PDA

View Full Version : dual quad



titane357
02-07-2007, 11:38 AM
is there anybody that test LW, fprime, Kray , maxwell with dual quad (8 processors) ?
Do they all really use all 8 proc power ? :question:

cresshead
02-07-2007, 12:12 PM
they would for rendering but not for modeling/layout interaction

titane357
02-07-2007, 01:03 PM
I meant for rendering for sure...:D
Because I read in a PC test that maya with mental ray don't use all 8 core (just 4)...

cresshead
02-07-2007, 01:10 PM
ahh...that's not a 3d app restriction but an imposed render node liecence restriction i believe....

3dsmax's mental ray has unlimited rendering via backburner
3dsmax scanlien has a limit of 10,000 nodes
xsi foundation has 2 cpu liecences
xsi...advanced has 16 or 32 [can't be sure]

lightwave is 'restricted'' to 999 nodes

re nodes...also 'threads'...currently lightwave9.0 has limit of 8 threads for the host machine..this is just for upto lightwave 9.0 and not what ''may'' appear in the next free update [9.2]

MrWyatt
02-07-2007, 01:16 PM
I meant for rendering for sure...:D
Because I read in a PC test that maya with mental ray don't use all 8 core (just 4)...

that might be the case for the complete version but the unlimited version renders with 8. I got such a setup at work and I can say, it renders on all 8.

Captain Obvious
02-07-2007, 02:14 PM
Lightwave can render with up to eight threads. If you're using the Perspective or Advanced cameras in LW9, they'll use all eight cores. With the Classic camera, the performance will be lower.

FPrime seems to use all eight cores just fine.

Maxwell too.

Don't know about Kray.

modo screams on the eight-cores.

(Yes, I've actually used one of those dual quads; they're awesome.)

titane357
02-07-2007, 11:58 PM
dual x5355 ? :D

cresshead
02-08-2007, 12:26 AM
Hmm...where's apple 8 core macpro?....

Animapper
02-10-2007, 08:03 PM
re nodes...also 'threads'...currently lightwave9.0 has limit of 8 threads for the host machine..this is just for upto lightwave 9.0 and not what ''may'' appear in the next free update [9.2]


I'm running LW 9 on a Mac Pro 266 running winXP32 and I get 8 threads now. It's a dual dual though, not a dual quad. I haven't seen anyone with a dual quad xeon yet. I posted some render times last week that show off how fast the mac pro is. If you didn't see it, I can repost if anyone is interested. Hopefully, 9.2 will open LW up to more than 8 threads. If it doesn't, i may not move to a dual quad right away.

Regards,

Red_Oddity
02-11-2007, 04:33 PM
Well, i'm gonna find out real soon, as our computer vendor finally has those 12 dual Xeon 5355 processors for us.
(Can't wait to play with those new machines)

gerry_g
02-11-2007, 04:55 PM
Hmm...where's apple 8 core macpro?....

On my wish list ;-)

Sensei
02-12-2007, 03:35 AM
In K-Ray you set number of rendering threads by putting f.e. "threads 8" in tailer cmds string field..

erikals
02-12-2007, 06:51 AM
Some tests will show up here after some time...
http://www.blanos.com/benchmark/

Animapper
02-12-2007, 07:54 AM
so if LW only sees a maximum of 8 threads, when the dual quads come out the potential for 16 threads will not be utilized on the modeling machine? In rendering a final render perhaps we can split the machine to look like two 8 threaded machines but it would be great to have this power in preview renders.

Regards,

Captain Obvious
02-12-2007, 08:17 AM
so if LW only sees a maximum of 8 threads, when the dual quads come out the potential for 16 threads will not be utilized on the modeling machine? In rendering a final render perhaps we can split the machine to look like two 8 threaded machines but it would be great to have this power in preview renders.

Regards,
Dual quads have EIGHT cores, not sixteen. Lightwave renders just fine on all cores, at least if you use the new perspective or advanced cameras. It does not utilize it as well with the old classic camera.

Animapper
02-12-2007, 08:53 AM
Threads and cores are different here. I have the latest macpro 2.66 and I get 8 threads now (with four cores). If the dual Quad comes out it could have 16 threads; Make sense?

Are you saying the new advanced cameras can utilize more than 8 threads?

Regards,

Captain Obvious
02-12-2007, 08:54 AM
Lightwave can currently render with eight threads on ANY machine, regardless of how many cores it has.

Eight is the maximum right now.

Your machine has four cores.

Animapper
02-12-2007, 09:11 AM
So is there a diminishing return on how many threads to use when you are using a machine with 1, 2 or 4 cores? How do you see 8 cores working?

Thanks,

StereoMike
02-12-2007, 10:36 AM
No matter how many threads...
in the current "iX" magazine they tested a double quad xeon x5355 (clovertown) against a 4x dualcore opteron 8218 (santa rosa). Both 8 cores (easy math, eh?).

The intel machine reached upto ~153% of the amd in single process benchmarks, but....scaled very badly in multi process benchmarks. The amd got upto 140 % of the intel in multiprocess benchmarks (these numbers were generated wirh SPECint, SP and MP).

Funny thing, I didn't expected that. In fact they said in that test, that an 8 core intel machine won't make sense right now, cause under full throttle you get just the power of 4 cores. They recommended the amd solution for everyone who wants to crunch numbers.

cause for the good results of the opteron were the fact, that only 2 cores shared the infrastructure on the chip and experience less bottlenecks.
mike

Captain Obvious
02-12-2007, 11:04 AM
There is a near-linear increase for eight-core Intel machines when it comes to rendering. It makes plenty of sense to buy an eight-core Intel machine for rendering.

StereoMike
02-12-2007, 11:47 AM
Sure it makes sense to have many cores. It just seems to be a fact that an opteron scales much better:

from www.spec.org/cpu2000/results

CFP2000 rates on
Fujitsu Siemens Computers PRIMERGY BX630

Intel Xeon processor X5355, 2.66 GHz: 4 cores, 1 chip, 4 cores/chip
Base/Peak: 61.0 / 61.0

Intel Xeon processor X5355, 2.66 GHz: 8 cores, 2 chips, 4 cores/chip
Base/Peak: 101 / 101

AMD Opteron (TM) 285: 2 cores, 1 chip, 2 cores/chip
Base/Peak: 36.8 / 39.0

AMD Opteron (TM) 285: 4 cores, 2 chips, 2 cores/chip
Base/Peak: 72.9 / 77.6

AMD Opteron (TM) 885: 8 cores, 4 chips, 2 cores/chip
Base/Peak: 138 / 149

mike

StereoMike
02-12-2007, 12:04 PM
of course this is only one benchmark of many, and depending on the type of test the results will differ.
But this spec.org benchmark (here: cfp rates: floating point- rates) is especially built for multiprocessor benchmarking and has quite a reputation.

Animapper
02-12-2007, 12:27 PM
Well, I guess cost to build these may be the bigger factor for the near future. I don't mind using AMD at all. I've built several nice AMD rigs but I'm anxious to use anything that helps speed up my preview renders.

Regards,

Lightwolf
02-12-2007, 12:33 PM
But this spec.org benchmark (here: cfp rates: floating point- rates) is especially built for multiprocessor benchmarking and has quite a reputation.
Did they run CineBench?
The problem with SpecFP is that many of the contained benches are _very_ memory intensive (and it doesn't surprise me to see the Opteron come out first in that case).
3D rendering doesn't seem to depend as much on the memory throughput however (more or less, depending on the renderer) so the result may be different.

The only problem is that the current Cinebench uses a fairly small scene...

Cheers,
Mike

Intuition
02-12-2007, 12:48 PM
No matter how many threads...
in the current "iX" magazine they tested a double quad xeon x5355 (clovertown) against a 4x dualcore opteron 8218 (santa rosa). Both 8 cores (easy math, eh?).

The intel machine reached upto ~153% of the amd in single process benchmarks, but....scaled very badly in multi process benchmarks. The amd got upto 140 % of the intel in multiprocess benchmarks (these numbers were generated wirh SPECint, SP and MP).

Funny thing, I didn't expected that. In fact they said in that test, that an 8 core intel machine won't make sense right now, cause under full throttle you get just the power of 4 cores. They recommended the amd solution for everyone who wants to crunch numbers.

cause for the good results of the opteron were the fact, that only 2 cores shared the infrastructure on the chip and experience less bottlenecks.
mike


Which is why I am waiting for the AMD true quad core chips. They will have less bottleneck and less watt usage which is less heat while giving great and stable performance.

I am soooo tempted by the 8 core setups but I must be patient until AMD ships the quad opterons.... :rock:

StereoMike
02-12-2007, 12:48 PM
@ Lightwolf

No, just CPU2000 and CPU2006 (base, without SmartHeap libraries). But it's quite an interesting read, I recommend you read at least that article in the kiosk.

Yeah, the problem of the 5355 is the bucketload of data that comes in on too small lanes. A less memory intense test could even the field, but I doubt, that they come close, no matter what SMP test you drive.
Just a note, I cited the old 285/885 numbers, the newer 2218 / 8218 is even faster.

Lightwolf
02-12-2007, 12:55 PM
@ Lightwolf

No, just CPU2000 and CPU2006 (base, without SmartHeap libraries). But it's quite an interesting read, I recommend you read at least that article in the kiosk.
I was going to get the issue, I just didn't know it was published already...

Cheers,
Mike - waiting for Barcelona ;)

Captain Obvious
02-12-2007, 01:31 PM
According to the modo benchies I've seen, the dual quad performs well within 90% of a linear increase. If that's not acceptable, I don't know what is.

jasond
02-12-2007, 01:50 PM
2.66 Xeons from Boxx.

I couldn't wait for the better technology to bubble up, and I don't have a fetish about the hardware, so I jumped at the extra horsepower.

LW can use all 8 cores to their fullest when rendering.

The investment has justified itself already (1.5 months use) and made my life a little easier. My output is faster, I'm tackling things I wouldn't have considered before, and the clients are happy. Multitasking is much better. Every day has been a mad dash for a couple of months, so I haven't taken the time to play with nuances of optimization for renders... but then I haven't needed it.

I'd say that as long as you don't shop for computers the way you might for a wife the current monster jalopies work just fine.

Rick Hall
02-18-2007, 07:13 PM
so if LW only sees a maximum of 8 threads, when the dual quads come out the potential for 16 threads will not be utilized on the modeling machine? In rendering a final render perhaps we can split the machine to look like two 8 threaded machines but it would be great to have this power in preview renders.

Regards,


If you have more than 8 cores, then you can run screamernet to take advantage of all CPUs. I would be cool to have a render farm in one box.

Rick

Intuition
02-18-2007, 10:38 PM
According to the modo benchies I've seen, the dual quad performs well within 90% of a linear increase. If that's not acceptable, I don't know what is.

ARG!

QUIT tempting meeeeee!!!!

Between you and JasonD making these statements my wallet is going to bleed again..... stop.. I beg of ye :devil: :devil: :devil:

Ivan D. Young
02-19-2007, 01:44 AM
Quad cores with hyper threading should be released in the second half of 2007. So being able to render with 16 cores will be a reality very soon. but what is interesting is that both Intel and AMD seem to think that 16 cores will be the limit in the near future. something about load balancing and all that sort of complicated info sharing. I can't wait to get an 8 core myself though.

Lightwolf
02-19-2007, 01:48 AM
Quad cores with hyper threading should be released in the second half of 2007.
Hm, last time I read about it intel denied plans to add HT this year. But times are interesting nonetheless.

Cheers,
Mike

Captain Obvious
02-19-2007, 02:46 AM
Maybe Intel will ask IBM for help to make their SMT suck less? :)

Lightwolf
02-19-2007, 05:02 AM
Maybe Intel will ask IBM for help to make their SMT suck less? :)
Well, they are working on a HyperTransport like technology (CSI or something like that). Why they just don't use HT I don't know...

Cheers,
Mike