G4, and varying cache/memory bandwidth, benchmarks?

[diehlr]

Looking very seriously at buying a used Powermac G4. Have a few looming questions.

How do the various L2/L3 cache configurations affect real-world performance in common apps such as movie playback, A/V encoding, and overall system responsiveness?

How much does memory bandwidth play a roll in overall system performance? Are most/all Powermac G4s based on 100/133 mhz SDRAM?

Has anyone done an exhaustive benchmark comparison between various flavors of G4 at similar (as close as possible) clockspeeds with varying amounts of memory bandwidth, or across different G4 motherboards? (Yikes vs Sawtooth vs Digital Audio vs Quicksilver).

 

[ewinemiller]

Check out

http://www.barefeats.com

They do a lot of that kind of benchmarking.

 

[oingoboingo]

Looking very seriously at buying a used Powermac G4. Have a few looming questions.

How do the various L2/L3 cache configurations affect real-world performance in common apps such as movie playback, A/V encoding, and overall system responsiveness?

How much does memory bandwidth play a roll in overall system performance? Are most/all Powermac G4s based on 100/133 mhz SDRAM?

Has anyone done an exhaustive benchmark comparison between various flavors of G4 at similar (as close as possible) clockspeeds with varying amounts of memory bandwidth, or across different G4 motherboards? (Yikes vs Sawtooth vs Digital Audio vs Quicksilver).

First up, check out http://www.lowendmac.com which has a comprehensive listing of the specifications of practically every Macintosh Apple has ever produced. You will be able to find out details like which RAM type, bus speed, cache size etc etc that each of the PowerMacs use.

Another site which you might find useful is http://www.barefeats.com. These guys do various benchmarks on all sorts of Mac hardware, and you might be able to find the comparison you're after there. Also take a look at http://www.macspeedzone.com. And come to think of it, the reviews section of http://www.xlr8yourmac.com contains lots of reports from people who have upgraded their Macs using various G3 and G4 upgrade cards, and they usually post benchmarks of their before and after speed results. You could probably find some complementary benchmark information there which would aid you in your comparisons.

With regards to different memory speed and bus speed, the December 2003 MacWorld has a review of the new 12", 15" and 17" Aluminium PowerBooks. "What the hell has that got to do with PowerMacs?" I hear you say. The review contains a comparison between the 1GHz 12" PowerBook, and the 1GHz 15" PowerBook. Same speed G4, but the 12" PB has a 133MHz front-side bus and DDR266 memory, while the 15" PB has a 167MHz FSB and DDR333 memory...which makes for a good controlled comparison concerning the effect of FSB and DDR RAM speed. The benchmark results from the MacWorld tests are:

SpeedMark 3.2 Overall Score (700MHz G4 eMac = 100, higher is better):
12" PB G4/1GHz: 142
15" PB G4/1GHz: 138

Photoshop 7.0.1 Suite (time elapsed, lower is better):
12" PB G4/1GHz: 0:56s
15" PB G4/1GHz: 0:52s

iMovie 2.1.2 Render (time elapsed, lower is better):
12" PB G4/1GHz: 0:53s
15" PB G4/1GHz: 0:52s

iTunes MP3 encode (time elapsed, lower is better):
12" PB G4/1GHz: 0:45s
15" PB G4/1GHz: 0:43s

Quake III frame rate (higher is better):
12" PB G4/1GHz: 64
15" PB G4/1GHz: 73

Cinema 4D XL 7.303 3D Render (time elapsed, lower is better):
12" PB G4/1GHz: 7:00m
15" PB G4/1GHz: 6:54m

MPEG-2 encode (time elapsed, lower is better):
12" PB G4/1GHz: 14:45m
15" PB G4/1GHz: 13:59m

As you can see, there isn't a whole lot of difference (excepting the 3D benchmarks, but the 15" PB has a faster video chip), except in the MPEG-2 encoding benchmark. Make what you will of these results...obviously the logic board in the 12" and 15" are probably different, they contain different graphics chips, and they may contain different hard drives too. But all in all, they are fairly closely matched, despite their differences in FSB and RAM speed.

If I was looking for a second-hand G4 system, the main thing I'd look for would be an AGP graphics slot. Quartz Extreme can really give GUI performance a nice boost and make the system feel more responsive (especially important for an older Mac). QE isn't supported (officially) on non-AGP Macs.

 

[diehlr]

Check out

http://www.barefeats.com

They do a lot of that kind of benchmarking.

Cool site, although most of their benchmarks are quite old. I would like to see many of those benchmarks updated to reflect the current scenario in modern applications running under Panther. I wonder how much this would change the results?

 

[diehlr]

First up, check out http://www.lowendmac.com If I was looking for a second-hand G4 system, the main thing I'd look for would be an AGP graphics slot. Quartz Extreme can really give GUI performance a nice boost and make the system feel more responsive (especially important for an older Mac). QE isn't supported (officially) on non-AGP Macs.

Thanks for the benchmark results and URLs. I guess bandwidth isn't too crucial for overall performance in most apps. I seem to recall a while back reading that beyond a certain point, bandwidth on the G4 usually doesn't make much of a difference because of the way the G4 was designed (either the chip itself or the underlying motherboard architecture, I forget which). Specifically with respect to using DDR memory.

You're right about AGP. It's definitely one of the main reasons why I am wanting to go with a G4 over a G3. I don't care at all about 3D performance, but I want very snappy 2-D and desktop performance. For that reason I will probably eventually upgrade the video card in whatever lower-end G4 I decide on buying, since most people are selling them with 16 meg Rage 128's.

 

[ewinemiller]

Here's one relatively real world benchmark, or at least it is for the stuff I do. I have a dual 800 Powermac and an 800 ibook G4. Same 133 mhz bus, same 256k of L2 cache, the Powermac has 2meg of L3 cache, the ibook has none.

Cinebench Render portion
ibook 390.2 secs 67 CB-Cpu
Powermac (using only one processor, disabled the other using the developer tools) 374.0 secs 70 CB-Cpu

Hardly noticable difference. As a guess, I would suspect the newer G4s with the 512k of L2 cache would get about a 10% boost in performance depending on the task. When the PIIIs went to 512k of L2 cache, they were about 10% faster per clock than the previous 256k ones.

 

[oingoboingo]

Cool site, although most of their benchmarks are quite old. I would like to see many of those benchmarks updated to reflect the current scenario in modern applications running under Panther. I wonder how much this would change the results?

This page:

http://macspeedzone.com/html/hardware/machine/comparison/all

Contains 'more modern' benchmarks of 14 different Macs (ranging from a B&W G3 PowerMac all the way up to a dual 2GHz G5 PowerMac), running Cinebench 2003, Altivec Fractal, QuickTime encode, a few games benchmarks, some disk copying benchmarks, a PhotoShop action suite test, a Microsoft Office benchmark, and an iTunes music rip benchmarks. The machine lineup in this test includes a dual 1GHz G4 PowerMac (Jan '02 model), a single 1.25GHz G4 PowerMac (June '03 model) which might help you out, plus the usual assortment of iMacs, eMacs, PowerBooks, iBooks and a few G5 systems.

Hopefully you find this useful.

 

[oingoboingo]

Thanks for the benchmark results and URLs. I guess bandwidth isn't too crucial for overall performance in most apps. I seem to recall a while back reading that beyond a certain point, bandwidth on the G4 usually doesn't make much of a difference because of the way the G4 was designed (either the chip itself or the underlying motherboard architecture, I forget which). Specifically with respect to using DDR memory.

You're right about AGP. It's definitely one of the main reasons why I am wanting to go with a G4 over a G3. I don't care at all about 3D performance, but I want very snappy 2-D and desktop performance. For that reason I will probably eventually upgrade the video card in whatever lower-end G4 I decide on buying, since most people are selling them with 16 meg Rage 128's.

There was a bit of controversy when the first DDR PowerMacs were released, because as you correctly point out, the G4 CPU itself cannot properly utilise a DDR memory bus. As I understand it, the Apple system controller *can* use DDR, so things like hard drive DMA transfers can take advantage of the extra memory bandwidth afforded by DDR RAM, but the G4 itself cannot. Being cynical, probably the biggest advantage gained by using DDR RAM is simply the wide availability of DDR versus SDR RAM these days!

http://www.macsales.com (aka: Other World Computing) used to have GeForce 2MX AGP cards relatively cheaply, but they seem to have disappeared. They have AGP 2x/4x Radeon 9000 Pro cards for $169, but this might be a bit more than you want to spend on an old G4 PowerMac. If you take a look around http://www.xlr8yourmac.com, there are plenty of stories about flashing PC GeForce cards to make them compatible with the Mac. You might be able to pick up a PC GeForce card very cheaply and save yourself some cash.

 

[diehlr]

http://www.macsales.com (aka: Other World Computing) used to have GeForce 2MX AGP cards relatively cheaply, but they seem to have disappeared. They have AGP 2x/4x Radeon 9000 Pro cards for $169, but this might be a bit more than you want to spend on an old G4 PowerMac. If you take a look around http://www.xlr8yourmac.com, there are plenty of stories about flashing PC GeForce cards to make them compatible with the Mac. You might be able to pick up a PC GeForce card very cheaply and save yourself some cash.

I'm currently using a GeForce2 GTS 32 MB AGP 4x card in my PC. Once the next batch of DX9 cards come out this spring, I plan on upgrading the card in my PC and the GeForce2 will be moved onto whatever G4 I end up buying. I have no hesitations in performing whatever necessary alterations to get it running on a Mac. I imagine OSX 10.3.x desktop graphics performance with a GF2 will be dramatically faster than a Rage 128.

 

[oingoboingo]

I'm currently using a GeForce2 GTS 32 MB AGP 4x card in my PC. Once the next batch of DX9 cards come out this spring, I plan on upgrading the card in my PC and the GeForce2 will be moved onto whatever G4 I end up buying. I have no hesitations in performing whatever necessary alterations to get it running on a Mac. I imagine OSX 10.3.x desktop graphics performance with a GF2 will be dramatically faster than a Rage 128.

Something I just remembered...a Rage 128 (even the AGP version) will not support Quartz Extreme acceleration. QE requires that the video chip support arbitrary texture sizes, and I think the first ATI chips to allow this are from the Radeon series of cards. So, yes, a GF2 will result in a substantial speedup (above and beyond the fundamental hardware performance advantages that a GF2 has over a Rage128).

Arstechnica has a good overview of how QE works (and why the Rage128 doesn't support it) here:

http://www.arstechnica.com/reviews/02q3/macosx-10.2/macosx-10.2-8.html