Dezamagire procesor

xeleron, on 07 februarie 2019 - 03:10, said:

O fi citit omul pe aici recitalurile numaratorilor de nuclee. Din martie 2017 e tot un RAGE pe forum cu invazia softurilor multithreading, de la banalul player pana lz ... de ce nu?... ce se discuta pe acest topic.

Redau mai jos zicerile de bine ale unui consultant AutoDesk, ziceri care infig un cutit virtual in inima de piatra a armatei moar coars:
"Multi-threading isn't a magic wand, its not that easy to implement, and won't give as much "MASSIVE performance increases" as some would think.  Given that AutoCAD is rarely the only application running I'd rather it not hog all the cores.

There can only be one process modifying one data record (drawing is a database) at a time, otherwise the results are not predictable which is a Bad Thing in computing.  And with heavily data-driven programs like Civil3D the order of change becomes important, you can't have something change data A after its been read to make changes for data B; otherwise the state of data B no longer reflects the actual state of data A.  That kind of dependency doesn't lend itself well to parallel processing.

The best comparison is for image processing and rendering.  Each pixel in an image is fully independent of all the others, so every one of them can be processed in its own thread without affecting the others.  With a little organization, iterative processing can be done on the image without having to wait for the entire image to finish processing at each stage.

Some others had the idea of unloading the UI elements to their own processes, which might not be a bad starting place.  The same problems still apply, even in a read-only use ie. they would still need to wait until changes are complete before updating the displayed values.  They could constantly monitor for changes"

E ca scenariul apocaliptic prezentat in primele pasaje seamana nitel si cu gamingul?

Durerea e ca si nici astea care stiu sa foloseasca toate nucleele la randarea/encodarea finala nu stiu sa foloseasca toate firele cat lucrezi la proiect. Cand am recomandat Puget  pentru analiza mai aprofundata, mare parte dintre amedisti l-au declarat sorosist doar pentru ca nu le placeau ce vedeau acolo.

Edited by xvmhunt, 07 February 2019 - 06:03.

ia uite ce zice nene asta:
"Re: Dedicated render machine for 3ds max vs one single multi core machine? Time
A whole lot to go through here...
Hardware choices for rendering are based on what renderer you are using - the commonly used Mental Ray and Arnold are CPU based ie. it doesn't make use of the video card.  For these you want more cores, and you also want more RAM to go along with it (typically 1 - 2 GB per core).  Other renderers are GPU based ie. it *does* use the video card.  For these you want enough VRAM on the video card to hold the scene being rendered.  Optimized scenes are very important because unlike CPU rendering if it cannot be loaded into the VRAM it just. won't. run.  Multiple cards can be added to reduce rendering time.  Mental Ray can now benefit from Quadro/Tesla video cards as well, so do some further investigation first.
A dedicated rendering workstation can be a good idea if you need to render during work hours (whenever those may be - I've seen "hot seating" shift work situations where you have people on a computer 24/7), in which case the users computer needs to be free for work.  If you have idle time overnight it's not a bad idea to investigate doing your heavy rendering work during that time.  There's also the option of distributed rendering, where you can have multiple workstations on a network acting as slaves for the rendering process.  Others are best suited for describing that, so I'll leave it to them to chime in.  You don't necessarily want a heavy duty render station (dual oct-core processors, 64 GB RAM, dual 1080Ti video cards, etc.) as a daily workstation - most of that power isn't used on day to day work, only rendering.
You don't want a laptop for rendering.  They're built for portability not heavy lifting.
----------------------------------
If you are going to fly by the seat of your pants, expect friction burns.
"I don't know" is the beginning of knowledge, not the end."

Edited by arg, 07 February 2019 - 10:44.

MinettoMatoi, on 07 februarie 2019 - 03:57, said:

Vezi setarile pentru AVI (codec in special).
Nu-l fa direct AVI, fa imagini separate (PNG), apoi le faci ce format video vrei tu (cu FFMpeg sau alte programe, simple, gratuite).

Acuma ca merge f bine pe quiksilver hardware renderer, sunt f multumit,  nu stiam de chestia asta, vechile versiuni de 3D Studio Max nu aveau asta, foloseau doar procesorul si nu si placa video, ca erau prea slabe acum 20 ani, deci Max 1.2 si 2.0, la astea ma refer.

Pai, serios, mai pune si tu mina si studiaza, daca ai ramas acum 20 de ani..... cum ai vrea sa intelegi ceva din ziua de azi ?!
Vezi, de aia vii aici cu calculele astea mirobolante !!!!

Ai si tu dreptate, nu m-am mai ocupat de calucri de ani, am pierdut notiunile, pana in 2008 din 1998 eram la curent. O sa pun mana sa ma pun la punct.

 shmu, on 04 februarie 2019 - 15:54, said:

 MinettoMatoi, on 04 februarie 2019 - 19:49, said:

 super8, on 05 februarie 2019 - 15:46, said:

 MinettoMatoi, on 05 februarie 2019 - 17:43, said:

 MinettoMatoi, on 05 februarie 2019 - 17:53, said:

 buyonevideo, on 05 februarie 2019 - 18:46, said:

 super8, on 05 februarie 2019 - 19:08, said:

Pentru toti userii care nu l-au crezut pe @MinettoMatoi ca si-a cumparat AMD Ryzen Threadripper 2950X:
 Screenshot1.png   133.44K   16 downloads
 Screenshot2.png   125.03K   15 downloads
 Screenshot3.png   168.49K   12 downloads
 Screenshot4.png   201.15K   11 downloads
 Screenshot 5.png   126.25K   9 downloads

Pacat ca are doar 16GB Ram  
Trebuia minim 32GB !

nu stiu daca a zis cineva ca nu si-a cumparat.

problema e cu ce compara si mai ales CUM compara.

e ca si la masini: aproape orice rabla merge cu 100km/h. dar ca sa ajungi la 300 iti trebuie enorm mai multi CP, aerodinaminca, greutate mai redusa, consum si alte bazaconii. nu e ca si cum ai nevoie de doar 3 ori mai multa putere.

Edited by shmu, 17 February 2019 - 10:53.

Am rezolvat eu, am schimbat renderer ul in art și acum trage de cpu la maxim. Scanline e pt altceva.

Nu ati inteles, eu stiu ca nu se compara cum am comparat eu numeric, dar am zis asa doar raportul babeste, adica cum ar veni am facut eu o metoda de masura.
De ex pe x calc cu proc de 1,5 Ghz, merge cu indicativul 1, si ala la 7 Ghz merge de 4 ori mai rapid, si in acest sistem de masura am zis.
DAR paradoxal orice ziceti voi, sa stiti ca, cam da si raportul numeric, nu 100 %, dar un 80 % da, in general.

Este ceva ciudat, dar daca nu prea ma pricep bine la noile versiuni de max care au mai multe tipuri de randare, aia e, deci:
La scena de test cu acele 12 cuburi ultra complexe ca structura, deci sarcina maxima, pe scanline mergea cel  mai incet, le ART si Quicksilver hardware mergea avion.
In schimb..mai noi am vazut ca la o scena mult mai simpla, cu o casa simpla facuta de mine doar pe afara, nu si interior, o strada, cateva lumini fara reflezii si con de lumina, fara umbre, pe scanline merge f rapid, pe Quicksilver merge foarte incet si pe ART merge infect de incet.
Am impresia ca diferenta tip randare NU sta in doar pe proc sau si cu placa video, ci sta in tipul de randare pur si simplu care randeaza in alt mod, si ptr fiecare caz in parte, un fel de randare merge rapid, altul mult mai incet desi ala s-ar potrivi poate.

Oricum daca e asa, adios scene cu reflexii, con de lumina, umbre, fac din alea mai simple, ca doar nu stau 100 ore sa randez in loc de 5 ore.
De ex am vazut pe youtube unu care a facut o scena cu lumini, randata pe scanline dura 10 min dar nu se randa bine efectul, cand atingea lumina suprafetele nu era okey, umbrele, cand a randat cu Mental Ray, se vedea brici DAR a durat de 27 ori mai mult adica 270 minute, asta e infectie, pacat....

Ce infectie? daca umblii pe situ cu virusi da.

Nu mai, infectie de viteza la randare am zis , era la figurat.