First Cinbench scores for "MacPro 8 core"

[lllab]

oh yeah 2100:-)))
this`ll be mine...maybe for christmas.

cheers
stefan

[RenatoT]

Only 2100 ?:(

A friend of mine got 960 with the Dual Core 2 Duo @ 3.2 ghz... so this result must almost 3x of this.

cheers :)
Renato

[NWoolridge]

Only 2100 ?:(

A friend of mine got 960 with the Dual Core 2 Duo @ 3.2 ghz... so this result must almost 3x of this.

cheers :)
Renato

A couple of things to remember: these were 2.66 Ghz chips, and it was an aftermarket mod to the Mac Pro. In other words, there may not have been enough RAM to satisfy eight cores in a render task, and it was running an unmodified EFI, so any mods necessary to optimize for eight cores were not made...

Nick

[rsquires]

yeah renato

a current Mac Pro Dual core 3 ghz has a Cinebench around the 1600's or so. So 8 cores should theoretically be 3200. Pretty poor scores I'd say.

regards

rich

[Srek]

There is no linear increase in renderspeed when adding cores. twice the number of cores don't double the renderspeed, it's more in the range of a factor 1.8 for each doubeling of core numbers.
Cheers
Björn

[willsud]

Interesting that the QX6700 got almost exactly the same scores as my dual dual-Xeon 5150..

not surprising I guess but costs hugely less

I'm happy with the dual 5150s though... at least I won't need the central heating on this winter :)

[3DBond]

Interesting that the WinXP numbers in Cinebench are slightly better with 8 cores yet slightly more worse with 4.

I wonder if Leopard will affect that.

Exciting times in any event. I'm hoping for a serious performance bump right around the time CS3 comes out.

[jddog]

There is no linear increase in renderspeed when adding cores. twice the number of cores don't double the renderspeed, it's more in the range of a factor 1.8 for each doubeling of core numbers.
Cheers
Björn

Björn... but this will not mean that the calculation of a image after shadow or radiosity calculation will be not faster using a 8 core als a 4 ? And in this case we can benefit of more that a 1.8 factor ?

jd

[lllab]

well a quad 2.66 has 1400CB.
it is known that the efficiency get worse with more cores.

the absolue maximum would be 1,8 as srek pointed out. thats the max you could get out, probably less for more cores. more realsitic for so much cores is 1.6-1,5 i think.

1core 100%
2cores180%
4cores 320%
8cores 583% (not 800%)

so that would be 435CB for one 2.66 core
783CB for 2cores
1409 for 4 cores
2536 for 8 cores 8x1.8)

as keeping the factor 1.8 is unrealistic for more cores(apple/intel did already great in keeping that for 4cores) calculating with 1,5 you get:
1409 x1,5 = 2113CB and thats exactly what this machine has.

sio i really would not count that as bad:-) at least well enouigh for my expectations.
in some techmags there was written stuff about multicore chips that let me expect much less at least...

in the end what is best on these quad chips is that for the price of one 5150 you get 2 now!

cheers
stefan

[imashination]

Might I also add that, in my opinion the current CB render test wasn't particularly well intended to scale up to so many processors. On a dual it finishes the test in 40 seconds, on a quad in 25 seconds, when the render times get so low, the numbers become less and less reliable, as they don't give enough time for the renderer to really get going. It takes a second or so for each render thread to initialise, so its already partly finished rendering before all the cores even get their teeth stuck in.

[lllab]

exactly what mash says, in my experience larger scenes benefit more of more corse than small fast ones. same as with DR. the CB test scene should be more complex to be more precise for todays machines, and it should use GI.

cheers
stefan

[laurent]

One of the problem with an 8cores machine is that I would want to use it for my work,
test scenes would fly, I could have the GI, AA always turned on. but then I would have to swap back and keep keeping with my Ol' 2cores while the big ass renders are calculated, that would suck ;)

1core 100%
2cores180%
4cores 320%
8cores 583% (not 800%)

That's also why Cinema needs proper network distributed bucket rendering:
2cores x 4(computers) = 720 again, we're talking about large renders wher distribution time is meaningless.

[Creature]

How does RAM influence multi-core rendering? I'm currently working with 2Gb RAM on a single core machine. Am I supposed to put 16Gb of RAM in a 8 core machine? :argh:

[Srek]

No, the memory overhead per thread isn't nearly that big. It should only nead a very few percent more memory per thread.
Cheers
Björn

[imashination]

Well, depends. Each thread needs a buffer to render to. If you do 8k billboards, youll need 100s of megs just for the buffers for them to render into.

[Creature]

Another thing about multicore (I'm looking into those as I'm upgrading my machine in the next month or two and currently try to figure out what to get):

Is it true that with Particle-, Cloth-, and Dynamicsimulations multicores aren't really of any use as that can only be done by one core? As for example Particles at frame 2 are based on the outcome of the calculation of frame 1 so you can't really split it among cores as core 2 won't know what the result of the calculations of core 1?

So basically the only speed up you get with using multicore systems is at rendertime?

What I'm looking at is a Dell XPS 700 (Core 2 Duo Extreme 2.93GHz, 6Gb RAM) or an Alienware S-4 7500 (same processor but only with 4Gb RAM)

[kromekat]

Well, at CB 2100 to 2500 for an 8-core Mac Pro, it will be between 3 and 4x faster than my current Dual 2.5 G5, so I guess I am in for one of those babies when they launch! :D

Adam

[imashination]

Is it true that with Particle-, Cloth-, and Dynamicsimulations multicores aren't really of any use as that can only be done by one core? As for example Particles at frame 2 are based on the outcome of the calculation of frame 1 so you can't really split it among cores as core 2 won't know what the result of the calculations of core 1?

Correct. Its essentially down to parallel or serial calculations. Most physics based stuff is serial, in that you cant work out the next frame until the current one is done. What would be nice is setting off a core to precalculate various parts, one to prepare the particle stuff, another for the cloth, one for the hair caching, another for......

So basically the only speed up you get with using multicore systems is at rendertime?


Well, it depends how hard you push your system. Half an hour ago I was burning 4 dvds at once, zipping a 16 gig file, playing music, downloading and playing a 3d game at the same time, so, the cores come in useful if you like to make your system sweat. Otherwise yes, its basically a rendertime thing.

What I'm looking at is a Dell XPS 700 (Core 2 Duo Extreme 2.93GHz, 6Gb RAM) or an Alienware S-4 7500 (same processor but only with 4Gb RAM)[/QUOTE]

Dont get the alienware, you'll lose all credibility ;) they're overpriced systems in a silly case, the dell xps arent too far behind either. Variable disco lights behind the speaker grill anyone?

[lllab]

....syes alianware isnt worth it.

i'd buy a macpro, the best system i ever had running xp;-)
(also the best speed/quality/price ratio- a c4d dream machine)

cheers
stefan

[Creature]

Sorry for hijacking this thread for my hardware questions :)

Currently looking into Mac hardware ... as I can use WinXP on them I guess it'd be fine for me. Using a lot of PC only software here.

[Srek]

What would be nice is setting off a core to precalculate various parts, one to prepare the particle stuff, another for the cloth, one for the hair caching, another for......
Sadly this all depends on one another. Just for an example take a look at the Hair TP Interaction example. If Hair and TP were calculated independently it would simply not work.

Cheers
Björn

[Creature]

So for 10.1 we could have an option in the scene-settings dialog which lets us choose if we want to spread simulations among cores and loose interaction of those, or if we want to have it computed by a single core and have TP water droplets on hair which collides with cloth :thumbsup:

:)

[Srek]

So for 10.1 we could have an option in the scene-settings dialog which lets us choose if we want to spread simulations among cores and loose interaction of those, or if we want to have it computed by a single core and have TP water droplets on hair which collides with cloth :thumbsup:

:)
Not realy, my example was only a simple one, the connections are much deeper and the consequences would make practicaly every animation that contains more then simple PSR animation useless.
Cheers
Björn

[CGTalk Moderation]

This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.