Return-Path: Received: from smtpauth01.prod.mesa1.secureserver.net ([64.202.165.181] verified) by media-motion.tv (CommuniGate Pro SMTP 4.2.10) with SMTP id 4746060 for AE-List@media-motion.tv; Wed, 13 Jun 2012 23:46:06 +0200 Received: (qmail 19529 invoked from network); 13 Jun 2012 21:48:49 -0000 Received: from unknown (173.247.4.230) by smtpauth01.prod.mesa1.secureserver.net (64.202.165.181) with ESMTP; 13 Jun 2012 21:48:48 -0000 Subject: Re: [AE] (OT) An example of a PC version of a Mac Pro with today's tech for AE References: From: Greg Balint Content-Type: multipart/alternative; boundary=Apple-Mail-699FE3A5-78E9-4FD8-8698-4634E6FF54B5 X-Mailer: iPhone Mail (9A405) In-Reply-To: Message-Id: <0D8B7B54-DCF9-465E-9F9D-D20C9DB00CFA@delrazor.com> Date: Wed, 13 Jun 2012 17:49:17 -0400 To: After Effects Mail List Content-Transfer-Encoding: 7bit Mime-Version: 1.0 (1.0) --Apple-Mail-699FE3A5-78E9-4FD8-8698-4634E6FF54B5 Content-Transfer-Encoding: quoted-printable Content-Type: text/plain; charset=us-ascii Forgive me for the constant questioning of this statement from the product m= anager, but if I have a 6 core cpu with hyperthreading on, and I can choose i= n the preferences to use 9 cores, and it launches 9 separate processes in my= OS with 2 Gb of RAM reserved for each. And I can see those processes all ru= nning when I render, and those 9 "cores" that windows or OSX shows are all m= axed out at 100% usage, and the render shows 9 frames at a time being render= ed in chunks, how is AE only using the 6 physical cores and not also utilizi= ng the extra cores available due to hyperthreading? By the statement made, Wouldn't I see only 6 of them pegged at 100% usage, w= hile the other 6 aren't doing anything? ////Greg Balint ///Art Director / Motion Graphics Designer delRAZOR.com/ On Jun 13, 2012, at 4:28 PM, Chris Meyer wrote: > I probably have my terminology on hyperthreading vs. multithreading wrong,= and for that I apologize. >=20 > But this does not change the fact - confirmed by the AE product manager, f= er crissakes - that when you enable the preference for multiprocessor render= ing, this specific feature can only really use the physical cores, and one s= hould not spend more than needed buying RAM strictly to feed virtual cores d= uring multiprocessor rendering, as you won't see a gain.=20 >=20 > To give an example: With my 12 real + 12 virtual core Mac (which the OS re= ports as being 24 cores), a sweet spot would have been 36 or so GB of memory= so I could reserve the recommended 3 GB per 10 cores (reserving 2 cores and= 6 GB of RAM for other functions), rather than spending money to install (22= x 3) + 6 =3D 72 GB when it would not be taken advantage of for multiproc re= ndering. (Me, I underspent and only put 24 GB in it, riffing on outdated rec= ommended practices that you only needed 1.5 or maybe 2 GB per process. I reg= ret that, as I'm not getting the multiproc boost I should have. OTOH, runnin= g two Quadro 4000 cards has been nice, but that's a separate well-tread disc= ussion.) >=20 > Aside from that specific use case, in general more RAM is good (esp. for l= arge format work like SV2's, as well as more caching etc.), and more cores -= both real and virtual - are good to accelerate other parts of the program. I= never said otherwise.=20 >=20 > I understand various people have real empirical data that changing certain= numbers gave certain results, and I don't dispute that. But there's a diffe= rence between speculating at how the two may be connected, and learning what= is actually going on underneath the hood. >=20 > Hey - I'm just trying to share some obscure (and unexpectedly controversia= l!) knowledge to help fellow users spend their money wisely. >=20 > back to work -=20 > Chris --Apple-Mail-699FE3A5-78E9-4FD8-8698-4634E6FF54B5 Content-Transfer-Encoding: quoted-printable Content-Type: text/html; charset=utf-8
Forgive me for the constan= t questioning of this statement from the product manager, but if I have a 6 c= ore cpu with hyperthreading on, and I can choose in the preferences to use 9= cores, and it launches 9 separate processes in my OS with 2 Gb of RAM reser= ved for each. And I can see those processes all running when I render, and t= hose 9 "cores" that windows or OSX shows are all maxed out at 100% usage, an= d the render shows 9 frames at a time being rendered in chunks, how is AE on= ly using the 6 physical cores and not also utilizing the extra cores availab= le due to hyperthreading?

By the statement made, Wo= uldn't I see only 6 of them pegged at 100% usage, while the other 6 aren't d= oing anything?

////Greg Balint
///Art Director / Motion Graphics D= esigner

On Jun 13, 2012, at 4:28 PM, Chris Meyer <chris@crishdesign.com> wrote:

I probably have my terminology on hyp= erthreading vs. multithreading wrong, and for that I apologize.

But this does not change the fact - confirmed by the AE product manag= er, fer crissakes - that when you enable the preference for multiprocessor r= endering, this specific feature can only really use the physical cores, and o= ne should not spend more than needed buying RAM strictly to feed virtual cor= es during multiprocessor rendering, as you won't see a gain. 

To give an example: With my 12 real + 12 virtual core Mac (w= hich the OS reports as being 24 cores), a sweet spot would have been 36 or s= o GB of memory so I could reserve the recommended 3 GB per 10 cores (reservi= ng 2 cores and 6 GB of RAM for other functions), rather than spending money t= o install (22 x 3) + 6 =3D 72 GB when it would not be taken advantage of for= multiproc rendering. (Me, I underspent and only put 24 GB in it, riffing on= outdated recommended practices that you only needed 1.5 or maybe 2 GB per p= rocess. I regret that, as I'm not getting the multiproc boost I should have.= OTOH, running two Quadro 4000 cards has been nice, but that's a separate we= ll-tread discussion.)

Aside from that specific u= se case, in general more RAM is good (esp. for large format work like SV= 2's, as well as more caching etc.), and more cores - both real and virtual -= are good to accelerate other parts of the program. I never said otherwise.&= nbsp;

I understand various people have real empiric= al data that changing certain numbers gave certain results, and I don't disp= ute that. But there's a difference between speculating at how the two may be= connected, and learning what is actually going on underneath the hood.

Hey - I'm just trying to share some obscure (and unexpe= ctedly controversial!) knowledge to help fellow users spend their money wise= ly.

back to work - 
Chris
= --Apple-Mail-699FE3A5-78E9-4FD8-8698-4634E6FF54B5--