Quad G5 with QuadroFX 4500 benchmarks

#1

Ok, busy working on getting some details together. Basically, I ordered the system with 4GB memory. It shipped with 512MB memory (2x 256) but the dealer had given me an additional 4GB of memory (4x 1GB) to put in the system. I did install the memory myself, as well as removed the QuadroFX 4500 and both hard drives, because I can… :smiley:

I am still installing software and running system updates at this time, so I am hoping to spend more time working on benchmarking tomorrow. I installed and tested Maya and noticed that the viewport runs smoother on OS X with the QuadroFX 4500 than my dual Opteron 250 does with the QuadroFX 3400. That is my personal take on it. I haven’t recorded any render benchmarks yet. I did notice thogh that by default, Mental Ray will only use 2 threads for rendering, so the most I saw on ZooRender was benchmark results similar to that of the dual 2.5GHz G5 on the front page of their site. I’ll run a re-render tomorrow, but try and use the flag -n 4 to see if I can get it firing on all cylinders.

I’ll be downloading SPEC Viewperf tonight and readying it tomorrow for running. I ran Cinebench but hadn’t recorded the results as I was doing stuff in the background. I did a test on the regular Cinebench and the G5 optimized ones, so expect those results as well. So far I’m fairly happy with this system. Then again, the best systems I have used so far are my 1.5GHz Powerbook and our 1.8GHz iMac G5.

If there are any suggestions for benchmarks, let me know and I’ll see what I can do.

Right now, on the hot plate are ZooRender, SPEC Viewperf and Cinebench (both regular and G5 optimized).

#2

Try using Specapc for Maya as well. http://www.spec.org/

The reason why this benchmark test is so great is that it actually uses YOUR copy of Maya, and uses scenes that you can open up and take a look at, so there is absolutely no way it can be “tweaked” to get fake results because it’s using .mb files with your copy of Maya. So it’s actual real world results, that are calculated and recorded.

You can even substitute your own scenes with the provided ones if your so inclined to do so, because all it really does is measure the FPS of the view port with a script and nothing else. Then input the data to an excel Sheet for calculation.

So it’s a really great tool for testing hardware with Maya.

#3

I noticed that it says the benchmark is for Windows. I assume it could be run on OS X as well?

#4

Dunno, I never ran it on OS X, it’s essentially a Mel Script and runs from within Maya. So it’s worth a try :slight_smile:

#5

Ok thanks. I was looking at that benchmark, but the Windows thing turned me off. I’ll grab it and see if there are any Windows specific things involved in the file, and hopefully there are no issues when running on Maya 7, since it says Maya 6.5 on the page. I’ll take a look and se if I will have any issues. I still need to download SPEC Viewperf so I’ll grab abc while I’m at it.

#6

waiting on the cinebench results I’ll be extremely interrested to see what difference the quadro card makes

#7

Ok, I have gotten some results thrown together in Cinebench. Needless to say, I’m not impressed in terms of OpenGL rendering. CPU rendering is damn nice though, so it would be nice to see something done about the OpenGL performance on the 4500. Anyways, here are the Cinebench results (Which includes my dual Opteron 250 workstation):

The results are paired in threes. The order is…

  1. Quad 2.5GHz G5 (QuadroFX 4500) Cinebench non G5 Optimized
  2. Quad 2.5GHz G5 (QuadroFX 4500) Cinebench G5 Optimized
  3. Dual 2.4GHz Opteron (QuadroFX 3400) Cinebench Windows

Rendering (Single CPU): 271 CB-CPU
Rendering (Single CPU): 358 CB-CPU
Rendering (Single CPU): 333 CB-CPU

Rendering (Multiple CPU): 824 CB-CPU
Rendering (Multiple CPU): 1005 CB-CPU
Rendering (Multiple CPU): 615 CB-CPU

Multiprocessor Speedup: 3.04
Multiprocessor Speedup: 2.81
Multiprocessor Speedup: 1.85

Shading (CINEMA 4D) : 339 CB-GFX
Shading (CINEMA 4D) : 356 CB-GFX
Shading (CINEMA 4D) : 387 CB-GFX

Shading (OpenGL Software Lighting) : 989 CB-GFX
Shading (OpenGL Software Lighting) : 1136 CB-GFX
Shading (OpenGL Software Lighting) : 1664 CB-GFX

Shading (OpenGL Hardware Lighting) : 2461 CB-GFX
Shading (OpenGL Hardware Lighting) : 2537 CB-GFX
Shading (OpenGL Hardware Lighting) : 3260 CB-GFX

OpenGL Speedup: 7.26
OpenGL Speedup: 7.12
OpenGL Speedup: 8.42

I’ll try and roll out some SPEC benchmarks soon. I’m busy trying to get it to compile under OS X.

#8

Of course you wouldn’t be, because cinebench is just a benchmark for Cinema 4d that dosen’t use the full palette of open GL API calls like a professional OpenGL application like Maya, XSI, 3dsmax, autocad…and the rest of the industry standards.

Again though, Cinema 4d competes with apps like hash and truespace and the hobbyist/enthusiast applications at a price point of 700 bucks so you really can’t expect it to use hardware as well as a 20,000 dollar program.

Some people do think that ALL programs use hardware the exact same way, which is just a naive and inexperienced opinion.

The spec benchmarks would be a more accurate representation of Open GL performance.

But the drivers for the Quadro on OS X have always come under suspicion about the quality. It should be interesting to see the spec benchmarks and see what’s doing on the low level programming side.

#9

From the looks of things, Apple really needs to work on thier OpenGL support. I’m no Mac expert, but doesnt Apple custom write the drivers for OSX? That, and OpenGL isn’t quite up to par, as it is on Windows?

Though the CPU rendering is quite impressive… Wonder how that stacks up to a pair of 280s… and a Quadro 4500.

#10

±---------+
| PLEASE |
| DO NOT |
| FEED THE |
| TROLLS |
±---------+
| |
| |
.|.||/…

Yes, apple does write their own drivers, thats why Im so interrested in what happens here. As the main difference between geforce/quadro and radeon/firegl is that the drivers are written by different teams with a different focus, I was always wondering what possible difference there might be. Im assuming that the same driver teams are writing the geforce and quadro drivers for osx, which leads to the question, what IS the difference between the 2 cards on the mac. They have the same memory and gpu and the drivers are (im assuming) written by the same team, so they don’t have the one big difference which the windows versions have.

The closest comparisson I have to the 280 is a 275 with a 7800gtx attached. This gets 1381 render points and 3254 opengl. Running the 64bit version of c4d would likely push the render score just past 1500 as that version drops support for old cpus.

I have got a result with a quadro 4500 and that gets a feeble 2701 opengl score, but that will be getting hampered by the 1.8gig opteron that was driving it.

#11

Btw, I don’t think anyone has said this yet…so I will.

Thanks Enygma for doing this,

It’s really cool of you. We can get the cold hard data which most professionals can’t publish instead of opinion :slight_smile:

Which is always the best.

#12

No problem. I’m still trying to get the SPEC Viewperf benchmark working. I’m getting a compiler error of some sort. Basically it is saying something about not being able to find malloc.h, which makes no sense since that header is a standard library, so I’ll try and get that issue taken care of and hopefully have some SPEC Viewperf benchmarks ready.

#13

I’m not attempting to feed trolls here. I just wanted a comparison to hardware I know well. That way I can see how this new mac stacks up in price/performance with what I consider its competition :slight_smile:

I’m really interested in seeing the OSX developers improve thier drivers, as thier CPU power is quite respectable… It would only be fitting if they had the realtime performance to match.

If only more of us had the resources (time, money, know how) to pull off such a comparison, we’d probably get a more informed picture of what is generally good and what is generally bad for CG work.

#14

/me raises hand

I’m taking donations… :smiley:

#15

After buying a Quad G5, I bet! :slight_smile:

#16

any news on that osx spec ?

i would love to check how my system performs

thanks for doing this !

#17

Yes, I was going to do some modifications to the source files to make it compile on OS X, but I found that someone else had actually written a batch of files that does that for me. I have other things to do at the moment so that is nice to have. Unforunately, I had downloaded the SPEC Viewperf 8.01 package while the modification scripts are for the 8.1 package. I’ll get to it in a bit. I’ll have to download 8.1. Thanks for reminding me. Just got quite busy at the office.

#18

I would have had one myself if they didn’t have such long order times on custom built ones. The quads are available here now, but if you want the gf7800 or quadro card then they add 2 months to the order :confused: Otherwise they come with the relatively cruddy 6600

#19

I’m getting on my Quad G5 a multiple CPU score above 1100. Did you switch the processor setting from “auto” to “max” in the Energy Saver preference panel?

#20

Could someone possibly explain to someone with less experience (such as myself) what those numbers mean? How does 1005 CB-CPU compare to 615 CB-CPU / 2537 CB-GFX compare to 3260 CB-GFX - what are those numbers actually referring to with regard to time/performance?

Also, do these numbers predict what type of performance might be possible with Maya or could there be a large variation between applications?

Thanks,