hi all
i very rarely post here mainly cos of workload and NDAs BUT i have been around for almost two years.
On monday night when the latest beta was made available i thought that finally the time had come to get an almost perfect renderer for the jobs we do, especially after a 3 month or so gap that has followed the release of 1-45. The bugs on it were plenty so one would have thought that the main issues were fixed.
So i went ahead and downloaded the latest and decided to replace the 1-45 version which we have already started using for production mainly because of the proxy.
I am afraid i ll have to go bak to that version as whatever i do with the new one is always at least 3 times slower than before and i am only using irmap and qmc, no lightmap.
here are two images that show that (work in progress)
settings: all default apart from
min -4
max-2
hsph 20
prim irmap mult 1
sec qmc mult 0.5
qmc subd 3, sec bounces 1
The scene was made in max7 and the materials are mostly multisubs Standard and vray (1-46) mats for all the glass
Trees are vray mats( 1-46) with alpha in op.
There are no glossies in the scene.
I also have been completely unable to render the same scene with lighmap regardless of the combination of settings i used because it always freezes before it even starts the calculation.
Lightmaps with teapots work great but i dont think our clients would like teapots insted of buildings(although i wish they did).
Any ideas or comments would be really appreciated.
regards
alkis
ps. i do not wish to upset anyone and i am very happy with vray ,
i only want to get my old rendertimes with this a fixed 1-45 version whatever its name might be
hi guys
thanks for the replys
natty: here are the settings
i dont have a clue about the primitive count
its exactly the same scene
another bug maybe?..
longjohn: light map , when it works is the coolest and fastest addition to vray . i havent been using it for production up to now because most of my mats are standard and now that i can use it it freezes up.
i ll continue to try and figure it out though as i can spare two more days on testing and upgrading
as mentioned in other posts it would be great if there was an extensive list of changes between version and maybe some quick tips for optimisations and different settings
thanks
alkis
Do you have a dual-processor machine? If yes, then I think the difference in render times comes from the Dynamic geometry option. You can try to render with Static geometry and see if there is any difference (if all the geometry fits into the memory, of course).
In 1.45, each rendering thread had its own geometry pool, where is in 1.46, all threads share the same geometry which means that when one of the processors loads geometry, the others must wait for it. If this needs to be done often, processors waste more time waiting for each other than doing any actual work. It’s one of them memory/speed issues. It’s not a bug, so there is nothing to fix (if this is the problem, of course).
As for lightmaps, they will freeze (cycle infinitely) when you have opacity-mapped standard materials; this is what I managed to find out. In that case, using VRay materials does the trick.
vlado hi
thanks for the reply
specs : dual xeon 3.2 , 2gb RAM with 3GB switch enabled
i have made a render with static geometry which is significantly faster but still twice as slow as 1-45
if all threads now share the same geometry and it causes a problem with dual machines when you are using dynamic does it mean you can never use dynamic cause it ll always be slower?
thanks for the light map suggestions, i ll do some more tests and report back any further probs
thanks again for your time and effort
regards
alkis
PS vlado: do you have any idea why the primitives count keep on changing on the same scene if i use different methods of rendering or different setups
should i ignore it or could it be part of the problem
The primitives change because with the dynamic geometry option, some geometry may never be loaded, if it’s not needed. Using static geometry will always give the same primitive count (unless you have displacement or fur, which always create dynamic geometry).
if all threads now share the same geometry and it causes a problem with dual machines when you are using dynamic does it mean you can never use dynamic cause it ll always be slower?Well, the purpose of dynamic geometry was to save memory, more than anything. The 1.46 build uses even less memory for dynamic geometry on dual machines than 1.45, although sometimes this may be at the cost of rendering time.
Well, the purpose of dynamic geometry was to save memory, more than anything. The 1.46 build uses even less memory for dynamic geometry on dual machines than 1.45, although sometimes this may be at the cost of rendering time.
that makes sense but the differences in render time are way to big to justify something like that
thank you again for your time and explanations
regards
alkis
Well, in your scene the slowdown is probably not only because of the dynamic geometry; have you tried comparing the number of samples in the irradiance map (as stated in the Irradiance map rollout)?
hi again
here are some of my findings
the number of samples is as follows
vray1-46 :29161 samples - times as above (8min)
vray1-45 :21033 samples - times as above (4 min)
thats with the settings as above
after that i tried another test in 1-45
changed the min max and hsph subd to -3 -1 and 30 accordingly everything else exactly the same
this gave me the following image at a still a faster time than 1-46
now the number of samples for that are…
…70511…
which is more than twice the amount of 1-46 render ( and still faster)
this is very interesting and even more puzzling
i hope that we can reach the bottom this
regards
alkis
You can also take a look at the number of rays that were cast by VRay in the VRay messages window to see if there is any difference (it says Number of raycasts: XXXXXXX after each frame).
vray 1-45 : it doesnt give me a number. the message window is below
both dynamic and static options have the same messages in that window
regards
alkis
Alkis … as far as i know the newer version takes more samples than the last build … show me a screen shot of your setting’s please, all your settings…
also have been completely unable to render the same scene with lighmap regardless of the combination of settings i used because it always freezes before it even starts the calculation.
here are further results
v 1-46
qmc noise threshold 0,01
min samples 8
result : 7m 27 sec , rays 35942907
qmc noise threshold 0,02
min samples 3
result : 5m 18 sec , rays 24025846
qmc noise threshold 0,03
min samples 3
result : 4m 36 sec , rays 19028980
this seems that it does the trick
in the irmap rollout it stills shows 29148 samples for the last and fastest try
should we use these as a default maybe now as the settings i had before were the default ones
there is no change in the image quality , however due to the resolution its very difficult to judge that accurately.
i m glad we got somewhere
thanks for all your time and suggestions vlado
it looks like after all that 1-46 might be staying…
actually there is other way to increase speed, by decreasing the amount value(right above the noise value) to 0.6 for example, and decrease the hsph to 20-25. I think, vray cuts out to fast samples. This will work very well on exteriors even with animation.
hi again
losbellos: the issue in this post was not mainly how to optimize the specific scene as the method you mention etc i know and have been using in previous vray versions
the problem is that both 1-45 and 1-46 with the same default values behave completely differently.
if there is a need for every scene to change qmc noise etc to get a scene rendering roughly the same as 1-45 then maybe those default values should change to something else and furthermore people should be made aware from now on.
i only mean it in a very positive way so everyone can get even better results out of vray.
regards
alkis
If you have the time, try to follow some threads about 1.46 in the general and bug sections of the forum. Quite a lot of people are testing the differences between 1.45/1.09 and 1.46. Especially IR map and qmc settings need different values…
For IR map, change hsph to half the value you used before (and best to lower interp also). The qmc iq already discussed above. These are just my thoughts and are based on test with 1.09 vs 1.46, not sure if they apply to 1.45 vs 1.46.
Vray 1.46 has new default settings, but when you load an old scene it takes of course the values of that scene… I would also like to see a shift in values so that old and new builds render a bit more the same. But as vlado already said, this is different for every scene so probabely there’s no general rule to do so.
