More about preliminary Maya testing on dual 2GHz Power Mac G5

[mark_wilkins]

BTW, note that rendering (which is where the dual G5 will do the best, as it's multithreaded) is also where you really need that extra speed kick most.

I also believe that when Alias recompiles Maya with G5 optimizations, as they're likely to do, the machine should do somewhat better. At the moment the performance numbers are probably similar to the 2 GHz dual Xeon.

Also, does anyone but me think it's something that the 1533 MHz Athlon 1800+ comes close to the performance of the 3.2 GHz Pentium 4?

-- Mark

[alphatron]

Hi Mark,

AMD chips have always done extremely well in relation to P4's when running Maya. Maya is not really optimized for P4's (SSE2, etc..), which is why the AMD can compete with it. In programs like Lightwave or Max, which have strong P4/SSE2 optimizations, the AMD chips, usually do not come close.

BTW, are you testing with only one G5 configuration? If yes, what are the exact system specs of this machine?

[mark_wilkins]

I posted the specs:

Power Mac G5 Dual 2GHz
2 GB DDR400 RAM

the spec I didn't mention (because I don't think it has much impact on the numbers) is that it's running an ATI Radeon 9800 Pro video card.

Since the tests aren't multithreaded, only one processor is used for most of the work.

(By the way, why don't they optimize better for the P4? Is it that they want to sell MIPS boxes? :D) (kidding!)

-- Mark

[GregHess]

Mark,

Thanks for taking the time to do this.

Though not many have commented, I'm sure its greatly appreciated.

[dmeyer]

Thanks in advance Mark...mine is delayed until Oct 21! :annoyed:

[mark_wilkins]

I'll bet you get it earlier.

-- Mark

[ZrO-1]

:thumbsup: Awesome stuff Mark, I can't wait to see what your numbers turn out like. And thanks for the links too. Your book never leaves my side when I'm working in Maya.

[mark_wilkins]

Don't have the G5 numbers yet, but here are the results for a dual 2.8 GHz Xeon running Red Hat 7.2.2, 2 GB RAM, using the Tehama (i850) chipset:


wrap total: 6.58721
solid total: 24.2328
particle total: 22.4773


wrap dg: 6.21468
solid dg: 20.8653
particle dg: 13.6751

The wrap deformer was half the time of any of the other machines! Could it be threaded? Have to refresh my memory on the G5 results to see!

-- Mark

[kromekat]

Thanks for posting this Mark, I look forward to seeing the rendering figures!

[dmeyer]

Originally posted by mark_wilkins

The wrap deformer was half the time of any of the other machines! Could it be threaded? Have to refresh my memory on the G5 results to see!

-- Mark

Hopefully the horrid optimization of Maya on the Mac has been improved in 5.0. I remember the good 'ol version 3.5 that was neither MP aware nor AltiVec optimized.

[alphatron]

When approximately do you think the G5 scores will be up? Do you think you'll be able to test some Athlon64 FX's in the near future as well?

Thanks for keeping us updated!

[alphatron]

Edit: Double post.

[mark_wilkins]

OK, I have the dual 2 GHz numbers up on the benchmark results page.

The particle and solid body sims scale a little better than clock speed. The performance of the G5 is decent, but a ~3 GHz class Intel processor is faster.

However, the wrap deformer is a good example of what must be multithreaded code: the dual Xeon and the dual G5 both tore through it. The G5 beat the Xeon by a comfortable margin, too!

I've added rendering numbers for this scene:

http://www.melscripting.com/glassinroom.mb

(sorry, it's not a very appealing scene -- just thrown together for the test.)

I'm running the G5 test right now and will update when I have a number. As it is, though, you can see that Hyperthreading provides a nice speed increase for the 3.2 GHz P4.

It seems pretty clear to me that IF you're able to run multithreaded code, the G5's performance is pretty strong. (Unfortunately, I won't be able to run the render on the Xeon.)

If someone else wants to, here's how to run that render:

* Place the .mb file in a project's scenes directory

* Type


mayarender_with_mr -proj <project> -file <project>/scenes/glassinroom.mb


replacing <project> with the path to your project directory.

Unfortunately, I don't have access to an Athlon 64 and I'm not planning to buy one. If someone here wants to run the tests, though, I'll be happy to put up their results.

Remember that these are not great benchmarks. If all you do are wrap deformers, the G5 may be a great performer, but this test says nothing about, say, lattices. That said, the G5's occasional exceptional performance suggests that it's probably a great overall choice for Maya depending on what you're doing.

Also, the G5 is known to be optimization-sensitive. Optimizations that may come in future versions of Maya could help this out quite a bit. (No guarantee that such optimizations are forthcoming, but I'd be surprised if it weren't at least a topic for discussion...)

-- Mark

[alphatron]

Interesting results Mark. Judging solely from the wrap deformer test, it looks like the G5 handles dual processing very efficiently. What has really surprised me though is how well the Athlon XP 1800 holds up to the G5, and all the intel systems as well! especially since it's system only has 768mb's of RAM.. that's one cheap, and still competitive machine! Incidentally, it's also one of the processor I use for 3D and compositing at work.

[mark_wilkins]

OK, I put up the render numbers. The dual G5 renders that scene in 64 min. vs the 3.2 GHz Pentium 4's 84, which means the G5 is doing somewhat better per-clock. Doubtful that a P4 Extreme Edition would catch up, and hard to say about a dual Xeon, but Athlons are probably still the king of the hill, so to speak...

-- Mark

[alphatron]

Will you be doing the render test with the dual 2.8?

[mark_wilkins]

Unfortunately, as I said before, I won't be able to run the render on the Xeon.

-- Mark

[danylyon]

I just ran your bench on my machine at work.

Dual Xeon 3.06Ghz (533Mhz FSB)
2GB of RAM (Kingston)

46.44 min

I used an existing project.. I hope that was correct?

Edit: on Windows XP

[1000101]

Mark; I have access to a couple of dual xeons here in my office. If I get some spare time I'll run your bench and get you some numbers.

I will also compare linux vs windows on the systems to rule that out on your wrap deformer test.

I'm still at a loss as to why my particles time is lower than everyone elses on my personal machine though hehe.

strangeness.

[alphatron]

danlyon,

wow, 46 minutes.. that slaps the other P4's and the G5 silly!

[dvornik]

For some reason the page with the results is displayed strangely formatted - it's hard to read.

[mark_wilkins]

Yeah, I'm not too surprised at 45 min. or so for a dual Xeon 3 Ghz.

Mental Ray (because the threads can often be truly independent) seems to cut the time just about in half when you run on two procs, and since my P4 3.2 was getting around 84 min, that number's in the ballpark for what a dual could do with it.

The real issue for the G5 right now is that there's no optimization in Maya. All the cases I've seen where people have presented benchmarks where the dual 2 GHz G5 outperformed dual 3 GHz Xeons were with code that had been carefully G5-optimized using either IBM's XLC compiler or manually with Apple's Shark tools in combination with the PPC970 flags in GCC 3.3.

I have no idea whether a G5-optimized version of Maya's in the cards, but if it is then that might goose some of this performance by as much as 20-30%, which would put the dual in the ballpark of the 3 GHz Xeon.

At the moment, though, I'd put it like this: if sheer rendering throughput for Maya is your sole standard, the G5 is not the best choice out there. If price/performance is your standard, AND you don't want to build yourself, the G5 is pretty good, but again not the best. On the other hand, if you're attracted to the Mac platform for its UNIX-like environment, free development tools, Final Cut Pro, or whatever, the performance of the dual G5 is definitely good enough for it to be an attractive solution.

-- Mark

[mark_wilkins]

Originally posted by dvornik
For some reason the page with the results is displayed strangely formatted - it's hard to read.

Yeah, umm thank Microsoft Excel for that.

(I don't know why MS Excel can't produce web pages that look good on Internet Explorer. Come on, let's see some monopoly power here!)

-- Mark

[dvornik]

Could you please run some viewport benchmarks on a G5 as well while you're at it? Like SpecAPC, testcenter or your own.

[mark_wilkins]

I don't have any such benchmarks, and I'm not sure how I'd go about writing one.

Anyway, wouldn't such things be limited by the video card rather than the CPU? Video card performance is very comparable to the PC versions of the same cards, when stuff is GPU bound.

-- Mark

[dvornik]

I have a very strong impression that maya viewport is extremely processor-dependant. Obviously it's hard to expect anything groundbreaking from a Radeon, but still.

There are no maya for mac viewport benchmarks published anywhere, so it is very interesting to see some.

http://www.specbench.org/gpc/apc.static/maya5.html
http://www.maya-testcenter.de/

[alphatron]

Remember that openGL is not multithreaded, so you won't really see any performance gains on a dual proc rig in the viewports.

[dvornik]

Well, you can see an Operton with 980XGL beating P4s 3.2 with FX1000 on the official SPEC results page. By a rather significant margin. So who knows where G5 would stand.

http://www.specbench.org/gpc/apc.data/apc_maya50_summary.html

[dvornik]

OK, if I understand the test correctly here are my results (from "Time since last global timer reset:" in the script editor window):

Athlon XP 1900+, 512 RAM, A7V333, Quadro4 750 XGL:

wrap total: 11.9849
solid total: 18.1117
particle total: 17.4187

wrap dg: 11.6436
solid dg: 14.0534
particle dg: 13.9314

Are you sure the time is not affected by the video card? Otherwise why my 1900 would beat your 1800 by so much on the solid test? And... umm... What's "dg"?

[edit] As a possible proof that the test is affected by the video card here are the results in "compatibility" OGL mode (the slowest one) on dual monitors:

wrap total: 12.223
solid total: 21.8459
particle total: 27.7139

wrap dg: 11.7295
solid dg: 14.9814
particle dg: 13.8668

So wrap looks like the most processor-dependant test and G5 is way ahead of the rest on that one. I believe we need more testing to come to some solid conclusions, it's very helpful that you wrote this script.

As a total speculation I would expect some very decent viewport performance from a G5/Geforce combination.

[mark_wilkins]

I'm speculating that video performance isn't such an issue because the time doesn't get any better when you cover up almost all of the viewport with another window. Usually, OpenGL culls objects that are not visible, so that should greatly accelerate drawing.

Maybe it's doing stuff like caching geometry in video RAM or something??

-- Mark

[dvornik]

The easiest way to test it would be to run the test on the same machine with different video cards. I may do it over the weekend.

And i don't know if you can make sense of this page on particles from that famous 3dchips article (all i could come up with is "Nvidia cards suck at particles" - meaning that particles are videocard-dependant):
http://www.3dchips.net/content/review.php?id=63&page=15

[Thalaxis]

Maya uses OpenGL to render particles, so maybe that explains
the particle issue.

The viewport coverage issue might be a result of the way that
Quartz works; it's based on a realtime compositor, so it may
actually be rendering the viewport even when it is covered. The
OpenGL renderer can't change its behavior if it doesn't know that
it is rendering to an obscured viewport.

Both of those are just theories, of course :)

[dvornik]

If I'm not mistaken my render result for that glassinroom on a 1900+ is 2:16:41 (137 min). That's a lot. It was in "MSG 0.0 info : wallclock 2:16:41.86 total" - am I right?

It seems like different processors may be good for different tasks. I can't think of a reason why old Athlons would get scores like that in supposedly processor-intensive viewport tasks though (the first test).

Would be nice to try this rendering test on a network renderer.

[mark_wilkins]

Originally posted by Thalaxis
The viewport coverage issue might be a result of the way that
Quartz works; it's based on a realtime compositor, so it may
actually be rendering the viewport even when it is covered.

The way OpenGL works I can't imagine how you could have an OpenGL viewport that were rendered that way. If you did it would be dog slow.

-- Mark

[dvornik]

OK, so here are the results from the same PC using different video cards. It's an old P3 1GHz, 768 gigs of RAM,

Matrox G400 Max / Quadro4 750 XGL

wrap total: 38.0656 / 35.8446
solid total: 56.3517 / 47.1088
particle total: 55.1844 / 42.5435

wrap dg: 36.9636 /35.0265
solid dg: 41.6422 / 39.4878
particle dg: 34.615 / 35.7606

It definitely proves that the realtime solid and particles tests depend on the videocard as well the processor. I still have no idea what "dg" is though.

Here are the testcenter results to demonstarte the videocards performance difference:

wire 8.4 / 39
shaded 1.8 / 5.2
texture 2.1 / 8.2
t&l 0.6 / 6

[mark_wilkins]

DG is the time spent calculating the Dependency Graph while total time includes time spent doing other things like servicing display redraws.

-- Mark

[dvornik]

Thanks. So the Dependency Graph results are the most accurate for the processor-intensive viewport tasks, as far as I can tell.

I think AMD processors (even the old ones) perform surprisingly well in these tasks. That's kind of unexpected. Why would that happen? BTW SPECapc runs particle and deformation tests, so that's probably the reason for the high Operton scores.

G5 seems to be on the level of the best PC processors so far. Maybe not "the best" or "the best value" but it's totally holding it's ground and it's performance is rather balanced between viewport and rendering.

[murtle]

Thanks to Mark and everyone for benchies.

My results are;

wrap total: 10.344
solid total: 27.5724
particle total: 17.8772

wrap dg: 9.66997
solid dg: 23.5909
particle dg: 13.3598



Epox 8RDA+
1700+ @ 2GHz (166x12)
2x512 MB Kingston PC2700 (2.5-3-3-7)
ATI Radeon 8500 (275/275) Catalyst 3.5
Win2000 SP3
Maya5


I wonder how it's effect Quadro FX1000 on solid benchmarks?

[mark_wilkins]

Originally posted by dvornik
G5 seems to be on the level of the best PC processors so far. Maybe not "the best" or "the best value" but it's totally holding it's ground and it's performance is rather balanced between viewport and rendering.

The thing is that raw performance is rarely the only reason to go for a particular platform. The things that make me happier using my dual G5 Mac for Maya than the PC are:

* One system for editing and DVD output, for which I use Final Cut Pro and DVD Studio Pro.

* UNIX-like OS, with easy access to Perl and Python.

* It just happens to be meaningfully faster than my next-best machine, the P4 3.2 GHz.

None of this is meant to be trashing the Windows PC. After all, I have a nice one.

If I were a big Avid user, had the inclination to install a different shell and scripting tools, and wanted to get a top-end dual Xeon or Opteron, I could have these same advantages on the PC side. I've never liked Windows much, though, and going for those options would not realistically have saved me any money.

-- Mark

[dvornik]

Hey, fair enough. I'm very happy to see fast Macs.

I have to mention that apple bought and killed the direct PC competitor to DVD Studio Pro (Spruce with it's DVDMaestro) - so there's a lot of dirty business going on on both mac and PC sides.

[alphatron]

We still use the last version of Maestro for DVD authoring at work, and prefer it over DVD Studio Pro for the moment, which we also own.

[mark_wilkins]

Have you tried DVD SP 2? I understand it's a significant update, but I haven't tried it.

-- Mark

[alphatron]

Yes we own DVD SP 2, and we like using Maestro more because of the added level of control.

[dmeyer]

Originally posted by mark_wilkins


* One system for editing and DVD output, for which I use Final Cut Pro and DVD Studio Pro.


Final Cut Pro is reason enough alone for me to keep a current Mac. :)

[mark_wilkins]

Originally posted by alphatron
Yes we own DVD SP 2, and we like using Maestro more because of the added level of control.

Can you describe the differences? (i.e. what can you do with Maestro that's difficult or impossible on DVD SP 2.)

I'm not familiar with either one.

-- Mark

[alphatron]

Well, this isn't really the thread to be talking about this, as it's already gettting OT, but if you'd like I will PM you a comparison.

[Saurus]

Originally posted by mark_wilkins
The thing is that raw performance is rarely the only reason to go for a particular platform. The things that make me happier using my dual G5 Mac for Maya than the PC are:

* One system for editing and DVD output, for which I use Final Cut Pro and DVD Studio Pro.

* UNIX-like OS, with easy access to Perl and Python.

* It just happens to be meaningfully faster than my next-best machine, the P4 3.2 GHz.



I could be wrong, but isnt Maya's Mac version is only available in Complete version, not Unlimited version?

Saurus

[mark_wilkins]

Originally posted by Saurus
I could be wrong, but isnt Maya's Mac version is only available in Complete version, not Unlimited version?

Sure is, but my use of fluids, Maya Live, cloth, and fur is limited.

-- Mark

[Saurus]

Originally posted by mark_wilkins
Sure is, but my use of fluids, Maya Live, cloth, and fur is limited.

-- Mark

Almost all the studios I worked for, we always used most of Unlimited components beside Maya Live. I don’t think we can't work properly without
those components.

Saurus

[mark_wilkins]

I'm talking about home use here...

Anyway, most studios I know of have a mix of complete and unlimited. Depends on what they're doing.

Where it might make a lot of sense to run Macs in a large studio with Maya is in, for example, a matte painting department where they're expected to do some 3D work but are mostly living in Photoshop, or in a layout department doing animatics with quick turnaround to FCP 4.

Also, there are 3rd-party hair and cloth solutions available for OS X, if a studio wanted to head down that route.

-- Mark

[tyson12zoll]

Originally posted by mark_wilkins


Also, there are 3rd-party hair and cloth solutions available for OS X, if a studio wanted to head down that route.

-- Mark

e.g Syflex kicks maya cloths bootay

However, nothing on the mac can come close to fur, IMO

[beaker]

Originally posted by tyson12zoll
e.g Syflex kicks maya cloths bootay

However, nothing on the mac can come close to fur, IMO
Shave and a Haircut?

[tyson12zoll]

Originally posted by beaker
Shave and a Haircut?

They are both great tools. I had the impression that fur was more powerful, though i've never used either.

[Marc Andreoli]

I don't think Shave and Haircut is available for OSX at the moment... but once it is released you can get maya complete, syflex and Sn'H for about $5000. That is still $2000 less than unlimited. So you could throw in one of those sexy 23" cinema displays with the money you 'save'...

If you don't need fluids and live, that looks like a pretty nice package to me ;)

[benvu]

Does anyone know if and or when there will be a Quadro FX 2000/3000 caliber graphics card for the PowerMac G5?

This would make the 'switch' over to Mac more worthwhile.

Hopefully the card will have dual DVI ports instead of one DVI and one Apple ADC port.

STILL WAITING FOR MAYA 5 UNLIMITED FOR MAC!

[agreenster]

Mark_Wilkins:

Quick question--do you know if Maya 4.5 OS X is Multi-Processor and AltiVec aware? Im considering getting a G5 but I have Maya 4.5, not 5.0. Do you think the performance will still compare?

Thanks! :wavey:

[mark_wilkins]

I'm not sure to what extent, if any, Maya is vectorized. Probably not much, as AltiVec does not support double-precision floating-point arithmetic, which is necessary for most of what Maya does.

However, 4.5 definitely will support multiprocessor rendering with Maya's software renderer.

-- Mark

[agreenster]

Thanks Mark

I've been on these boards for a while now, but just now noticed theres a small Mac community here. (I mostly hang around the gallery sections, getting motivation from everyones work)

So I checked some of your recent posts and learned a lot from just skimming the discussions. Thanks for spreading your knowledge to us "lesser" CG'ers (aka -animators! ;))

I do think Im gonna get one of dem dual G5s........:drool:

[eple]

Originally posted by benvu
Does anyone know if and or when there will be a Quadro FX 2000/3000 caliber graphics card for the PowerMac G5?

This would make the 'switch' over to Mac more worthwhile.

Hopefully the card will have dual DVI ports instead of one DVI and one Apple ADC port.

STILL WAITING FOR MAYA 5 UNLIMITED FOR MAC!



Would you like a Wildcat instead??? .:buttrock:
Ok, this is purely rumor, but read this:
http://www.appleinsider.com/news/

Also, please send 3D labs (Steve.Cox@3Dlabs.com ) an email letting them know how important this is to all of us Mac enthusiasts.

[OC-NightHawk]

Originally posted by mark_wilkins
I also believe that when Alias recompiles Maya with G5 optimizations, as they're likely to do, the machine should do somewhat better. At the moment the performance numbers are probably similar to the 2 GHz dual Xeon.

Thats it just a 2GHz Xeon? Thats only one step from the bottom of the barrel?! Are you sure? This is in a dual G5 with the option to use all available processors on? Ouch. :hmm:

Pats dual 2.93GHz (OC'd) Xeon main rig .... I've got all the rendering power I'll need for a while.:buttrock: And it cost less than 2 grand 6 months ago.:thumbsup: Man I've got to up my voltage so I can get my processors to 3.2GHz.

Edit: next quarter SCAD will have some G5's set up along with some HP x8000 workstations (Dual 3.06GHz Xeon). I can't wait try them out.:bounce:

[mark_wilkins]

Alias has to do a bunch of things to get Maya properly G5 optimized and they're not that easy, so it may be a couple of releases. Applications that are properly optimized do much better for the clock speed.

However, even now my dual 2 stomps my 3.2 P4 for rendering and cost about the same in September, when I bought both, so no complaints.

-- Mark

[OC-NightHawk]

Originally posted by mark_wilkins
Alias has to do a bunch of things to get Maya properly G5 optimized and they're not that easy, so it may be a couple of releases. Applications that are properly optimized do much better for the clock speed.

However, even now my dual 2 stomps my 3.2 P4 for rendering and cost about the same in September, when I bought both, so no complaints.

-- Mark

The top end intel stuff is always marked up a lot . :beer:

I had to work my butt off building and selling 5 machines to get enough money to buy what I got now. At the time there was no G5 just the G4 :cry: I looked at Dells site and just about died. My machine without the ability to overclock was in the ball park of $3000. :surprised Then I found out newegg sold xeon parts and I saw my chance.

I wonder if Alias will optimize Maya 5 for the Mac and give it free as a update or if they'll pawn it off as Maya 5.5. Hoping for the later. My one gripe with Apple is their refusal to stop selling one button mice.

Edit: must have been tired, when I said the optimised either being 5.5 or 5 and hopin for later, I meant I hope its not the later. What a nasty typo.

[mark_wilkins]

ability to overclock

Well, I like building my own systems as much as the next guy (which is what the P4 was for), but overclocking is just too much of a risk to take for my work machine. (I did get the P4 up to a stable 3.6 GHz for testing though, that was fun, but I just couldn't bear to leave it that way.)

I will say one thing, though, both the P4 3.2 (running Windows XP Pro) and the dual 2 GHz G5 (running OS X 10.3) are the most stable non-linux systems I've run in many years.

My one gripe with Apple is their refusal to stop selling one button mice.

They have mountains of studies showing that new computer users don't know what to do with the extra buttons, and the system's built so that you don't really need them, except for certain ported applications like Maya. Besides, it's a $10 upgrade, so who cares?

-- Mark

[OC-NightHawk]

Originally posted by mark_wilkins
ability to overclock

Well, I like building my own systems as much as the next guy (which is what the P4 was for), but overclocking is just too much of a risk to take for my work machine. (I did get the P4 up to a stable 3.6 GHz for testing though, that was fun, but I just couldn't bear to leave it that way.)

I will say one thing, though, both the P4 3.2 (running Windows XP Pro) and the dual 2 GHz G5 (running OS X 10.3) are the most stable non-linux systems I've run in many years.

My one gripe with Apple is their refusal to stop selling one button mice.

They have mountains of studies showing that new computer users don't know what to do with the extra buttons, and the system's built so that you don't really need them, except for certain ported applications like Maya. Besides, it's a $10 upgrade, so who cares?

-- Mark

Yeah I hear you. I slowly and I mean slowly increased the multiplier back up to 22 after forcing the 533MHz FSB. I did it over a period of a week making sure that it didn't do anything out of the ordinary for 24 hours. It made it all the way to 22*133 and remains stable as a rock. The machine folds around the clock, about 99% gromacs so I'm reasonably sure that the stability remains. Of course the second I try to make it 23*133 it gets flaky under stress and reboots. It won't even post at 24*133. I know because I had the misfortune to reset the bios for some stupid reason and forgot to place the jumper back into default fsb. So when it wouldn't post I freaked until I realized it was trying to boot at the spead I can only dream of it doing without volt moding it. That was scary, I thought for sure I had a huge paper weight on my hands.

In regards to the stability I think its got a lot to do with the newer OS releases for both platforms. Windows 9X for desktops might as well have been a paper cup in a tub with no bottom. OS9 wasn't much better in terms of stability. Prefrences aside OSX and XP have done a lot for building up the reputations of both companies. I haven't run unix yet mainly because I've had no reason to, but I've been tempted to install linux as a secondary OS for my machine after the debacle with houdini on windows. I tried a network render and lost all of my shaders and the ones that worked where rendered wrong. The linux copy at the labs ran better so I think I'll look into a non com for linux in replacement of the windows one.

In regards to the mouse, I use the right button all the time when I'm too lazy to use keyboard short cuts. Not to mention I couldn't imagine using a modeling package without at least a two button mouse.

:hmm: looks like I've derailed the topic. Crap. Sorry about that. Ok enough about the PC.How is the Mac version in terms of MEL scripting and such? Is it a good port? Is it a port from the Irix, linux, or windows build?

[mark_wilkins]

Originally posted by OC-NightHawk
How is the Mac version in terms of MEL scripting and such? Is it a good port? Is it a port from the Irix, linux, or windows build?

There are some weird issues when dealing with Maya environment variables and system() commands (just like in Windows.) Otherwise, MEL works great, and actually the OS X port does not have some UI bugs that make the Windows version irritating.

-- Mark

[ristopuukko]

for the record:

dual AMD Athlon MP 2800+
Tyan Tiger MPX S2466 266 FSB
2G Kingston Reg ECC

the parts I mentioned above affect most to the actual math being done (or I atleast think they do) with glassroom.mb so I spare you the details of my setup so...

test scene rendering time 52:20


/risto

[ristopuukko]

Originally posted by mark_wilkins
... Otherwise, MEL works great, and actually the OS X port does not have some UI bugs that make the Windows version irritating.

I ran those mel - benchmarks (which was really illuminating -
thanks, Mark ), and I was wondering (just out of the blue) has
anyone tried that SPECapc Maya Benchmark on G5 ??

/risto

[mark_wilkins]

I have not tried the SPECapc benchmarks. However, I want to point out that the tests I posted above don't benchmark anything specific to MEL -- they spend almost all of their time doing the same things you'd be waiting around for interactively.

-- Mark

[ristopuukko]

Originally posted by mark_wilkins
...However, I want to point out that the tests I posted above don't benchmark anything specific to MEL -- they spend almost all of their time doing the same things you'd be waiting around for interactively...

aa, yes

I was a bit misleading here (as usual) - I'm finnish so my english
doesn't always output my thoughts as it is supposed to do :-)

I don't think the SPECapc benchmark does justice to a system
anymore than any other benchmark (except of course your clever
MEL thingies !!) but as I'm seriously thinking of getting in to the
OSX-camp, i'm interested in any test results (real life or simu-
lation) that tells something about the G5

I guess the best solution (I've been trying to avoid this) is to
have separate workstations on 3D and 2D work

the Mac - camp in my neighbourhood is very limited so you've
been more than helpful

/risto

[matty429]

Wouldn't a dual 2ghz Xeon Out render
a Singe 3.2 as well?

I'm not un justifying you're time spent on theese benchmarks...but wouldn't it be a fair comparison to only use one cpu on the G5 ...since all other comparisons only have one cpu?

[mark_wilkins]

I can only compare machines to which I have access. That's one reason that I made the tests themselves available.

However the Xeon in the table is a dual 2.8 GHz (a fact I failed to note on the table, but which I just added). It's faster than the P4 3.2 on the wrap deformer test but slower on the solid and particle dynamics tests because those operations either aren't multithreaded or aren't aided by multithreading.

I was unable to test the Mental Ray render on the dual Intel 2.8 because of MR installation problems on the machine on which I was running. However, I'd expect it to be about 5 to 10 seconds faster than the G5 based on other people's reports.

On the other hand, it's unclear how things would work out -- I note that the (multithreaded) wrap deformer is faster on the dual 2 G5 than it is on the dual 2.8 Xeon, which I'd say implies that either memory bandwidth or some efficiency issue related to multithreading is at work. Or maybe it's just an accident of the code interacting well with the G5 architecture?

BTW I realized after the fact that OS differences were not represented on that table. I've added them too.

-- Mark

[mark_wilkins]

By the way, this is a little off topic, but if you'd like to discuss specifics of using Maya on the Mac, please stop by the new forum I've kicked off at forums.melscripting.com:

Maya for Mac OS X Discussion
http://forums.melscripting.com/index.php?showforum=10

-- Mark

[Matt]

I'm interested like most others, in seeing fair results, in a comparison of 64bit vs. 64bit.

So that leaves it to:

G5 vs. Itanium vs. Opteron vs. AthlonFX vs. Athlon64.

I'm positive that the AthlonFX system will beat out everything else.

Edit: Clarification, Itanium2.

[beaker]

Originally posted by Matt
I'm interested like most others, in seeing fair results, in a comparison of 64bit vs. 64bit.

So that leaves it to:

G5 vs. Itanium vs. Opteron vs. AthlonFX vs. Athlon64.

I'm positive that the AthlonFX system will beat out everything else.

Edit: Clarification, Itanium2.
The only true test of that would be Prman since it is supposed to be 64 bit clean. Atleast for G5, they haven't said anything about 64 bit clean versions for Athlon64/FX yet. All the other software packages are just 32 bit with 64 bit enhancements.

[mark_wilkins]

Originally posted by Matt
I'm interested like most others, in seeing fair results, in a comparison of 64bit vs. 64bit.

There may be other ways to do this test, but you can't do it for Maya, as there is no 64-bit version of Maya.

As the previous poster pointed out -- maybe PRMan?

Anyway, for real-world users (as opposed to the benchmark-obsessed :D) the main performance question about the G5 is whether it's close enough to make Apple a compelling choice for the combination of included development tools and a UNIX-like environment (like Linux), and easy access to support and popular commercial applications (like Windows).

Here's my opinion: On a price/performance basis, the dual G5 models are competitive with anything but bargain-basement homebuilt machines. They may lag somewhat in some benchmarks, but it doesn't take much for the other factors (development tools, support) to outweigh that as long as it's close enough.

That is, if the tools are available -- if you need Maya Unlimited, at the moment the Mac simply isn't an option anyway.

-- Mark

[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.