PC renders a lot slower than my MAC! Why?

[beagle]

Hi,

I've been using Lightwave at work for a while now, on a 800mhz MAC G4.
Not the fastest thing around but not too bad either.
I've recently installed Lightwave (7.5) on my new PC at home to do some work over the weekend. It's an AMD 2500 Barton CPU running at 2Ghz.
The problem is, Lightwave's render times are at least double that of my (slower) MAC. Consequently no work was done :(
Any ideas on why this could be? I'm running it on Win XP, and it has a 64MB ATI graphics card.
Are there any tricks to speed up Lightwave under Win XP?
Also, I can't get Eki's Plug Pak (overcaster etc) to work. It installed fine, and everything appears to work (sliders etc) except that no light is actually emmitted. Nothing show in the preview or renders. Has anyone had problems with this on the PC version of Lightwave?

Thanks in advance for any ideas.

[Mylenium]

I'd say this may be due to poor configuration of your XP or technical problems with your PC in general. How much RAM do you have on the PC? If it's only 256 or even 128 Meg that may not be enough 'cos Win will try to swap to disk (XP needs lots of RAM for itself). Try shutting down any unnecessary services in XP (Media, Index, Network, Security etc.) Also LW is not optimized for AMD so it will not take full advantage of its power.

Mylenium

[beagle]

thanks for the advice Mylenium. I have 512MB of PC2700 Corsair RAM.
Windows is pretty slim already as my PC is optimized for audio applications like Steinberg's Cubase. However, this shouldn't interfere with anything else, and Photoshop runs like a dream.
Maybe it is just an AMD/Lightwave issue.

[mattclary]

Doubtful. Go look at www.blanos.com and see how these machine SHOULD compare in render speed. Have you copmpared any of the benchmark scenes? This will let us base our advice on known quantities.

[beagle]

Thanks for that Matt, I'll look into that this week and post back when I've got some results.

[Extent]

I don't know why people insist on saying that LW isn't optimised for AMD chips! I specifically remember a press release stating such optimisations *before* it was every done for the P4. The reason Intel chips are faster is because they have superior extra instruction sets. SSE2 and Hyperthreading. LW is optimised to take advantage of 3DNow! and 3DNow Pro that's in the AMD chips but they are AMDs answer to intels much older SSE instruction set. When Opteron is out it will have support for SSE2, and all the optimisations for that will help make that run faster as well. It's a matter of hardware support, not that the software favors one over the other.

[mattclary]

Extent, granted, the word "optimized" may be overused. It would be more accurate to say LightWave uses SSE2 which makes it render faster in the Intel platform. Of course, LightWave is making the most of what's available with that processor.



http://dictionary.reference.com/search?q=optimize

op·ti·mize ( P ) Pronunciation Key (pt-mz)
tr.v. op·ti·mized, op·ti·miz·ing, op·ti·miz·es
To make as perfect or effective as possible.
Computer Science. To increase the computing speed and efficiency of (a program), as by rewriting instructions.
To make the most of

[Antimatter]

another thing to keep in mind isn't those mac g4 dual cpu system correct?

so you got two 800mhz cpu chugging along in that g4

then another limitation on your render speed is memory bandwidth.

don't buy cheap motherboard or memory chip they often have horroriable memory bandwidth, for example my old 1ghz p3 system has an via mobo and some no name memory but the memory bandwidth was barely 1 gig/sec which is terriable, anyway my new system a 2.8ghz p4 with 1 gig of corsair memory will have around 6 gig/sec memory bandwidth which is much better.

anyway i would highly recommend at minum of 512 meg of ram if not 1 to 2 gig of ram for an windows machine.

[Beamtracer]

Love your avatar, Antimatter!

A Mac should be about 50% faster than a Pentium at the same clockspeed, due to differences in the way the PowerPC processor works.

So an 800MHz PowerPC might be about the same as a 1200MHz Pentium, depending on how well the software is optimized.

On the other hand, Pentiums have higher clockspeeds, so you'd expect a 2GHz Pentium to still be faster than a 800MHz Mac.

Dual processing will add to the speed (whatever the platform) but it is never twice as fast. There is always some overhead in multiprocessing.

[Antimatter]

hehe thanks, i found it one day and thought, bingo that's so perfect for me :)

now about the overhead, i wouldn't think it would be that bad on an modern day processor plus if the overhead was like what 50% off processing speed, then... it would be very nasty for those clustered supercomputers, anyway i think its more like 1-5% not sure. anyway i think the overhead is more related to memory or something.

now about the way that the PowerPC processor work, similar thing with the athons vs intel's cpu,

i got a couple of athons and theyre around what 1.5 to 2 ghz and they run almost as fast as my 2-3 ghz intel cpu'


now about the clock speed its not the end all of all, i was boried and did some reading up on the cray's series of supercomputers, and i found out that their cpu ran at about 300 mhz, anyway their pipeline is just fater than most personal computer, like good example is the athon and intel cpu thing.

athons got an nice and fat pipeline so for each clock cycle they execute more instruction. intel on other hand got thin one so they only execute a few instruction so they got to really ramp up the clockcycles.

[Extent]

The overhead on a dual system isn't bad, like you say. Superclusters are a little different tho, they don't run desktop applications on them, and the sort of process intensive software they're running is written for that kind of massive clustering. The longer you can make a thread last the less allocation you have to do, less overhead you run. The problem with memmory and mulitiprocessing is that you've only got one northridge controller (unless you've got a dual Opteron system :D ) so both processors can't be accessing memmory at the exact same time.

I did a rough raytrace benchmark and found that with two processors it renders at 53% speed of a single. (rough estimate only as I didn't dedicate the machine to the bench but it still gives the idea)

as far as IPC raitings for each chip (if I remember correctly) Intel does 4, AMD does 7, and Macs do about 20 ins


mattclary: Right. By definition LW is optimized for AMD. Intel has a superior special instruction set. Saying it wrong makes it look like Newtek dosen't care about AMD, which isn't true.

[beagle]

Cheers guys,

According to the benchmarks at blanos.com, my machine is indeed slow at it's default settings (1.83Ghz).
However, if I overclock to 2.1-2.2Ghz, the benchmark tests show times that are correct for that speed of machine. very odd.

So, I've found a workaround, although OC'ing my chip is a bit risky in this heatwave we're having here in the UK at the moment!

[Antimatter]

extent: thanks for that info on the instruction per clocl cycle. so from the looks of it, if you could overclock those mac proc you would have one damn good system :)

anyway 53% redcludition in time is pretty damn good, its not an perfect 50% but still 3% aint that bad but i'm not sure but i think there's a few motherboard out there that has an different build for it, i think its for the intel xeron's but anyway from what i heard about the motherboard, is that you HAVE to have at minum 2 stick of memory because one processor gets one set of memory, the other processor gets the other set of memory. anyway if you wanted to have dual channel support you would need four stick of ram.


beagle: i'm sort of guessing here but i have no idea what is happening with your machine but i'm guessing it has to do with the motherboard or something,

[facial deluxe]

May I suggest that you play with multithreading ?
1-2-4-8 give very different result.
I have a dual 2.6Ghz Xeon, on a very simple scene, rendering typos flat 2D polys, I have a gain of 500% going from 8 thread to 1 thread !!!!
On rather heavy scene though (raytrace, SSS, more polys), 8 threads gave me better results, around 200 %