when will blender take advantage of every core

most things i do on blender like particles or cloth don’t use more than one core

the view port performance isn’t good at all i start lagging at around 60k faces on gtx 560 or gtx 460

duplicating objects takes FOREVER when you have many objects

any idea when these things will be better im sure many people would like to know

:spin:

sooner or later

Have you tried a build with openMP enabled? Not sure if it’s on in the official builds but you can get them from graphicall.org

^ ill check that out

ok the max it got was 70… is it because i have 8 cores?

From my experience openmp uses more cores but it doesn’t max them out. I think it’s because it is experimental.

Enable VBOs (user prefs>System tab) and you’ll notice your viewport is as smooth as a baby’s bottom for millions of polys (had a 66 million poly sculpt in object mode at 33 fps with gtx 570). VBOs are off by default because they can be unstable in some cases (some simulations) and are slower when changing the scene (adding/removing meshes). But for a static scene, it’s heaven :slight_smile:

Otherwise yeah, it’d be nice if more simulation stuff was threaded. But complaining and demanding it from part/spare-time devs will only have them ignore you more.

VBOs also lead to crashes on intel GPUs :frowning:

Multithreading can be really difficult and there are some algorithms that just cannot be multithreaded.

Just an FYI, other 3D apps are in the same state as Blender. Certain processes lend themselves well to multiple CPU acceleration, but others , that require sequential processing, can only be safely implemented using a single thread (CPU). I am not saying that improvement could not be made, but the OpenMP builds are a step in the right direction and worth a download.

Let me calculate…umm 100/2 …30/2*3- … Around 5 years :wink:

VBOs do nothing for me there is no differenace… the max i can get is around 2 mil before i cant move at all lol
but in sculpt mode i can get 32 mil before i cant move at all

CPU have lost the game to the GPU as far as graphics is concerned. With GPUs being even 100 times faster in some case than a CPU, its better to get a better GPU than worry why blender does not full take advantage of your cores.

Saying that 64k face lagging does not sound reasonable, blender was not lagging at 500k when i was making my dragon back in 2.49 , so something fishy is happening here. Will give a try home to see how it performs.

OpenMP isn’t experimental.

Recently I’ve been looking into using OpenMP to do basic threading tasks for operations that can run in parallel. eg.
http://markmail.org/message/nc2jbal4vrw6qrxn
http://markmail.org/message/sn3qw6prurl5kp2r

Its pretty nice how you can just tag sections of code as being able to run parallel in any order, without a rewrite, however overall speedup quite small still - there are many more areas that would give bigger gains to thread, but they are much more work too and often mean re-structuring or rewriting existing code.

Probably we have most to gain from threaded modifier stack at the moment.

How I understand it, GPUs are conceptually the same as CPUs, only they have hundreds of weak cores, whereas CPUs have just a couple strong ones. So if something isn’t able to run on multiple CPU cores, it’d be even worse running it on a single GPU core

well your gpu does not do everything that your cpu can that’s why you want all your cores to work. you cant just replace cpu with gpu

i have heard that the new cards like the GTXs do not do good in view port… they really need to fix that because most of us blender user’s are game players and the gtx is where its at for games XD

if you have double sided meshes, turn that off and you will gain some speed. switching of selection highlighting also helps in object mode

^ i have done that it does nothing at all

well and a Swordfish is conceptually the same as any fish, but not any fish can reach speeds of 130 kms per hour.

Specialised hardware acceleration is not only a virtue of GPUs, audio cards with asio drivers can give speeds unreachable with plain CPU. And yes the diffirency can be massive for the average musician.

I think the benefits of hardware acceleration are evident in blender, just change cycles from gpu to cpu and watch the whole think crawl. Now what most dont realize is that even GUIs , like blender for example which is based on early version of opengl , are also hardware accelerated. Removed hardware opengl acceleration and watch even your guis crawl. The entired macos guis is basically opengl based and benefits from gpus directly.

The only reason why multicores are still around is sadly that GPUs are not capable of offering awesome speeds up for all scenarios. They are specialized tools afterall. But as far as music , audio and graphics is concerned no you would not want to rely on CPU at all. Unless speed does not concern you.

Wasn’t 2.67 planned to make overal performance better?
I don’t remember where I got that…

first, gtx’s do fine w/ viewport performance, if not better than any other card I know of
second, the problem with intel cards and VBO’s id basicaly intel cards, or intel cards drivers.
third, turning off object selection outlines will also increase performance at the cost of knowing what id selected :wink:
which almost completely defeats the purpose and another reason for colored wires, or even better colored layers…but whatever…
(in my human torch voice)blend on!