Resources Thread !
(17 replies)
  spam alert!!!
(4886 replies)
  Maxforums member photo gallery index
(114 replies)
  Maxforums Member Tutorials
(89 replies)
  three cheers to maxforums...
(240 replies)
  101 Things you didnt know in Max...
(198 replies)
  A Face tutorial from MDB101 :D
(95 replies) Members Gallery
(516 replies)
(637 replies)
  Dub's Maxscript Tutorial Index
(119 replies)

Maxunderground news unavailable

First page  Go to the previous page   [01]  [02]  Go to the next page  Last page
mf-bench 1.0 (comparing the performance of different real-time raytracers & GPUs)
show user profile  advance-software
As discussed, it is now possible to run vray-rt gpu on both AMD & nvidia hardware, so the question is - which is faster ?

To figure this out, we'll need a benchmark. Unless anyone knows of one that already exists, we'll have to make it ourselves.

The benchmark will require a test scene that can be uploaded & shared.

It should ideally be configured so it can be rendered in vray-rt & other real-time raytracers.

Something that features a wide variety of useful techniques would be best as this would enable us to get a feel for the overall level of performance available now, and in future hardware/software.

The reasons for doing this are to figure out which products are best value for money and to encourage competition.

E.g. If the benchmark renders faster on AMD hardware, nvidia will push to increase performance & vice versa. The aim being to accelerate the development of real-time raytracing technology.

Building a test scene is not my area - I'm a programmer, not an artist, so someone else will have to do this. It doesn't have to be asthetically appealing (though that wouldnt hurt), just give us a standard test we can apply on different systems.

For example, half a dozen shiny, shadow casting, refractive teapots on a table might do as a starting point.

Doesn't really matter what, just as long as the same test is run on different systems & you have some way of measuring how long it took to render.

Thoughts welcome - you guys know a lot more about raytracing than I do.

read 1117 times
2/1/2011 10:53:32 AM (last edit: 2/6/2011 12:23:27 PM)
show user profile  Mr_Stabby
assuming it uses direct compute and double precision, nvidia will currently win as amd has some issues in the area (probably fixed by drivers in upcoming months)

read 1098 times
2/1/2011 11:08:19 AM (last edit: 2/1/2011 11:08:19 AM)
show user profile  advance-software
Vray-RT GPU doesn't use DirectCompute (DX11), it uses OpenCL.
No idea whether it's double precision, though we should of course compare like with like as far as possible.
Quality and speed are both worth measuring.

Please see lower half of this page :

read 1097 times
2/1/2011 11:11:07 AM (last edit: 2/1/2011 11:14:15 AM)
show user profile  losthatter
A-S, do u have it working on your system?
read 1039 times
2/2/2011 2:22:51 AM (last edit: 2/2/2011 2:22:51 AM)
show user profile  advance-software
No. I'm working on real-time stuff (a 3D web browser), so don't raytrace anything yet .
The tech is too slow for me at the moment ... but it'll get there.

I run 3dsmax occassionally but only to test our export & content creation tools.

I'm interested in seeing some kind of a benchmark put together but won't be creating it.

This is my contribution. Others will have to put the test scene together, benchmark on different systems & post results back to the board.

We have a number of nvidia vray-rt gpu users on the board, so if something was put together, I would have thought they'd participate. All that would be required would be to load the scene, hit render & post back system specs & time to render + any feedback on how the test scene might be improved to make it more useful.

This would actually make a great student project - way more interesting and useful than the pointless surveys we see from time to time. If you want to figure out how to raytrace things quickly, get involved !

The 1.0 benchmark doesn't need to be particuarly fancy or take long to develop / use.

If it proves to be useful we could develop it further in future by perhaps scripting a bunch of different scene configurations together into some kind of automated test, but to get started I'd suggest keeping it simple.

We want to answer the following questions :

1. Which is better value for money for GPU accelerated (real-time) raytracing - AMD or nvidia hardware ?

2. Who has the biggest e-penis ?

3. Which product should receive the much coverted golden fronat award ?

4. Why does mineral water that has trickled through mountains for centuries have a use by date ?
read 1019 times
2/2/2011 10:37:44 AM (last edit: 2/2/2011 11:22:05 AM)
show user profile  advance-software
bump. Some people here still think raytracing should be done on the CPU.

Is no-one interesting in figuring out which hardware provides the best (fastest) solution ?
read 949 times
2/3/2011 9:19:15 AM (last edit: 2/3/2011 9:20:26 AM)
show user profile  losthatter
im working on some numbers and scenes for a Bench. im trying to find an equal NV card to compare.

Thus far. my Juniper Hd5770 does seem to render faster then my 4 CPU's by atleast 20%.
when it comes to Vray anyway. i havent seen any other GPU rendering plugins/engins that do this so ist kinda hard to compare results if u only have 1 tool. But im digging.

Stay tuned for more updates

read 923 times
2/3/2011 7:54:20 PM (last edit: 2/3/2011 7:58:25 PM)
show user profile  advance-software
I think iray is also GPU accelerated - Nik ?

Thanks for doing this. If possible try to figure out whether the calculations are being done in single or double precision as the performance will differ significantly between the two. dp slower but more accurate.
read 919 times
2/3/2011 8:18:03 PM (last edit: 2/3/2011 8:22:53 PM)
show user profile  Nik Clark
iRray is GPU accelerated. That is true. With iRay, when you exceeded the memory of the card, the renderer drops back to using the CPU.

read 905 times
2/3/2011 9:00:23 PM (last edit: 2/3/2011 9:00:23 PM)
show user profile  advance-software
Here's a first candidate benchmark scene, donated by Vella (thank you)

I can load but not render as I don't have vray installed. Should we go with a vray specific benchmark or should we stick to standard materials/objects so we can compare performance of other raytracers ?
read 882 times
2/4/2011 11:54:59 AM (last edit: 2/4/2011 11:54:59 AM)
show user profile  Nik Clark
You should have one scene for each renderer, if you want to see which hardware is best.

I can supply iRay scenes, if needed.

read 878 times
2/4/2011 12:00:25 PM (last edit: 2/4/2011 12:00:25 PM)
show user profile  advance-software
Thanks. Is it possible to have one universal scene that works with all raytracers (by using standard mats), or do you have to use raytracer specific materials ?

A universal test would be better because we could then compare renderers and hardware.
read 875 times
2/4/2011 12:06:32 PM (last edit: 2/4/2011 12:07:18 PM)
show user profile  BishBashRoss
I don't think you're going to be able to do a fair test of different renderers like that. Too many variables and most renders have their own optimised materials.

read 868 times
2/4/2011 12:22:23 PM (last edit: 2/4/2011 12:22:23 PM)
show user profile  Nik Clark
I agree, it would be hard to have one scene that was fairly rendered by each system, as there are loads of options to set, and it is not a fair comparison.

read 864 times
2/4/2011 12:24:40 PM (last edit: 2/4/2011 12:24:40 PM)
show user profile  advance-software
Thanks guys - so, as vray appears to be the most popular renderer, it sounds like it makes sense focus there for the time being.

If vella's scene is suitable for initial testing, please post GPU & render time.

Don't make any changes to the scene unless they are necessary (in which case we'll have to post an updated scene so everyone is running the same test). All renders must be the same size - I'd suggest 1024x768.

Feel free to hit me with the stupid stick if I've missed anything or if you have a better test scene.
read 859 times
2/4/2011 1:07:49 PM (last edit: 2/5/2011 5:06:04 PM)
First page  Go to the previous page   [01]  [02]  Go to the next page  Last page
#Maxforums IRC
Open chat window