Return-Path: Received: from mail-ob0-f169.google.com ([209.85.214.169] verified) by media-motion.tv (CommuniGate Pro SMTP 4.2.10) with ESMTP-TLS id 4710353 for AE-List@media-motion.tv; Wed, 09 May 2012 18:52:36 +0200 Received: by obbwd18 with SMTP id wd18so575916obb.28 for ; Wed, 09 May 2012 09:54:11 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=20120113; h=date:from:to:message-id:in-reply-to:references:subject:x-mailer :mime-version:content-type:x-gm-message-state; bh=8LLICBInyn5/JDbhdlUb5Nfu8PSXYPothpxERdXsTaY=; b=SoRv0a3Hz7LtCf/hr2OCoM7iUKV7czh9Oyh2j2Rhzv5EznQWUD8A+O0bmuIEXhy9eQ BiSTaUudhjw76rvA2TQ0mgf7jqhizQwkPYHSt/ho3l1IEiraT0Gf4nonbdRVBx1w4XRO Z+7q1+pMU9EQVM0krdfoWtcnLtUP/CULDxBS2/lcCr9OOZZJ7VVXbMxUY/HR26CWRko8 odpMDOd8yC0oYnsRZ/xiAi5/sntKqjyH906dJ+BWs6zFvbba0MZuEY71UQmyxgwrz9Ye fqtt5bpqQuab4hX/OFw/o8qjJzZmn2MKMhNJse15KEYYOhQjRJdjCq1Yc+3xLGBLwUh3 x6QA== Received: by 10.182.159.5 with SMTP id wy5mr1295846obb.24.1336582450992; Wed, 09 May 2012 09:54:10 -0700 (PDT) Return-Path: Received: from gutsblow-iMac-2.local (cpe-76-94-181-188.socal.res.rr.com. [76.94.181.188]) by mx.google.com with ESMTPS id j4sm3201436obn.19.2012.05.09.09.54.09 (version=TLSv1/SSLv3 cipher=OTHER); Wed, 09 May 2012 09:54:10 -0700 (PDT) Date: Wed, 9 May 2012 09:54:08 -0700 From: Satya G Meka To: After Effects Mail List Message-ID: <78202DB77EB74F6EA649143D81691EF9@rowbyte.com> In-Reply-To: References: Subject: Re: [AE] made in CS6 X-Mailer: sparrow 1.5 (build 1043.1) MIME-Version: 1.0 Content-Type: multipart/alternative; boundary="4faaa130_73a1821b_5a7" X-Gm-Message-State: ALoCoQkVtMpaN++q1XJR+zz1aYmmOfcTLAg5h1bATUBaavkHzIfEBZH/Vx55i6zvt59WDTIE52uk --4faaa130_73a1821b_5a7 Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: quoted-printable Content-Disposition: inline I hope Adobe somehow standardizes access to the GPU within AE and opens i= t to 3rd party devs. Right now, every plugin with GPU Rendering functiona= lity tries to checkout huge GPU buffers/use all the resources without any= co-ordination & control, which usually leads to crashes. Another issue, = GPU features are highly dependent on the hardware make & model and its im= possible to test features on all modern GPU cards for 3rd party developer= s. =20 Satya G Meka. =20 On Wednesday, May 9, 2012 at 9:21 AM, mylenium=40mylenium.de wrote: > =20 > Nah. Cause and effect, you know. The problem is not GPU usage vs. CPU u= sage, the problem is simply this whole system being so poorly optimized. = I mean, so far all the examples using this raytrace stuff are trivial bey= ond belief, yet all the brute force of a GPU can't compensate for those d= eficiencies when rendering some simple cheesy text. And you are wrong abo= ut CPUs not having evolved. Today's Xeons are compeltely different beasts= than their predecessors were a few years ago jsut as a 3rd generation co= re7i is a different one than a first generation one 3 years ago. Still, e= ven that isn't really the point - my core7i here still renders such a sce= ne in the same time in Cinema4D, but without any GPU usage simply because= its CPU code has evolved over the years and is pretty optimized. And C4D= doesn't even have the fastest renderer. modo, Lightwave, Mental Ray and = others are even further optimized, not to speak of hybrid GPGPU renderers= like Arion, Maxwell RT, iRay or even pure GPU renderers like Octane buil= t on CUDA rather than OptiX. And you are over-obsessing over =22raytraced= =22 and =22interactive=22. I've said many times that I could do this sinc= e 2003 in =46Prime for Lightwave and that one was 90% CPU based with some= OpenGL thrown in here and there. So no matter how you pose the question,= something is not right about the road Adobe chose. My only consolation r= eally is that hopefully they really are working on reworking all the API = underpinnings and then we may be talking when we have alternatives and ne= ed not rely on Adobe's good will. Until then I will remain extremely crit= ical and skeptical of what I see so far. As far as I'm concerned, they si= mply have their priorities backwards and if you allow me - all this will = ever be to me is a bling-bling feature that makes some nice marketing, bu= t quickly falls apart in practical use. None of the people I talked to wo= uld upgrade to CS6 for this, but of course most of them are 3D artists...= =20 > =20 > =20 > =20 > =20 > =20 > =20 > Mylenium =20 > =20 > =20 > =20 > =5BPour Myl=C3=A8ne, ange sur terre=5D =20 > ----------------------------------------- =20 > www.mylenium.de (http://www.mylenium.de) =20 > =20 > =20 > =20 > =20 > =20 > Steve Oakley = hat am 9. Mai 2012 um 18:00 geschrieben: =20 > =20 > > Ok, let me make some clarifications with the project open in front of= me this morning - =20 > > =20 > > I did make ONE render pass at the project with really high photons/pi= xel that did take a long time. it was too glossy looking. thats the one t= hat stuck in my head > > =20 > > the pass it seems I posted was actually 6 photons/pixel, render time = 12min 12 sec - thats 1080p24 with Do=46 and motion blur cranked up. lots = of Do=46 and MB make for longer render times. turn them down / off and re= nder time can drop to next to nothing. > > =20 > > why GPU =3F because it provides a level of interactivity you can't ge= t w/o massive processing. remember this is ray traced, not some OGL phong= shader. while 3D does work on the CPU, yes its slow. =20 > > =20 > > the next answer is simpler - CPU's have really kinda of stagnated the= last couple of years. their performance gains have been marginal while t= he price of the top performers is thru the roof. its easy to spend =241k+= per Xeon CPU if you want the current screamer. =20 > > =20 > > the latest GPU's are far far far cheaper. the newest 680 is what - =24= 600 =3F adding that to your current and probably a bit old machine will g= ive it a lot more life then dropping =245k to just get started with a dua= l Xeon box. seriously my 8core MBP with Dual GPU's is still very viable f= or every day work. the GPU investment as stretched out my purchase cycle = for an entirely new machine quite a bit. =20 > > =20 > > you don't have to get a Quadro... I'm going to have a blog post about= this soon... but the short answer is if you do this at home, casually at= work, ect a G=46X class card will probably be fine for you. thats a few = hundred =24. OTH if you edit all day long, do 3D in AE all day, render at= nite and just pound hard on the machine then a Quadro is what you want b= ecause they are built for this level of continuous use and abuse. if you = make your living from the machine, its not that hard a decision. =20 > > =20 > > Prem Pro also greatly benefits from multiple GPU's too, or just faste= r ones. =20 > > =20 > > So does speed grade > > =20 > > and davinci resolve > > =20 > > and photoshop > > =20 > > given that all my main apps are seeing a lot of great improvements fr= om more / better GPU's I'm there. also note that Prem Pro has OCL acceler= ation now - so ATI GPU's are getting some love too now. > > =20 > > it just amazes me.... people say =22I want you to max out the CPU's, = RAM, GPU's of my machine. I've got'm, please use them. Then when an app d= oes....=22 =20 > > =20 > > S > > =20 > > On May 9, 2012, at 9:53 AM, Greg Balint wrote: =20 > > > I agree Mylenium, =20 > > > =20 > > > I was super excited to hear about the 3d stuff being added to AE. I= thought. =22man, this will save me a lot of time and effort coordinating= renders from c4d by just doing simple things right in AE.=22 But I have = to say I'm severely disappointed with this. I figured from the simple ext= rusion feature that this would be some quick turn around 3d feature, but = with all of that horsepower, and still 40 minutes render=3F I could prob= ably render the same scene, with native Do=46 and motion blur directly in= C4d in about 5 minutes in C4D. =20 > > > =20 > > > Now I'm not bashing the example piece itself. But I can't see how I= 'd even be able to stand watching all of that processing power take forev= er for a 3d extrusion of some text with reflections and shadows. =20 > > > =20 > > > No doubt this is a welcome feature for people who don't have any 3d= package to work with, but for those of us that do, I don't see any of us= using this feature for production environment speedy work. =20 > > > =20 > > > =20 > > > I also wonder why the 3d implementation was decided to work solely = on gpu processing, when CPU is obviously more versatile and powerfulas se= en with 3d packages that don't use gpu at all for rendering. As a PC use= r with a beefy ATI card installed, it surely feels like Adobe has its poc= kets lined with Nvidia money. =20 > > > =20 > > > Just my opinion here, but I'm way more excited about the caching fe= atures they've added. =20 > > > =20 > > > ////Greg Balint =20 > > > ///Art Director / Motion Graphics Designer > > > delRAZOR.com/ (http://delRAZOR.com/) =20 > > > =20 > > > =20 > > > On May 9, 2012, at 10:20 AM, =22 mylenium=40mylenium.de (mailto:myl= enium=40mylenium.de) =22 < mylenium=40mylenium.de (mailto:mylenium=40myle= nium.de) > wrote: =20 > > > =20 > > > > Nobody is arguing the need for some sort of 3D in AE (though not = exactly that kind of 3D as far as I'm concerned) and the marketability of= such a feature, but seriously, if it takes 40 minutes on a machine with = a beefy 4000 series Quadro and an additional Tesla board, then something = is severely is askew and one can only wonder why the AE team chose this a= pproach in the first place. I wouldn't even wanna know how long it takes = to render this if AE falls back to its software mode. 2 hours=3F 3 hours=3F= 5 hours=3F In a day and age where software renderers in 3D programs chur= n out frames with full global illumination and all the bells and whistles= in minutes even without any GPU involvement, what AE has to offer compar= es poorly. =20 > > > > =20 > > > > Mylenium =20 > > > > =20 > > > > =20 > > > > =5BPour Myl=C3=A8ne, ange sur terre=5D =20 > > > > ----------------------------------------- =20 > > > > www.mylenium.de (http://www.mylenium.de/) =20 > > > > =20 > > > > =20 > > > > =20 > > > > =20 > > > > Steve Oakley < steveo=40practicali.com (mailto:steveo=40practical= i.com) > hat am 9. Mai 2012 um 16:07 geschrieben: =20 > > > > =20 > > > > > I don't have or use C4D so I couldn't render anything 3D that w= ay. AE is my app of choice, its what I have in front of me to use. please= keep in mind this is all new for AE - ray tracing and the start of real = 3D. its an entirely new look / capabilities. for how long have people as= ked for real 3D in AE =3F forever. I don't think knocking is right. =20 > > > > > =20 > > > > > 16 photons was about right. I had done some lower passes and it= was too grainy. not saying that wasn't a good look...I'll go check the p= roject again. > > > > > =20 > > > > > I've seen some other really nice work done which will hopefull= y surface soon. > > > > > =20 > > > > > S > > > > > =20 > > > > > On May 9, 2012, at 8:37 AM, James WIlson wrote: =20 > > > > > > I think 16 photons might be overkill considering the dark nat= ure and minimal amount of objects to bounce light off of. Maybe it could = have been rendered in less time with the same results if you took that se= tting down. It does look nice. =20 > > > > > > =20 > > > > > > =20 > > > > > > On May 9, 2012, at 2:47 AM, mylenium=40mylenium.de (mailto:my= lenium=40mylenium.de) wrote: =20 > > > > > > > Yeah, and the same stuff probably renders in the same time = in C4D - without any GPU fancies required and on a 3 year old machine. Ri= diculous and nothing to write home about. =20 > > > > > > > =20 > > > > > > > =20 > > > > > > > =20 > > > > > > > Mylenium =20 > > > > > > > =20 > > > > > > > =20 > > > > > > > =5BPour Myl=C3=A8ne, ange sur terre=5D =20 > > > > > > > ----------------------------------------- =20 > > > > > > > www.mylenium.de (http://www.mylenium.de/) =20 > > > > > > > =20 > > > > > > > =20 > > > > > > > =20 > > > > > > > Steve Oakley < steveo=40practicali.com (mailto:steveo=40pra= cticali.com) > hat am 9. Mai 2012 um 07:27 geschrieben: =20 > > > > > > > =20 > > > > > > > > ok... now that AE CS6 is loose I'll show you some titles= I made for my web shows. CS6 rendered with Q4000 + Tesla C2075 in 40 min= as I recall... thats with cranked up Do=46, motion blur, 16 photons / pi= xel - ie real nice settings :) its quick, don't blink. totally done in AE= =20 > > > > > > > > =20 > > > > > > > > http://www.youtube.com/watch=3Fv=3DX-ZNVyGSVIY&feature=3D= youtu.be =20 > > > > > > > > =20 > > > > =20 > =20 --4faaa130_73a1821b_5a7 Content-Type: text/html; charset="utf-8" Content-Transfer-Encoding: quoted-printable Content-Disposition: inline
I hope Adobe somehow standardizes access to the GPU within AE and ope= ns it to 3rd party devs. Right now, every plugin with GPU Rendering funct= ionality tries to checkout huge GPU buffers/use all the resources without= any co-ordination & control, which usually leads to crashes. Another= issue, GPU features are highly dependent on the hardware make & mode= l and its impossible to test features on all modern GPU cards for 3rd par= ty developers. 

= Satya G Meka.
=20

On Wednesday, May 9, 2= 012 at 9:21 AM, mylenium=40mylenium.de wrote:

=20 =20 =20

Nah. Cause and effect, you know. The problem is not GPU usage v= s. CPU usage, the problem is simply this whole system being so poorly opt= imized. I mean, so far all the examples using this raytrace stuff are tri= vial beyond belief, yet all the brute force of a GPU can't compensate for= those deficiencies when rendering some simple cheesy text. And you are w= rong about CPUs not having evolved. Today's Xeons are compeltely differen= t beasts than their predecessors were a few years ago jsut as a 3rd gener= ation core7i is a different one than a first generation one 3 years ago. = Still, even that isn't really the point - my core7i here still renders su= ch a scene in the same time in Cinema4D, but without any GPU usage simply= because its CPU code has evolved over the years and is pretty optimized.= And C4D doesn't even have the fastest renderer. modo, Lightwave, Mental = Ray and others are even further optimized, not to speak of hybrid GPGPU r= enderers like Arion, Maxwell RT, iRay or even pure GPU renderers like Oct= ane built on CUDA rather than OptiX. And you are over-obsessing over =22r= aytraced=22 and =22interactive=22. I've said many times that I could do t= his since 2003 in =46Prime for Lightwave and that one was 90% CPU based w= ith some OpenGL thrown in here and there. So no matter how you pose the q= uestion, something is not right about the road Adobe chose. My only conso= lation really is that hopefully they really are working on reworking all = the API underpinnings and then we may be talking when we have alternative= s and need not rely on Adobe's good will. Until then I will remain extrem= ely critical and skeptical of what I see so far. As far as I'm concerned,= they simply have their priorities backwards and if you allow me - all th= is will ever be to me is a bling-bling feature that makes some nice marke= ting, but quickly falls apart in practical use. None of the people I talk= ed to would upgrade to CS6 for this, but of course most of them are 3D ar= tists...


Mylenium

=5BPour Myl=C3=A8ne, ange sur terre=5D
-----------------------------------------
www.mylenium.de


Steve Oakley <steveo= =40practicali.com> hat am 9. Mai 2012 um 18:00 geschrieben:

<= div> Ok, let me make some clarifications with the project open in front o= f me this morning -
 
I did make ONE render pass at the project with really high phot= ons/pixel that did take a long time. it was too glossy looking. thats the= one that stuck in my head
 
the pass it seems I posted was actually 6 photons/pixel, render= time 12min 12 sec - thats 1080p24 with Do=46 and motion blur cranked up.= lots of Do=46 and MB make for longer render times. turn them down / off = and render time can drop to next to nothing.
 
why GPU =3F because it provides a level of interactivity you ca= n't get w/o massive processing. remember this is ray traced, not some OGL= phong shader. while 3D does work on the CPU, yes its slow. 
 
the next answer is simpler - CPU's have really kinda of stagnat= ed the last couple of years. their performance gains have been marginal w= hile the price of the top performers is thru the roof. its easy to spend = =241k+ per Xeon CPU if you want the current screamer. 
 
the latest GPU's are far far far cheaper. the newest 680 is wha= t - =24600 =3F adding that to your current and probably a bit old machine= will give it a lot more life then dropping =245k to just get started wit= h a dual Xeon box. seriously my 8core MBP with Dual GPU's is still very v= iable for every day work. the GPU investment as stretched out my purchase= cycle for an entirely new machine quite a bit. 
 
you don't have to get a Quadro... I'm going to have a blog post= about this soon... but the short answer is if you do this at home, casua= lly at work, ect a G=46X class card will probably be fine for you. thats = a few hundred =24. OTH if you edit all day long, do 3D in AE all day, ren= der at nite and just pound hard on the machine then a Quadro is what you = want because they are built for this level of continuous use and abuse. i= f you make your living from the machine, its not that hard a decision.&nb= sp;
 
Prem Pro also greatly benefits from multiple GPU's too, or just= faster ones. 
 
So does speed grade
 
and davinci resolve
 
and photoshop
 
given that all my main apps are seeing a lot of great improveme= nts from more / better GPU's I'm there. also note that Prem Pro has OCL a= cceleration now - so ATI GPU's are getting some love too now.
 
it just amazes me.... people say =22I want you to max out the C= PU's, RAM, GPU's of my machine. I've got'm, please use them. Then when an= app does....=22 
 
S

On May 9, 2012, at 9:53 AM, Greg Balint wrote:

I agree Mylenium,
 
I was super excited to hear about the 3d stuff being added = to AE. I thought. =22man, this will save me a lot of time and effort coor= dinating renders from c4d by just doing simple things right in AE.=22 But= I have to say I'm severely disappointed with this. I figured from the si= mple extrusion feature that this would be some quick turn around 3d featu= re, but with all of that horsepower, and still 40 minutes render=3F  = ;I could probably render the same scene, with native Do=46 and motion blu= r directly in C4d in about 5 minutes in C4D. 
 
Now I'm not bashing the example piece itself. But I can't s= ee how I'd even be able to stand watching all of that processing power ta= ke forever for a 3d extrusion of some text with reflections and shadows.&= nbsp;
 
No doubt this is a welcome feature for people who don't hav= e any 3d package to work with, but for those of us that do, I don't see a= ny of us using this feature for production environment speedy work. =
 
 
I also wonder why the 3d implementation was decided to work= solely on gpu processing, when CPU is obviously more versatile and power= fulas seen with 3d packages that don't use  gpu at all for rendering= . As a PC user with a beefy ATI card installed, it surely feels like Adob= e has its pockets lined with Nvidia money. 
 
Just my opinion here, but I'm way more excited about the ca= ching features they've added. 

////Greg Balint
///Art Director / Motion Graphics Designer

On May 9, 2012, at 10:20 AM, =22 mylenium=40myleni= um.de =22 < mylenium=40myleni= um.de > wrote:
 
Nobody is arguing the need for some sort of 3D in AE (though= not exactly that kind of 3D as far as I'm concerned) and the marketabili= ty of such a feature, but seriously, if it takes 40 minutes on a machine with a beefy 4000 series Quadro and an additional Tesl= a board, then something is severely is askew and one can only wonder why = the AE team chose this approach in the first place. I wouldn't even wanna= know how long it takes to render this if AE falls back to its software m= ode. 2 hours=3F 3 hours=3F 5 hours=3F In a day and age where software ren= derers in 3D programs churn out frames with full global illumination and = all the bells and whistles in minutes even without any GPU involvement, w= hat AE has to offer compares poorly.

Mylenium

 

=5BPour Myl=C3=A8ne, ange sur terre=5D
-----------------------------------------
www.mylenium.de

 


Steve Oakley < steveo=40pract= icali.com > hat am 9. Mai 2012 um 16:07 geschrieben:

I don't have or use C4D so I couldn't render anything 3D th= at way. AE is my app of choice, its what I have in front of me to use.&nb= sp;please keep in mind this is all new for AE - ray tracing and the start= of real 3D. its an entirely new look / capabilities.  for how long = have people asked for real 3D in AE =3F forever. I don't think knock= ing is right.
 
16 photons was about right. I had done some lower pass= es and it was too grainy. not saying that wasn't a good look...I'll go ch= eck the project again.
 
 I've seen some other really nice work done which= will hopefully  surface soon.
 
S

On May 9, 2012, at 8:37 AM, James WIlson wrote:
I think 16 photons might be overkill considering the da= rk nature and minimal amount of objects to bounce light off of. Maybe it = could have been rendered in less time with the same results if you took t= hat setting down. It does look nice.
 

On May 9, 2012, at 2:47 AM, myleni= um=40mylenium.de wrote:

Yeah, and the same stuff probably renders in= the same time in C4D - without any GPU fancies required and on a 3 year = old machine. Ridiculous and nothing to write home about.

 

Mylenium 

 

=5BPour Myl=C3=A8ne, ange sur terre=5D
-----------------------------------------
www.myleni= um.de

 


Steve Oakley < ste= veo=40practicali.com > hat am 9. Mai 2012 um 07:27 geschrieben:

> ok... now that  AE CS6 is loose I'll sho= w you some titles I made for my web shows. CS6 rendered with Q4000 + Tesl= a C2075 in 40 min as I recall... thats with cranked up Do=46, motion blur= , 16 photons / pixel - ie real nice settings :) its quick, don't blink. t= otally done in AE
>
> http://www.youtube.com/watch=3Fv=3DX-= ZNVyGSVIY&feature=3Dyoutu.be
> 

 

  =20 =20 =20 =20 =20 =20

--4faaa130_73a1821b_5a7--