PDA

View Full Version : New Nehalem quad 2.66 compared to g5 quad 2.5




jschone
Apr 6, 2009, 04:13 PM
Last week I received the Nehalem 2.66 Quad. Memory has to arrive so it still has the standard 3 gb. This system is going to replace my g5 quad 2.5 (4.5 gb) which is going somewhere else in the studio. As a fine art print studio, the systems are mostly used for Photoshop. file size 600mb - 2 gb in general.
The first thing I did was to transfer the lacie esata pci-express over to the MacPro.(with almost empty striped Raid, lacie biggest s2s, as primary scratch and 2 wd digital studio editions as scratch 2 and 3, all e-sata)

Not having installed every plugin yet, I used the standard Photoshop CS4 reduce noise filter on the same image (1.2 gb (both running locally) )on the g5 and on the MacPro . To my surprise it ran equal on both machines and maybe even a little faster on the Quad g5.
Then I did the Photoshop speed test from the beginning of the MacPro thread, and the results where:
-Nehamelem quad 25 seconds (with striped raid for scratch)
-G5 quad 36 seconds (no scratch)

I know 3gb is low in the Nehalem but I expected better performance against the g5 quad(with 4.5 gb)

Is this what i should expect?!

Jochem



cmaier
Apr 6, 2009, 04:17 PM
Last week I received the Nehalem 2.66 Quad. Memory has to arrive so it still has the standard 3 gb. This system is going to replace my g5 quad 2.5 (4.5 gb) which is going somewhere else in the studio. As a fine art print studio, the systems are mostly used for Photoshop. file size 600mb - 2 gb in general.
The first thing I did was to transfer the lacie esata pci-express over to the MacPro.(with almost empty striped Raid, lacie biggest s2s, as primary scratch and 2 wd digital studio editions as scratch 2 and 3, all e-sata)

Not having installed every plugin yet, I used the standard Photoshop CS4 reduce noise filter on the same image (1.2 gb (both running locally) )on the g5 and on the MacPro . To my surprise it ran equal on both machines and maybe even a little faster on the Quad g5.
Then I did the Photoshop speed test from the beginning of the MacPro thread, and the results where:
-Nehamelem quad 25 seconds (with striped raid for scratch)
-G5 quad 36 seconds (no scratch)

I know 3gb is low in the Nehalem but I expected better performance against the g5 quad(with 4.5 gb)

Is this what i should expect?!

Jochem

On an image that size, with only a few gB physical RAM, I don't find those numbers too surprising - probably a lot of paging to disk.

Genghis Khan
Apr 6, 2009, 06:15 PM
hmmm...still, i would've expected better...despite the same amount of cores and near same processor speed, the nehalem is meant to be 4 years newer.........

grue
Apr 6, 2009, 06:36 PM
hmmm...still, i would've expected better...despite the same amount of cores and near same processor speed, the nehalem is meant to be 4 years newer.........


You expected wrong. The machine is being crippled by the meager amounts of RAM. When it has to start using disk space as RAM, you might as well have a can of tuna instead of a processor.

jschone
Apr 7, 2009, 06:27 AM
ok, thanks guys. RAM will arrive on friday, see what happens.

UltraNEO*
Apr 7, 2009, 07:05 AM
3Gb on a Core i7 rig... I have more than that on my games PC i've build purely for gaming. OP, I'd suggest a minimum of 6GB and 12Gb for multi-media work, gives it a bit more flexibility.

jschone
Apr 7, 2009, 03:44 PM
Thank you, UltraNEO, the Ram is on order, like I said in my first post.....

Sorry, I don't know anything about gaming. Last game I played was Hunchback on a Sony MSX...

Tesselator
Apr 7, 2009, 04:21 PM
Thank you, UltraNEO, the Ram is on order, like I said in my first post.....

Sorry, I don't know anything about gaming. Last game I played was Hunchback on a Sony MSX...

Hehehe, I think I remember that game... On the C=64 tho. :D

Anyway, I guess you won't see any difference at all with more memory installed. I went from 4 gigs to 12 gigs and my times actually got worse. They went from 28.53 Seconds to 32.77 seconds. (EDIT: Just now restarting my Mac and running the test again reveals a time of 30.47 Seconds. Closing and reloading the image and then re-running the test a second time produced a result of 27.67 seconds. So maybe there's no difference at all and it's not slower after all.)

http://forums.macrumors.com/showpost.php?p=6537868&postcount=729

But I think 25 and 36 is about right for those two machines anyway. You went from 4 physical cores at 2.5 to four physical cores at 2.66 with slightly faster RAM and a larger CPU cache? Looks right to me!

Concerning Photoshop, clocks count for the most! Cores seldom count at all. And it seems to successfully ignore any architectural improvements that have been engineered into the various CPU models and revisions - besides the caches maybe. :)

.

jschone
Apr 8, 2009, 04:25 PM
Hi Tessel, thank you! the name of the game was Hunckback from Notredame, the main character had to jump over stones and ring bells if I remember well...It might even have been on the Vic 20...Those huge sd/flash card kind of things were nice though, much better then those audio-cassettes that came afterwards..

Anyway, very much related to this, a G5 quad still performs well if you put enough memory in it even compared to the new Nehalem. Of course I am not talking 3d or movie rendering here, but regular Photoshop, Illustrator and Indesign tasks. 25 secs against 36-37 is not much of a difference...

The biggest change probably would be a solid state disk for apps...together with a 4 Quadra Lacie connected e-sata II as scratch...

Tesselator
Apr 8, 2009, 04:40 PM
Hi Tessel, thank you! the name of the game was Hunckback from Notredame, the main character had to jump over stones and ring bells if I remember well...It might even have been on the Vic 20...Those huge sd/flash card kind of things were nice though, much better then those audio-cassettes that came afterwards..

Hehhe sounds about right.
http://image.com.com/gamespot/images/bigboxshots/1/573751_377_front.jpg http://upload.wikimedia.org/wikipedia/en/4/48/Hunchback_screenshot.JPG
:D

Anyway, very much related to this, a G5 quad still performs well if you put enough memory in it even compared to the new Nehalem. Of course I am not talking 3d or movie rendering here, but regular Photoshop, Illustrator and Indesign tasks. 25 secs against 36-37 is not much of a difference...

The biggest change probably would be a solid state disk for apps...together with a 4 Quadra Lacie connected e-sata II as scratch...


LOL @ the bold text! :D

And, yep! Clocks still count for most of our computing experiences. Cores have been the focus and that's good for folks like me who always have a machine rendering off by itself (or a stack of machines as it was in the past) but what you say is very true. Ideally I would love to see a 4-core chip clocked at 5.2GHz Mmmmm... Yummy! All the better if it was SMP capable for multi-processor system building!

darrenscerri
Apr 8, 2009, 04:43 PM
Open Activity Monitor and post the number of Page Outs and Swap after you run the filters.

CubeHacker
Apr 8, 2009, 05:43 PM
And, yep! Clocks still count for most of our computing experiences. Cores have been the focus and that's good for folks like me who always have a machine rendering off by itself (or a stack of machines as it was in the past) but what you say is very true. Ideally I would love to see a 4-core chip clocked at 5.2GHz Mmmmm... Yummy! All the better if it was SMP capable for multi-processor system building!

You're just helping to propagate the false fact that more mhz = faster speed, something that Apple fought against for years with their "Megahertz myth", and something that even Intel has given up on now. The Nehalem might be running at a slower clock speed than the G5, but its easily twice as efficient at running code. Thats not to say that a Quad G5 is a slouch. But a quad core Nehalem should certainly be faster.

Tesselator
Apr 8, 2009, 06:49 PM
You're just helping to propagate the false fact that more mhz = faster speed, something that Apple fought against for years with their "Megahertz myth", and something that even Intel has given up on now. The Nehalem might be running at a slower clock speed than the G5, but its easily twice as efficient at running code. Thats not to say that a Quad G5 is a slouch. But a quad core Nehalem should certainly be faster.

Should be. The OP here is proof that clocks do actually matter. It's just not the ONLY thing is all.

Yes, CPU architecture matters a lot. Apple was not correct about their "Megahertz myth" assumptions however and finally came around to Intel. But now we're at a die saturation point so R&D offers the solution of multiple dies (cores) until we can move to a better overall technology than electrons running around on gold plated silicon glass dancing to the beat of a pulse clock.

cmaier
Apr 8, 2009, 06:58 PM
Should be. The OP here is proof that clocks do actually matter. It's just not the ONLY thing is all.

Yes, CPU architecture matters a lot. Apple was not correct about their "Megahertz myth" assumptions however and finally came around to Intel. But now we're at a die saturation point so R&D offers the solution of multiple dies (cores) until we can move to a better overall technology than electrons running around on gold plated silicon glass dancing to the beat of a pulse clock.

"gold plated?"

sigh.

Tesselator
Apr 8, 2009, 07:03 PM
"gold plated?"

sigh.

What? Gold is no longer evaporated onto the wafers as a production step? When I was working at National SemiConductor in Utah back in the early 80's that was the process.

cmaier
Apr 8, 2009, 07:17 PM
Gold is a deep donor. The last thing you want in a semiconductor device is deep donors.

And interconnect was aluminum back then (and now copper), with some tungsten plugs here and there.

Why on earth would they introduce gold, rendering the devices completely unpredictable?

CubeHacker
Apr 8, 2009, 07:19 PM
Should be. The OP here is proof that clocks do actually matter. It's just not the ONLY thing is all.

Yes, CPU architecture matters a lot. Apple was not correct about their "Megahertz myth" assumptions however and finally came around to Intel. But now we're at a die saturation point so R&D offers the solution of multiple dies (cores) until we can move to a better overall technology than electrons running around on gold plated silicon glass dancing to the beat of a pulse clock.

I guess it depends on the test being done. But barefeats compared some of the older Mac Pro's to the quad G5 here (http://www.barefeats.com/harper.html). The old 3.0ghz 4 core woodcrest processors should be about the same speed as the new nehalem processors, and you can easily see that its about 2x the speed of the G5's. Certainly not scientific, as some Photoshop actions might be more MP aware than others. But generally speaking, it should be about 2x faster.

Tesselator
Apr 8, 2009, 07:39 PM
I guess it depends on the test being done. But barefeats compared some of the older Mac Pro's to the quad G5 here (http://www.barefeats.com/harper.html). The old 3.0ghz 4 core woodcrest processors should be about the same speed as the new nehalem processors, and you can easily see that its about 2x the speed of the G5's. Certainly not scientific, as some Photoshop actions might be more MP aware than others. But generally speaking, it should be about 2x faster.

Yeah REALY REALY not scientific as most of those benchmarks came from individuals running in a whole host of different environments and conditions.

I'll not argue that the new procs are faster on a clock for clock basis. They are. Those tests even combined do not show an accurate depiction of how much tho and certainly not how much this difference affects mission critical (mission specific) computing.

jschone the OP, has the right idea. Set both machines up and optimize them for a particular workflow and then benchmark within that workflow.

In his case the differences between the Nehalem 2.66 Quad w/3GB RAM and the G5 Quad 2.5 w/4.5GB RAM were not astounding. And I would be willing to bet given the same monitor, mouse and keyboard, that a person would have to test them side by side to know which one (s)he was using (within jschone's stated workflow). There's probably that small of a difference outside the realm of benchmarks. I could be wrong of course. I don't have a G5. My Apple specific experience goes from the Apple ][ through to about the PowerMac 8100 and then drops dead till the 1st Mac Pro. So I missed pretty much all of OS9 through to OSX 10.4


.