POV-Ray : Newsgroups : povray.advanced-users : Speeding up media Server Time
5 Dec 2021 21:21:22 EST (-0500)
  Speeding up media (Message 23 to 32 of 32)  
<<< Previous 10 Messages Goto Initial 10 Messages
From: Ash Holsenback
Subject: Re: Speeding up media
Date: 14 Feb 2021 14:38:42
Message: <60297c42@news.povray.org>
On 2/14/21 2:33 PM, Ash Holsenback wrote:
> On 2/14/21 12:32 PM, Alain Martel wrote:
>> Le 2021-02-14 à 11:02, Ash Holsenback a écrit :
>>
>>> another question: are variance, confidence and ratio valid in method 
>>> 3... given current narrative i'd say yes to the first two and no on 
>>> the last
>>>
>>> made some progress... here's what i have so far
>>> http://wiki.povray.org/content/Reference:Sampling_Parameters_%26_Methods
>>
>> The confidense and ratio are to adjust the samples numbers between the 
>> various intervals and are not used.
>> As for variance, it also seems to be method 1 and 2 specific. It don't 
>> seems to have any effect when using method 3.
> 
> are you sure about that... given that the narrative after Sampling 
> Method 3: uses the term variance, then later on in the narrative in 
> Sampling method 1: last paragraph ( above the note ) links variance and 
> confidence together... did i make a bad leap? the role of ratio is a 
> couple of paragraphs up... please expand on this a bit ;-)

one more thing... would it be accurate to say that most if not all those 
media related keywords work for all three methods, however some of their 
roles have changed.


Post a reply to this message

From: Thomas de Groot
Subject: Re: Speeding up media
Date: 15 Feb 2021 02:37:16
Message: <602a24ac@news.povray.org>
Op 14/02/2021 om 20:38 schreef Ash Holsenback:
> On 2/14/21 2:33 PM, Ash Holsenback wrote:
>> On 2/14/21 12:32 PM, Alain Martel wrote:
>>> Le 2021-02-14 à 11:02, Ash Holsenback a écrit :
>>>
>>>> another question: are variance, confidence and ratio valid in method 
>>>> 3... given current narrative i'd say yes to the first two and no on 
>>>> the last
>>>>
>>>> made some progress... here's what i have so far
>>>> http://wiki.povray.org/content/Reference:Sampling_Parameters_%26_Methods 
>>>>
>>>
>>> The confidense and ratio are to adjust the samples numbers between 
>>> the various intervals and are not used.
>>> As for variance, it also seems to be method 1 and 2 specific. It 
>>> don't seems to have any effect when using method 3.
>>
>> are you sure about that... given that the narrative after Sampling 
>> Method 3: uses the term variance, then later on in the narrative in 
>> Sampling method 1: last paragraph ( above the note ) links variance 
>> and confidence together... did i make a bad leap? the role of ratio is 
>> a couple of paragraphs up... please expand on this a bit ;-)
> 
> one more thing... would it be accurate to say that most if not all those 
> media related keywords work for all three methods, however some of their 
> roles have changed.

Give that man a beer!

Much better assuredly.

Pending more learned answers on your last question, I would be wary to 
affirm this. The point being probably - if those are silently ignored by 
method 3 as Alain states - that people will wring themselves into hoops 
only to see some 'improvements' to their media, without any results 
obviously. If, on the other hand, those are /not/ ignored, we need to be 
dead sure about it. What does the code say, Mr Pokorny?

-- 
Thomas


Post a reply to this message

From: William F Pokorny
Subject: Re: Speeding up media
Date: 15 Feb 2021 06:00:30
Message: <602a544e$1@news.povray.org>
On 2/15/21 2:37 AM, Thomas de Groot wrote:
> Op 14/02/2021 om 20:38 schreef Ash Holsenback:
>> On 2/14/21 2:33 PM, Ash Holsenback wrote:
>>> On 2/14/21 12:32 PM, Alain Martel wrote:
>>>> Le 2021-02-14 à 11:02, Ash Holsenback a écrit :
...
> Give that man a beer!
> 
> Much better assuredly.
> 
> Pending more learned answers on your last question, I would be wary to 
> affirm this. The point being probably - if those are silently ignored by 
> method 3 as Alain states - that people will wring themselves into hoops 
> only to see some 'improvements' to their media, without any results 
> obviously. If, on the other hand, those are /not/ ignored, we need to be 
> dead sure about it. What does the code say, Mr Pokorny?
> 

Generally, I recommend we make the recommendation to use only media 3 
more strongly. Using the other methods with the current code base is 
potentially confusing given they won't get some of the normal defaults 
due how the flip to method 3 was done in the parsing phase.

I agree with Alain. The keywords: confidence, variance and ratio are 
used only with methods 1 and 2 (a). The use of the variance word in the 
method 3 wording is I suppose correct, but unfortunate. Another sample 
is taken between two existing, if the difference between the two end 
samples is > the aa_threshold.

(a) For the purposes of documentation see (a1) and ignore the rest of 
this footnote. Someone can set the intervals in method 3 to more than 1 
and if they do ratio does come into play with a lit and un-lit intervals 
scheme internally which 'probably' doesn't have much of an end effect 
for method 3, but... Always true, sometimes, almost never, never? Not 
sure... When using multiple medias with mixed sampling types, I know of 
some bleed through in the keyword settings between methods.

(a1) It would be good to somewhere add a note along the lines of: When 
using multiple medias, it's strongly recommend all use the same sampling 
method. With the current code there is some keyword setting bleed 
between methods.

Aside: The sentence about monte carlo.... only applies to methods 1 and 2.

Aside: FWIW. In povr I plan to at least make it appear methods 1 & 2 and 
their keywords do not exist at all (and I'll pull jitter from media 3). 
Ideally, I'll replace the whole of the current implementation with a new 
media 3 like one. There are standing issue with media 3 too with respect 
to speed, accuracy of result and when and if adaptive sampling actually 
kicks in. But, it's media - which is fuzzy. Most remaining issues don't 
matter much to an end result as I recall - so long as we sample enough 
that things look good. ;-)

Bill P.


Post a reply to this message

From: Ash Holsenback
Subject: Re: Speeding up media
Date: 15 Feb 2021 12:39:57
Message: <602ab1ed$1@news.povray.org>
On 2/15/21 6:00 AM, William F Pokorny wrote:
> On 2/15/21 2:37 AM, Thomas de Groot wrote:
>> Op 14/02/2021 om 20:38 schreef Ash Holsenback:
>>> On 2/14/21 2:33 PM, Ash Holsenback wrote:
>>>> On 2/14/21 12:32 PM, Alain Martel wrote:
>>>>> Le 2021-02-14 à 11:02, Ash Holsenback a écrit :
> ...
>> Give that man a beer!
>>
>> Much better assuredly.
>>
>> Pending more learned answers on your last question, I would be wary to 
>> affirm this. The point being probably - if those are silently ignored 
>> by method 3 as Alain states - that people will wring themselves into 
>> hoops only to see some 'improvements' to their media, without any 
>> results obviously. If, on the other hand, those are /not/ ignored, we 
>> need to be dead sure about it. What does the code say, Mr Pokorny?
>>
> 
> Generally, I recommend we make the recommendation to use only media 3 
> more strongly. Using the other methods with the current code base is 
> potentially confusing given they won't get some of the normal defaults 
> due how the flip to method 3 was done in the parsing phase.
> 
> I agree with Alain. The keywords: confidence, variance and ratio are 
> used only with methods 1 and 2 (a). The use of the variance word in the 
> method 3 wording is I suppose correct, but unfortunate. Another sample 
> is taken between two existing, if the difference between the two end 
> samples is > the aa_threshold.
> 
> (a) For the purposes of documentation see (a1) and ignore the rest of 
> this footnote. Someone can set the intervals in method 3 to more than 1 
> and if they do ratio does come into play with a lit and un-lit intervals 
> scheme internally which 'probably' doesn't have much of an end effect 
> for method 3, but... Always true, sometimes, almost never, never? Not 
> sure... When using multiple medias with mixed sampling types, I know of 
> some bleed through in the keyword settings between methods.
> 
> (a1) It would be good to somewhere add a note along the lines of: When 
> using multiple medias, it's strongly recommend all use the same sampling 
> method. With the current code there is some keyword setting bleed 
> between methods.
> 
> Aside: The sentence about monte carlo.... only applies to methods 1 and 2.
> 
> Aside: FWIW. In povr I plan to at least make it appear methods 1 & 2 and 
> their keywords do not exist at all (and I'll pull jitter from media 3). 
> Ideally, I'll replace the whole of the current implementation with a new 
> media 3 like one. There are standing issue with media 3 too with respect 
> to speed, accuracy of result and when and if adaptive sampling actually 
> kicks in. But, it's media - which is fuzzy. Most remaining issues don't 
> matter much to an end result as I recall - so long as we sample enough 
> that things look good. ;-)
> 
> Bill P.
> 
> 

thanks /everyone/ for guidance and support... please review
http://wiki.povray.org/content/Reference:Sampling_Parameters_%26_Methods


Post a reply to this message

From: Alain Martel
Subject: Re: Speeding up media
Date: 15 Feb 2021 18:58:20
Message: <602b0a9c$1@news.povray.org>
Le 2021-02-15 à 06:00, William F Pokorny a écrit :

> (a) For the purposes of documentation see (a1) and ignore the rest of 
> this footnote. Someone can set the intervals in method 3 to more than 1 
> and if they do ratio does come into play with a lit and un-lit intervals 
> scheme internally which 'probably' doesn't have much of an end effect 
> for method 3, but... Always true, sometimes, almost never, never? Not 
> sure... When using multiple medias with mixed sampling types, I know of 
> some bleed through in the keyword settings between methods.
> 
> Bill P.
> 
> 

This probably explain why using several intervals bog down method 3 so 
much. It tries to compute ratios and variances when it's not needed.


Post a reply to this message

From: Thomas de Groot
Subject: Re: Speeding up media
Date: 16 Feb 2021 02:36:33
Message: <602b7601$1@news.povray.org>
Op 15/02/2021 om 18:39 schreef Ash Holsenback:
> On 2/15/21 6:00 AM, William F Pokorny wrote:
>> On 2/15/21 2:37 AM, Thomas de Groot wrote:
>>> Op 14/02/2021 om 20:38 schreef Ash Holsenback:
>>>> On 2/14/21 2:33 PM, Ash Holsenback wrote:
>>>>> On 2/14/21 12:32 PM, Alain Martel wrote:
>>>>>> Le 2021-02-14 à 11:02, Ash Holsenback a écrit :
>> ...
>>> Give that man a beer!
>>>
>>> Much better assuredly.
>>>
>>> Pending more learned answers on your last question, I would be wary 
>>> to affirm this. The point being probably - if those are silently 
>>> ignored by method 3 as Alain states - that people will wring 
>>> themselves into hoops only to see some 'improvements' to their media, 
>>> without any results obviously. If, on the other hand, those are /not/ 
>>> ignored, we need to be dead sure about it. What does the code say, Mr 
>>> Pokorny?
>>>
>>
>> Generally, I recommend we make the recommendation to use only media 3 
>> more strongly. Using the other methods with the current code base is 
>> potentially confusing given they won't get some of the normal defaults 
>> due how the flip to method 3 was done in the parsing phase.
>>
>> I agree with Alain. The keywords: confidence, variance and ratio are 
>> used only with methods 1 and 2 (a). The use of the variance word in 
>> the method 3 wording is I suppose correct, but unfortunate. Another 
>> sample is taken between two existing, if the difference between the 
>> two end samples is > the aa_threshold.
>>
>> (a) For the purposes of documentation see (a1) and ignore the rest of 
>> this footnote. Someone can set the intervals in method 3 to more than 
>> 1 and if they do ratio does come into play with a lit and un-lit 
>> intervals scheme internally which 'probably' doesn't have much of an 
>> end effect for method 3, but... Always true, sometimes, almost never, 
>> never? Not sure... When using multiple medias with mixed sampling 
>> types, I know of some bleed through in the keyword settings between 
>> methods.
>>
>> (a1) It would be good to somewhere add a note along the lines of: When 
>> using multiple medias, it's strongly recommend all use the same 
>> sampling method. With the current code there is some keyword setting 
>> bleed between methods.
>>
>> Aside: The sentence about monte carlo.... only applies to methods 1 
>> and 2.
>>
>> Aside: FWIW. In povr I plan to at least make it appear methods 1 & 2 
>> and their keywords do not exist at all (and I'll pull jitter from 
>> media 3). Ideally, I'll replace the whole of the current 
>> implementation with a new media 3 like one. There are standing issue 
>> with media 3 too with respect to speed, accuracy of result and when 
>> and if adaptive sampling actually kicks in. But, it's media - which is 
>> fuzzy. Most remaining issues don't matter much to an end result as I 
>> recall - so long as we sample enough that things look good. ;-)
>>
>> Bill P.
>>
>>
> 
> thanks /everyone/ for guidance and support... please review
> http://wiki.povray.org/content/Reference:Sampling_Parameters_%26_Methods

Yes. Add that note (a1) from Bill. That is important indeed. One trivial 
thing I want to suggest: Add an empty line between Methods. It would 
increase the readability even more.

[Maybe Alain will not have to repeat indefinitely anymore about method 
3, intervals 1, and single samples.... ;-) ]

-- 
Thomas


Post a reply to this message

From: William F Pokorny
Subject: Re: Speeding up media
Date: 17 Feb 2021 08:23:18
Message: <602d18c6$1@news.povray.org>
On 2/15/21 6:58 PM, Alain Martel wrote:
> Le 2021-02-15 à 06:00, William F Pokorny a écrit :
> 
>> (a) For the purposes of documentation see (a1) and ignore the rest of 
>> this footnote. Someone can set the intervals in method 3 to more than 
>> 1 and if they do ratio does come into play with a lit and un-lit 
>> intervals scheme internally which 'probably' doesn't have much of an 
>> end effect for method 3, but... Always true, sometimes, almost never, 
>> never? Not sure... When using multiple medias with mixed sampling 
>> types, I know of some bleed through in the keyword settings between 
>> methods.
>>
>> Bill P.
>>
>>
> 
> This probably explain why using several intervals bog down method 3 so 
> much. It tries to compute ratios and variances when it's not needed.

Only ratios(a) I believe, but yes, expect this a part of the typical 
slow down (always? sometimes not?). I don't understand all the whys.

The most obvious reason is the number of interval acts as a multiplier 
for the starting number of samples with method 3 and this is 10.

(a) and the underlying light interval stuff. This code is tied to light 
types. My guess is it's aimed at speeding up (or better resolving 
perhaps?) spotlights and cylindrical lighting when using media(a1). 
Maybe it all more or less works - but I have doubts about how well 
beyond the simplest test set ups. I'd like to pull all the lit/unlit 
interval code out too.

Bill P.


Post a reply to this message

From: Alain Martel
Subject: Re: Speeding up media
Date: 17 Feb 2021 11:40:01
Message: <602d46e1$1@news.povray.org>
Le 2021-02-17 à 08:23, William F Pokorny a écrit :
> On 2/15/21 6:58 PM, Alain Martel wrote:
>> Le 2021-02-15 à 06:00, William F Pokorny a écrit :
>>
>>> (a) For the purposes of documentation see (a1) and ignore the rest of 
>>> this footnote. Someone can set the intervals in method 3 to more than 
>>> 1 and if they do ratio does come into play with a lit and un-lit 
>>> intervals scheme internally which 'probably' doesn't have much of an 
>>> end effect for method 3, but... Always true, sometimes, almost never, 
>>> never? Not sure... When using multiple medias with mixed sampling 
>>> types, I know of some bleed through in the keyword settings between 
>>> methods.
>>>
>>> Bill P.
>>>
>>>
>>
>> This probably explain why using several intervals bog down method 3 so 
>> much. It tries to compute ratios and variances when it's not needed.
> 
> Only ratios(a) I believe, but yes, expect this a part of the typical 
> slow down (always? sometimes not?). I don't understand all the whys.
> 
> The most obvious reason is the number of interval acts as a multiplier 
> for the starting number of samples with method 3 and this is 10.
> 
> (a) and the underlying light interval stuff. This code is tied to light 
> types. My guess is it's aimed at speeding up (or better resolving 
> perhaps?) spotlights and cylindrical lighting when using media(a1). 
> Maybe it all more or less works - but I have doubts about how well 
> beyond the simplest test set ups. I'd like to pull all the lit/unlit 
> interval code out too.
> 
> Bill P.

In my testings, intervals 5 samples 10 run MUCH slower than samples 100 
and default intervals 1.


Post a reply to this message

From: William F Pokorny
Subject: Re: Speeding up media
Date: 17 Feb 2021 14:22:53
Message: <602d6d0d$1@news.povray.org>
On 2/17/21 11:40 AM, Alain Martel wrote:
> Le 2021-02-17 à 08:23, William F Pokorny a écrit :
...
> 
> In my testings, intervals 5 samples 10 run MUCH slower than samples 100 
> and default intervals 1.
> 

OK. I'm sure you see what you see. If a small scene, post it and I'll 
add it to my media test scenes. It will get run, but it's unlikely I'll 
dig into it. The intervals > 1 cases are not something we should be using.

Using the recent media scene posted by Bald Eagle I see:

intervals, samples

Constant media
---------------
1, 100 : 46.27user 0.05system 0:12.60elapsed  (51.00)
5,  10 : 29.46user 0.07system 0:08.38elapsed  (9.36)
10, 10 : 51.06user 0.04system 0:13.81elapsed  (10.10)

Non-Constant media which in fact is constant:
( density { function { 1 } }  // Fools the POV-Ray constant test.
---------------
1, 100 : 140.91user 0.10system 0:36.46elapsed  (101.00)
5,  10 :  80.60user 0.07system 0:21.24elapsed  (11.00)
10, 10 : 151.94user 0.08system 0:39.21elapsed  (11.00)

Bald Eagles original gradient x density media
---------------
1, 100 : 132.72user 0.08system 0:34.38elapsed  (101.00)
5,  10 :  76.09user 0.09system 0:20.12elapsed  (11.00)
10, 10 : 143.69user 0.07system 0:37.18elapsed  (11.00)

Intervals > 1 being slower but not too MUCH slower.

Bill P.


Post a reply to this message

From: Mike Horvath
Subject: Re: Speeding up media
Date: 22 Feb 2021 14:00:04
Message: <6033ff34$1@news.povray.org>
On 2/7/2021 2:23 AM, Thomas de Groot wrote:
> Op 07/02/2021 om 04:41 schreef Mike Horvath:
>> On 2/5/2021 6:56 AM, Bald Eagle wrote:
>>> it's likely not your media - but your media settings.
>>>
>>> try:
>>>
>>>
http://news.povray.org/povray.newusers/message/%3C59a9ec1c%40news.povray.org%3E/#%3C59a9ec1c%40news.povray.org%3E

>>>
>>>
>>>
>>
>> I'm not sure how that applies to my case. I am not using emissive 
>> media or the "samples" keyword.
>>
> 
> There is nothing really wrong with your code as far as I can tell and 
> from own experience. From the outside (of the media) it is even rather 
> fast due to the default intervals 1 and samples 1 (Yes! you /do/ use 
> samples, even if you don't).
> 
> However, depending on the location of the camera within or without the 
> media, render times can be very different; you even may need to increase 
> the samples value to avoid splotches, and that will decrease render time 
> even more although not that much.
> 
> Using media is time consuming. There is no alternative.
> 


I am inside the cylinder, but the cylinder is clipped by another smaller 
cylinder to create a shell.



Mike


Post a reply to this message

<<< Previous 10 Messages Goto Initial 10 Messages

Copyright 2003-2021 Persistence of Vision Raytracer Pty. Ltd.