CGTalk - Benchmarks: PowerMac G5 and Athlon 64

View Full Version : Benchmarks: PowerMac G5 and Athlon 64

greyface

01-27-2005, 04:24 PM

Hey people, got my cutom built A64 system not long ago, and already own a PowerMac G5, so I decided to run Cinebench 2003 on both and see the results :p.

Contenders

-PowerMac G5 2x2.0Ghz PPC 970FX, 1Ghz HTT(FSB), 1GB 400Mhz DDR-SDRAM

-Athlon 64 3500+ NewCastle @2.5Ghz, 800Mhz HTT, 2GB 400Mhz DDR-SDRAM

I didn't go too much in detail with the specs, as most are irrelevant.
The Athlon 3500+ as you can see is overclocked from 2.2Ghz to 2.5Ghz, and is equivalent to what I would consider as the fastest chips on the market, ie the 4000+ and the FX-55. Sandra 2005 gives me results very close to the 2.6Ghz Opteron...

The G5 is a bit old now, first released in 2003, there are 2.5Ghz versions now, and probably 3Ghz will come soon.

Benchmark

I used Cinebench 2003 for this, as you may know, Maxon doesn't have an official G5 release of it, but a Beta is out, and that's what I used for the PowerMac. They say that it is not yet fully optimized, but it still works quite good. I ran each benchmark 3 times and made the averages, not that they were so different. And note that I only ran the CPU benchmarks as I don't think the others would be of interest, having a Radeon in my mac.
This benchmark renders a 620x620 scene with the Cinema4d render engine.

G5 1x2Ghz: 287 CB-CPU @ 91.8 sec
G5 2x2Ghz: 501 CB-CPU @ 52.5 sec

A64 1x2.5Ghz: 346 CB-CPU @ 76.1 sec

There you've got it, as you can see the G5 running in dual processor mode has an increased performance of 1.75x which I consider quite good. The G5 surprised me quite a bit, being a pretty old system now and still getting very good performance, especially in DP mode. The Athlon gets a good performance too with the overclock, closely matching that of the 4000+ and 3.4Ghz+ P4s without HT, note that HT is highly optimized in Cinebench and gives P4s up to 20% speed increase, don't know if its the case for other renderers like MR.

---

Ok that's it, hope it has informed you a bit on the G5 and how it compares to a good PC chip. I would like people to post their own Cinebench benchmarks so we can compare to this, especially people with dual Opterons and Xeons!

You can get CineBench for free at www.cinebench.com

cheers-

desty

OC-NightHawk

01-27-2005, 04:43 PM

I ran the test twice to get a margin of error and this is what I got.

1st time)
Rendering (Single CPU): 295 CB-CPU
Rendering (Multiple CPU): 618 CB-CPU
Multiprocessor Speedup: 2.10

2nd time)
Rendering (Single CPU): 293 CB-CPU
Rendering (Multiple CPU): 620 CB-CPU
Multiprocessor Speedup: 2.12

To put this in perspective this is from dual 2.4GHz 400FSB Xeons with 512 L2 Cache running at 2.93GHz 533FSB HT enabled and a GB of system memory on a iWill DP533 motherboard.

greyface

01-27-2005, 04:58 PM

Wow, mad. This means a DP is faster than 2 SPs, how can this be possible?

Thanks for the results!

OC-NightHawk

01-27-2005, 05:07 PM

I'll give you three reasons

1) HT
2) HT
3) HT

This is what HT is actually good for. :) It's a good ~35% boost over running the system with HT disabled for the test.

greyface

01-27-2005, 05:27 PM

Oh hmm I knew HT worked good with Cinebench, but not that good :thumbsup: . Are there similar improvements in Spec benchmarks and Mental Ray rendering? It's a pitty my P4 system is an old 2.4Ghz - don't got HT to test :shrug:

OC-NightHawk

01-27-2005, 05:39 PM

I guess it does.:shrug: If you want we can grab a free model off turbosquid and compare render times off rendering the same model. There's a wicked nice m1a1 model available for free download. :) http://www.turbosquid.com/FullPreview/Index.cfm/ID/190990/Action/FullPreview

uncon

01-27-2005, 08:18 PM

I had to render out a scene in Lightwave that crashed over night on my work laptop (Dell P4 1.8Ghz ((don't laugh))) so I loaded it up on a dual 2.4Ghz G5 we mainly use for video editing that wasn't doing anything. Rendering times went from 1.5 min to 10seconds, I setup the rest of my scenes to render and got to go home early!

Those G5 machines are very nice.

OC-NightHawk

01-27-2005, 08:24 PM

Don't take this the wrong way, but your office needs a serious reality check if they expect you to regularly render out production quality images with a old laptop. Its a good thing you where able to find a much better suited machine to get that job done. I don't even want to imagine how frustrating a laptop would be for a render job. :eek:

Edit:

AI-X,

I rendered this with MentalRay 640x480 default setttings except, samples min 16 max 64. Final Gather and gi are enabled with defualt settings. The only change I made to the scene is to put in a skylight. The render time was 1 min flat.

http://studentpages.scad.edu/~ejackm20/SoftQuadroFX4000/SoftQuadroFX4000_05.png

greyface

01-27-2005, 10:06 PM

Hey, I'll get that model, only thing is that I'm using XSI so no skylight there, gonna see if I can get my 3dsmax on my Athlon machine... If not - could you try that same render with HT disabled? Comparing it with the same CPUs would be the most interesting. I'll see how much I get though on XSI.

cheers

alexentremont

01-27-2005, 10:38 PM

OC-NightHawk

01-27-2005, 10:40 PM

yeah sure I can disable ht, but that will have to wait for a bit. The model I showed is here, http://www.turbosquid.com/FullPreview/Index.cfm/ID/205167/Action/FullPreview and available in multipul formats so you can bench it with maya on the g5.

Edit:

The DP is constantly drawing 4 parts of the picture at once as oppossed to a single P4C constantly drawing two parts of the picture at once.

I personally hope somebody with a 800FSB Xeon can share their render times. If my old 533FSB system is spitting distance behind the fastest G5 I can only wager that a Dual 3.6Ghz 800FSB Xeon system would have at least a small lead. Of course based on the single A64 I think a dual Opteron of the same speed would do a very good job as well. :)

imashination

01-28-2005, 12:27 AM

Here are the Cinebench 2003 results for the 2.5 GHz G5:

Rendering (Single CPU): 354 CB-CPU
Rendering (Multiple CPU): 655 CB-CPU
Multiprocessor Speedup: 1.85

It is true that on my PC (a 3 GHz P4C system overclocked to 3.3 GHz), enabling HT makes rendering about 20% faster, but I don't get it that it allows a DP system to be more than twice as fast as the same system using only one processor: both processors have HT, so how could adding a second identical processor result in a system 2.1 times faster??? I see it's real, but I don't understand the explanation.

Because the CB test may be slightly misleading. The single cpu test isn't a single CPU test, it's actually a single thread test. When c4d does the single threaded test, it of course cannot use the hyper threading feature.

So, you have a single threaded speed for a 3 gig p4 at 300CB
Two processors would give around 525 perhaps.
Add on the hyperthreading bonus of 20% and you get 660

alexentremont

01-28-2005, 12:38 PM

Because the CB test may be slightly misleading. The single cpu test isn't a single CPU test, it's actually a single thread test. When c4d does the single threaded test, it of course cannot use the hyper threading feature.

So, you have a single threaded speed for a 3 gig p4 at 300CB
Two processors would give around 525 perhaps.
Add on the hyperthreading bonus of 20% and you get 660

That makes sense. Thank you for the explanation!

jojojo

01-30-2005, 11:37 PM

I got:

356 in SP-mode and
642 in DP-mode

on a dual 2.5 G5 1.25GHz Bus with 2 gig of ram.

OC-NightHawk

02-01-2005, 06:01 AM

I got:

356 in SP-mode and
642 in DP-mode

on a dual 2.5 G5 with 1.5 gig of ram.

That's all you got in DP mode? Oh man ..... sorry but that's not a very far lead over a machine almost 2 years old. Ouch.

CGTalk Moderation

02-01-2006, 07:00 AM

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.