[Eug Wanker]

See here:

Dual 2 gig G5
Contributed by: jeremyw | Views: 5275

On Friday I got lucky and got about an hour on a pre-production PowerMac G5 (dual 2 gigahertz) with 2 gigs of RAM and I put it to some tests using Photoshop 7, Final Cut Pro 4 and Soundtrack. I was duly impressed with performance, stability and speed. The machine is huge (think Daystar machine size) but it is very quiet! The model I got access to (and the few others like it, apparently) have to travel around the country with an official Apple rep (like a G5 guard), but I didn't have to sign any NDA nor was I asked to keep my experience confidential.

In regards to ship dates, I have heard that the machines are NOT shipping yet, to anyone, as the FCC hasn't yet approved the computer. Think about this, though: their end of fiscal year is, what, the end of September? I think they need to ship them before this date.
Anyway, more information is included below...

It should be noted that this Dual G5 is a pre-release testing machine, not optimized or tweaked for real-world usage and these test results should be taken very lightly. I noted a sticker on the CPU that said "#1, 6-17", which I assume is a manufacture date for that chip. (Rumor sources have claimed that the IBM plant is not yet making the chips in any quantity, but that is unconfirmed.) Apple is guessing at a mid-September release date for the dual G5 machines and a lot of things could change by then (performance, speed, stability, etc). The last time I tested Photoshop on the then-new G4s and Photoshop was not yet optimized I saw similar results to the ones we have here. Also, with only 2 gigs of RAM and a 2+ gig file (when opened), the machine is hitting the disk all the time. Having a full 4 or 8 gigs of RAM would, I assume, give much better performance.

Some nice touches: the CD/DVD drive door slides down into the machine instead of swinging outward. Cool. The fans spin up and down as necessary and you can see which CPU is working harder based on fan speed. The machine looks great with side panels on or off. I imagine somebody adding neon tube lighting within days of Apple releasing the machines. There are virtually no cables or wires inside the machine. The Serial ATA drive connectors are super tiny and the driver power cables (like on standard IDE drives) are about twice the size. Having a few ports on the front is very convenient.

---

Testing the Dual 2 Gigahertz G5 Tower (2 gigs of RAM, Serial ATA drives, OS X.27) against a single 933-Mhz G4 (1.5 gigs of RAM, ATA133 drives, OS X.26)

In Soundtrack on the dual G5 I layered 64 audio files for about 20 seconds of length and played them all simultaneously without slowing down. Both CPUs were cranking at about 65%-70% usage. I have no idea what the computational limitations of audio files may be in Soundtrack (anyone have any idea?), but my ear can't discern all those sounds playing simultaneously!

In Final Cut Pro 4 the dual G5 recognized that the machine was a "PowerMac G5" and the most amazing thing to me was the fact that of all the video filters and transitions, only a couple filters were NOT real-time (no rendering needed) filters. I think only 1 or 2 transitions were not real-time. Scrubbing through the video files was extremely responsive and fast, even on non-rendered, layered video tracks. Again, in trying to slow down the machine, doing rapid scrubbing continuously on these layered tracks only used about 65% of the CPUs.

---

Photoshop 7, set to use 1 gig of RAM, unoptimized for G5?

First file (A): "couple_ethnic_gift.psd" layered photoshop file of 1.8 gigabytes
Second file (B): "couple_gift_exlrg.psd" layered photoshop file of 764 megabytes

Open file from internal disk:
G5 / A 4:40 minutes
G4 / A 5:01 minutes

G5 / B 1:47 minutes
G4 / B 2:15 minutes

Motion Blur 0 degrees 250 pixels:
G5 / A 0:46 seconds
G4 / A 1:55 minutes

G5 / B 0:32 seconds
G4 / B 1:04 minutes

Free Transform, Criss Cross Image (X-shape):
G5 / A 1:00 minutes
G4 / A 1:15 minutes

G5 / B 0:24 seconds
G4 / B 0:55 seconds

Twirl 999 degrees:
G5 / A 4:58 minutes
G4 / A 5:15 minutes

G5 / B 0:14 seconds
G4 / B 1:02 minutes

It would appear that RAM is critical as the smaller file, which fits in the available RAM, has much better performance. For all you large-file creators: I also took the 768 megabyte layered file and enlarged it to create a 5.6 gigabtye file. It created and saved out no problem.

[Skypat]

Well ... the difference does not seem to be that dramatic ... except for the "motion blur". I expected something like 50 to 100% faster. Am I dreaming ?

[turk.o]

Originally posted by Skypat:
Well ... the difference does not seem to be that dramatic ... except for the "motion blur". I expected something like 50 to 100% faster. Am I dreaming ?

well, i am not sure if you are dreaming or not, but you could be reading a little closer, for the 'a' files the G5 was faster, but not much because photoshop had to constantly go to disk because the file was bigger than the ram, whereas with the 'b' file, the speeds are almost always more than 100% faster than the g4 speeds because it could take the whole file into ram.

his 'a' file was 2GB, which is a HUGE photshop file, i work in PS every day and have never even gone over a one gig file, so.

anyway, all of this is a little irrelevant as he compared it to a single 933 G4, and it was a pre-production machine running on a non-optimized photoshop.

not a slam-dunk, but a good indicator of what might come. i am very anxious to see some real benchmarks on these when they start to ship. has anyone heard official news from adobe abdout a G5 plug-in or anything?

t.o

[osxisfun]

was he running 10.2.7?

also, i think someone mentioned that adobe was soon going to release a g5 plugin.

take away...

he was running on an unoptimized machine...

the g4 was not that much better than a g4 until people started to take adavantage of its altivec etc..

[axlepin]

I know that the G5 isn't quite done and software hasn't been tweaked for it, but I'd have thought that a G5 DP 2Ghz with 2gigs RAM would have CREAMED a single 933 G4 with only 1.5 gigs of RAM!

the test results, with a couple of exception, merely make the G5 seem like a faster G4..and not all that MUCH faster, considering the mhz disparity between the two, and it looks like, given a large enough file, the G4 really closed up the G5's lead.

I have a Dual 2-gig on order, and I'm REALLY hoping that between now and delivery date, there's going to be some serioso optimizing going on.

mucho serioso!

axle

[PixelPete17]

I'm not sure this post is legitimate.

For one, I'm a photoshop pro (truly an expert, I've worked national ad campaigns and smaller jobs too of course), and if this guy/gal really was testing a G5 against a G4, then I would think they would at least have the brains to use a file that fits in RAM in order to see the speed difference of the processor itself, and not the various variables involved a computer's subsystems (memory pipeline, bandwidth, etc).

If he/she had simply tested a 100MB file and run about seven different commands and filters, we'd have a much better idea of what the raw power of a G5 can do. Optimized it should be even better of course. With only a gig of ram alloted to PS 7 though, using a 768 MB file turns almost the whole test into scratch disc contest. Just plain stupid to put it bluntly. (no offense meant, but don't run that kind of test and then post to an internet board and expect to be taken seriously or have your post respected - this kind of post does more harm than good really as it just confuses everyone and ramps up the frenzy of confusion and anticipation)

I'm waiting for the August 20th Mac City presentation myself, and will then take THAT information with a grain of salt until I can get my own hands on a Dual 2 Gig with 2 Gig of RAM. That's the ideal machine for me, but I really trust Apple and their marketing about as far as I can see into the future.

Don't worry guys, for Apple's own sake, these new computers will be faster than these pseudo-benchmarks imply.

Thanks for sharing benchmark-guy, but unfortunately, a real opportunity for the first bit of accurate information was either truly wasted....or you're benchmarks never happened...either way, no food for thought has been given, so let's not start a "food" fight on these forums. (grin)

Take care,

Pixel Pete

[Eug Wanker]

I'm not sure this post is legitimate.

For one, I'm a photoshop pro (truly an expert, I've worked national ad campaigns and smaller jobs too of course), and if this guy/gal really was testing a G5 against a G4, then I would think they would at least have the brains to use a file that fits in RAM in order to see the speed difference of the processor itself, and not the various variables involved a computer's subsystems (memory pipeline, bandwidth, etc).

This guy doesn't sound like a Photoshop pro. So what? It just sounds like he was testing whatever was already installed on the machine. It's not as if he claimed he knew anything about Photoshop or was claimed he was doing an exhaustive review. It's just a very short preview, probably done by someone not quite in the know.

Don't worry guys, for Apple's own sake, these new computers will be faster than these pseudo-benchmarks imply.

I think the tests are probably not fake, and if even if they were, the comparative results would make sense.

1) Many Pshop filters are not dual optimized.
2) On one test the dual G5 is 4.4X as fast as the single 933.
3) This is using the non-G5-optimized Pshop presumably.

I think you'll be in for a disappointment if you expect the dual G5 to hugely faster (besides the boost from G5 optimization).

EDIT:

P.S. Here are some more numbers:

Originally posted by pbkobold:
Well, after watching some ADC TV, I have a few more numbers for you folks. These are using the Noble Ape behavioral simulator at NobleApe.com (get the source if you'd like). The numbers are the number of ape thoughts per second, and it is pretty self explanatory. The initial scalar implementation wasn't threaded and thus didn't use both processors but all the optimized versions did. Vector is self explanatory (although it is only the main loop in the offending function but I believe it takes about 80% of the total program time) and the Vector Optimized is just the vectorizing taken a bit further.

ASCII Time!!

Code:
/------------------+-------------------+----------------\
|                  | G4 - Dual 1.42Ghz | G5 - Dual 2Ghz |
+------------------+-------------------+----------------+
| Scalar           |       1200        |      1330      |
+------------------+-------------------+----------------+
| Threaded         |       2080        |      2490      |
+------------------+-------------------+----------------+
| Vector           |       4510        |     10200      |
+------------------+-------------------+----------------+
| Vector Optimized |       5700        |     16140      |
\------------------+-------------------+----------------/

And for that, I have one thing to say, <3 G5 bandwidth.

I can't say I'm overly impressed with the scalar score (and one can only hope this will be improved with G5 optimization), but unhindered bandwidth definitely does rule for the G5s vector performance.

[Eug]

More benches here using this benchmark app:

Dual G5 2.0: CPU 670
One G5 1.8: CPU 294

Dual G4 1.42: CPU 326
Dual G4 1.25: CPU 288
Dual G4 1.00: CPU 228
Dual G4 867: CPU 199

Eug's GHz TiBook: CPU 114

Note that my CPU speed is 17% the speed of the dual G5 2.0, and the dual 1.25 is slower than the single 1.8. This is of course a synthetic bench, but the G5 numbers are impressive nonetheless.

[HotSoup]

I know this is a bit off topic but does anyone know where I can get the score of a 1Ghz Ali PB compared to a single process 867? What their scores were?

[Eug Wanker]

Originally posted by HotSoup:
I know this is a bit off topic but does anyone know where I can get the score of a 1Ghz Ali PB compared to a single process 867? What their scores were?

The 1 GHz TiBook gets a CPU score of 114.

Note that the test scales pretty linearly with G4 processor speed. You'd get a lower G4 CPU score with a single processor 867. In most usage however, a desktop 867 with a good vid card would feel faster. (Faster hard drive, etc.)

[Superchicken]

Does this imply a Dual 1.42 is faster than a single 1.8?

[Eug Wanker]

Originally posted by Superchic[k]en:
Does this imply a Dual 1.42 is faster than a single 1.8?

Yes it does... for this benchmark. Not that I would be surprised though. Regardless of how bad the FSB speed is, 2.84 GHz of G4 speed is nothing to sneeze at.

[Eriamjh]

C'mon, not all of us use Photoshop. What about iDVD project rendering time? iMovie? iTunes ripping? Let's have some benchmarks for the "ordinary apps" that the non-photoshop users will use.

[Busemann]

Originally posted by Eriamjh:
C'mon, not all of us use Photoshop. What about iDVD project rendering time? iMovie? iTunes ripping? Let's have some benchmarks for the "ordinary apps" that the non-photoshop users will use.

Also games performance would be interesting, as well as VPC 6.1.

the "benchmark" apps are really useless IMO

[Eug Wanker]

Cinebench 2003 and Xbench results of the 1.6 here. The 1.6 does quite poorly on these non-G5-optimized benches.