PowerMac G5 1.6GHz PSBench 7 Results

[macrumors12345]

Re: Re: Re: Re: Re: One big note ...

Originally posted by Trekkie
[BNow I don't know if doing this in the desktop world is a good idea, because adding 256MB of RAM get's pointless whe you have to buy 4 64MB DIMMs to do it. But in a server it's different. [/B]

Yeah, it would be great if Apple went quad-channel, but I doubt that will happen. After all, it would mean that each time you wanted to upgrade the memory in the G5/1.6, you'd have to throw out all your DIMMs (since there are only four slots)! Okay, so maybe it wouldn't be so great... ;-)

 

[macrumors12345]

Re: not necessarily

Originally posted by AidenShaw
Multiples other than 2:1 are possible for the PPC970 FSB.

If the CPU speeds go up more quickly than memory speeds, they'll be able to keep the bus at 1GHz with faster CPUs - memory speeds won't limit the CPU speed.

Of course they can scale the CPU - I wasn't implying that was an issue (after all, they are currently using a 1 Ghz bus with only "800 Mhz" memory). My point was that the benefits of a faster FSB are limited if you don't have a faster memory bandwidth to complement it.

 

[Ertai]

I don't want to be pedantic about those Photoshop figures, but after all, that's where this thread started. So to sum up my previous post: A G5 1.6 runs the Photoshop test in 3 minutes, a Xeon 3.06/1 MB in 74.1 seconds, a Opteron 244 2x1.8 MHz in 71.2 seconds. That's not just a bit faster. That's roughly 2.5 times faster. Now do you think a 2x2 GHz G5 is 2.5 times faster than a G5 1.6? Dream on...
And by the way: this isn't some freaking PC-benchmark. This is Photoshop. This is Apple homeland. Remember Photoshop? The application MC Steve used to "slaughter" P4s in his ceremonies I mean keynotes.

 

[whooleytoo]

Re: Re: Re: Re: One big note ...

Originally posted by macrumors12345
I'm sure it will. FYI, the G5 FSB will go to 1.2 Ghz in 1Q 2004 (or half of whatever they clockspeed is increased to). =) Question is whether faster memory will be available by then to support it. I don't know.

Well, Hynix claims they'll have volume DDR 500 production by next month - which would be a good match (and in good time) for an early 2004 release of 2.4GHz or 2.5GHz G5s.

They also expect to deliver DDRII 1GHz memory early next year, though I'd imagine that's more tentative; plus I don't know how easily that could be integrate into the G5 architecture.

Mike.

 

[moped]

Re: 1.42 results

Originally posted by synthetickittie
Here are the results from my 1.42ghz dual g4 powermac (fw800), 2gb of ram, GeForce4 Ti 128mbs, running 10.2.6.

90 Degree Rotate-1.0
9 Degree Rotate-3.6
.9 Degree Rotate-7.4
1 Gaussian-5.5
3.7 Gaussian-6.7
85 Gaussian-8.1
Unsharp 50/1/0-5.3
Unsharp 50/3.7/0-10.1
Unsharp 50/10/5-7.9
Despeckle-5.5
RGB-CMYK-7.6
60% Reduction-2.3
LensFlare-9.9
Color Halftone-11.2
NTSC Color-10.8
Accent Edge-18.2
Pointilize-19.6
Watercolor-37.8
Polar Coordinates-8.4
Radial Blur-32.6
Lighting-7.6

-----------------------------------------------
Synthetickittie,

I don't know what happened to your 1.42 dual but my 1.25 dual
gave me the following result.

90 Degree Rotate-0.9
9 Degree Rotate-2.8
.9 Degree Rotate-2.6
1 Gaussian-0.7
3.7 Gaussian-2.3
85 Gaussian-3.3
Unsharp 50/1/0-1.2
Unsharp 50/3.7/0-2.7
Unsharp 50/10/5-3.2
Despeckle-0.8
RGB-CMYK-3.5
60% Reduction-0.6
LensFlare-5.9
Color Halftone-6.2
NTSC Color-7.3
Accent Edge-22.4
Pointilize-19.3
Watercolor-48.1
Polar Coordinates-4.8
Radial Blur-34.5
Lighting-3.3

 

[macrumors12345]

Re: 1.42 results

Originally posted by synthetickittie
Here are the results from my 1.42ghz dual g4 powermac (fw800), 2gb of ram, GeForce4 Ti 128mbs, running 10.2.6.

90 Degree Rotate-1.0
9 Degree Rotate-3.6
.9 Degree Rotate-7.4
1 Gaussian-5.5
3.7 Gaussian-6.7
85 Gaussian-8.1
Unsharp 50/1/0-5.3
Unsharp 50/3.7/0-10.1
Unsharp 50/10/5-7.9
Despeckle-5.5
RGB-CMYK-7.6
60% Reduction-2.3
LensFlare-9.9
Color Halftone-11.2
NTSC Color-10.8
Accent Edge-18.2
Pointilize-19.6
Watercolor-37.8
Polar Coordinates-8.4
Radial Blur-32.6
Lighting-7.6

Thanks for the DP 1.42 scores synthetickittie, that was exactly what we needed! All I can say is, WOW. The SINGLE G5/1.6 absolutely obliterates the Dual 1.42 G4 in virtually every test - in fact, the Dual 1.42 only wins in three tests, despite the fact that we are testing a DP machine against an SP machine!!! Sorry to ask, but are you sure these results are accurate? It's just incredible.

If they are accurate, this also indicates at least one of two things (and probably both). (1) PSBench is a *terrible* Photoshop benchmark (I have a lot of respect for the PPC970, but come on, it is not THAT much faster than the MPC 7455). (2) The Dual G5 will absolutely kill Xeon/P4/Athlon/Opteron on more comprehensive Photoshop benchmarks.

 

[macrumors12345]

Re: Re: 1.42 results

Originally posted by moped
-----------------------------------------------
Synthetickittie,

I don't know what happened to your 1.42 dual but my 1.25 dual
gave me the following result.

90 Degree Rotate-0.9
9 Degree Rotate-2.8
.9 Degree Rotate-2.6
1 Gaussian-0.7
3.7 Gaussian-2.3
85 Gaussian-3.3
Unsharp 50/1/0-1.2
Unsharp 50/3.7/0-2.7
Unsharp 50/10/5-3.2
Despeckle-0.8
RGB-CMYK-3.5
60% Reduction-0.6
LensFlare-5.9
Color Halftone-6.2
NTSC Color-7.3
Accent Edge-22.4
Pointilize-19.3
Watercolor-48.1
Polar Coordinates-4.8
Radial Blur-34.5
Lighting-3.3

Yes, I find those numbers slightly more believable. But it is still outstanding, since the single G5/1.6 still beats the Dual G4/1.25 in almost all the tests.

That said, the fact that these numbers are varying so wildly between similar machines throws into question whether this is even a remotely reliable benchmark.

 

[synthetickittie]

Re: Re: 1.42 results

Originally posted by macrumors12345
Thanks for the DP 1.42 scores synthetickittie, that was exactly what we needed! All I can say is, WOW. The SINGLE G5/1.6 absolutely obliterates the Dual 1.42 G4 in virtually every test - in fact, the Dual 1.42 only wins in three tests, despite the fact that we are testing a DP machine against an SP machine!!! Sorry to ask, but are you sure these results are accurate? It's just incredible.

If they are accurate, this also indicates at least one of two things (and probably both). (1) PSBench is a *terrible* Photoshop benchmark (I have a lot of respect for the PPC970, but come on, it is not THAT much faster than the MPC 7455). (2) The Dual G5 will absolutely kill Xeon/P4/Athlon/Opteron on more comprehensive Photoshop benchmarks.

thats what I was thinking, somethings gotta be wrong with these tests since my computer got killed. I did everything the way they said to, with the setting all put to what they want. I actually did a few of the first test again to make sure it was right and they all came out almost exactly the same.

 

[theRebel]

Originally posted by Ertai
So to sum up my previous post: A G5 1.6 runs the Photoshop test in 3 minutes, a Xeon 3.06/1 MB in 74.1 seconds, a Opteron 244 2x1.8 MHz in 71.2 seconds. That's not just a bit faster. That's roughly 2.5 times faster. Now do you think a 2x2 GHz G5 is 2.5 times faster than a G5 1.6? Dream on...

2 x 2.0Ghz = 4Ghz = 2.5 x 1.6Ghz

I know that a Dual 2Ghz is not actually equal to a 4Ghz system, but the Dual 2Ghz G5 PowerMac's PSBench numbers could very well be over twice as fast as the 1.6Ghz G5 PowerMac. The 2Ghz G5 is obviously faster than the 1.6Ghz G5 and the PowerMac has 2 of them. Plus the Dual 2Ghz system has slightly faster memory than the 1.6Ghz system.

 

[Rocketman]

I went to the chaosmint.com link and I found it interesting and notable that the slowest G5 Mac was generally on par with all those other wintel computers including MP computers and 3.06 Ghz computers in Photoshop tests.

This indicates that the bandwidth and pipelining of the G5/Apple hardware is notably fast and as chip speed, and driver and software optimizations are added, there is actually a good prospect for a change, of Apple actually surpassing Wintel in a visible and indisputable way.

Good job Steve Jobs.

Rocketman

 

[macrumors12345]

Originally posted by Ertai

And by the way: this isn't some freaking PC-benchmark. This is Photoshop. This is Apple homeland. Remember Photoshop?

It's PSBench, not "Photoshop." It is a Photoshop script that runs a very narrow subset of filters within Photoshop. I'm not sure who came up with it, but no self respecting graphic design artist would use it to test the overall performance of Photoshop. Honestly, I wouldn't trust Apple's Photoshop benchmark too much, but even it is a far more comprehensive benchmark than PSBench (they at least run about 35 unique filters/actions - PSBench only runs 15 unique filters/actions), which says a lot.

 

[twinturbo]

Originally posted by hvfsl
In the original article it said that the 1.6Ghz G5 is about the same as a Athlon XP2400. Well the Athlon XP 2400 is clocked at around 1.7Ghz with a 266Mhz bus, so this does not look so good for the G5.

Over on Ars Technica there's a discussion going on right now, and actually the 1.6G5 that's been benchmarked, using a normalized score system, is faster than the 2.6Ghz P4. That DOES look good for the G5.

 

[Mav451]

Originally posted by macrumors12345
Because the Opteron is quite similar to the Athlon but with four key advantages: better floating point performance, 64-bit instructions, better memory bandwidth, and SSE2 support. Of those four advantages, only the last two are even relevant for Photoshop, and they probably will not be enough to offset the Xeon's HT advantage.

Regarding the benchmarks you posted, I will admit that the Opteron performed better than I expected. Unfortunately, since they only give the aggregate score (of what appears to be a PSBench run, from comparing the Athlon MP 2200 times to Ace's Athlon MP 2200 times), we have no idea how the Opteron and the Xeon actually compare to each other - all we are doing is comparing four filters (accent edges, pointillaze, water color, and radial blur). For the record, I doubt that the GamePC website was being dishonest when they gave the aggregate PSBench score - they probably just do not understand that the aggregate score is not meaningful for PSBench (which is understandable since they area gaming site, not a graphic design site).

The strange thing is that the PSBench aggregate score seems to be heavily memory constrained - that is partially why the Opteron is doing much better than the Athlon and the Xeon/3.06 1 MB does much better than the Xeon/3.06. But the G5/1.6 actually has the fastest memory subsystem of all the system's in Ace's test, so why doesn't it *really* clean up? Possibly cache is more important than memory bandwidth in this test (the Xeon 1 MB and the Opteron both have big caches), or possibly some key PSBench filters have not yet been optimized for the G5. It would be very helpful if we could actually see some detailed, consistent G4 PSBench scores, but those have yet to appear.

Incidentally, I did not say I was "so sure" the Xeon would beat the Opteron. I said that I would be quite surprised if the Opteron won in Photoshop. And I will still be surprised if the Opteron wins in Photoshop. What I am SURE of is that the Opteron will not beat the Dual G5 in Photoshop. You can bet money on that. (and in case it was unclear, getting a high aggregate score in PSBench does not constitute "winning", anymore than winning in an Apple run benchmark at WWDC constitutes "winning" - at a very minimum you need to win a clear majority of the subtests).

Well i already know that the G5 and pretty much any kind of Apple hardware (G4) has always been heavily optimized for photoshop. So i know that even without these specific optimizations, that while the Opteron will be competitive, the G5 will (and it better) win out simply b/c it is an Apple computer.

I can still remember back when the G4 was released how they described the G4 slaughtering pentiums. The funny part is that i was already using AMD processors which were slaughtering pentiums in their own right. Pentiums in the socket 370 and 423 never really caught up with AMD's socket 462 until the northwood came out (early 2002).

If the G4/G5 didn't outperform their respective competing intel/amd chips in photoshop than Apple would have ALOT less to present in its keynote presentation

 

[macrumors12345]

Originally posted by twinturbo
Over on Ars Technica there's a discussion going on right now, and actually the 1.6G5 that's been benchmarked, using a normalized score system, is faster than the 2.6Ghz P4. That DOES look good for the G5.

Yes, the normalized score system obviously gives a substantially more accurate measure of the G5's Photoshop performance. And if you look at one of the tables they posted, you will see one reason why - the vast majority of the weight in the PSBench aggregate score is from filters that do not use Altivec (even though 67% of the filters do use Altivec).

 

[macrumors12345]

Originally posted by Mav451
Well i already know that the G5 and pretty much any kind of Apple hardware (G4) has always been heavily optimized for photoshop.

Yes. BUT, it is NOT heavily optimized for PSBench. As I posted above, it turns out that a vast majority of the weight in the PSBench aggregate score is from filters that do NOT use Altivec. So when you compare PSBench aggregate scores, you are comparing on a task which does not use much Altivec or floating point - primarily just scalar integer calculations. In other words, this is actually the WEAKEST scenario for the G5, since its comparitive advantage is in fp and vector ops (not scalar integer ops). Just something you should keep in mind.

If the G4/G5 didn't outperform their respective competing intel/amd chips in photoshop than Apple would have ALOT less to present in its keynote presentation

Indeed! And trust me, it will outperform in Photoshop.

 

[mspock]

www.barefeats.com

I juste read this on Barefeats.com Things are getting better and better !

8/27/03 -- I received a report from a new G5 owner running both Jaguar (10.2.7) and Panther (10.3) versions of OS X. In the tests he ran for me, Panther was up to 41% faster. Details soon.

I usually speak french, so sorry for my bad english

---------------------------
Dell 2,5 ghz 512 rambus, (nobody's perfect)
planning to buy a g5 macintosh next year, tired of XP!!!

 

[Mav451]

Originally posted by macrumors12345
Yes. BUT, it is NOT heavily optimized for PSBench. As I posted above, it turns out that a vast majority of the weight in the PSBench aggregate score is from filters that do NOT use Altivec. So when you compare PSBench aggregate scores, you are comparing on a task which does not use much Altivec or floating point - primarily just scalar integer calculations. In other words, this is actually the WEAKEST scenario for the G5, since its comparitive advantage is in fp and vector ops (not scalar integer ops). Just something you should keep in mind.


Indeed! And trust me, it will outperform in Photoshop.

Then all the more power to you. If this is indeed the "weakest" scenario, then it is a good sign.

Btw, a question to everyone out there: If anyone has an Opteron system, perhaps they could run these same filter tests as welll? I'd be interested in seeing those results.

In the same sense, i'd like to see a more contemporary comparison from the Intel/AMD side as well--use the a7n8x rev2.0 chipset instead of the 9 month old rev1.04. Use ddr400 instead of ddr333.

With intel, i'd like to see how the 2.8 and 3.0C's do on the 875 chipset versus the old granite bay. Of course, also using ddr400 that's actually RUNNING @ ddr400 (in Ace's test, they were indeed using pc3200, but running it at a crippled ddr266, b/c of chipset limitations on the 845)

 

[pilotgi]

Apple posts more benchmarks

According to the Register, a story posted last night shows new SPEC scores performed by VeriTest.

Look here

 

[mspock]

check this out too

http://www.barefeats.com/#quick

 

[Neotronik]

Has anybody done any tests with Macromedia Fireworks?

 

[arn]

thread split

 

[mustang_dvs]

Originally posted by arn
thread split

Thanks Arn.

I really just wish someone would do a full-up test of a 1.6GHz G5 side-by-side with a 1.25GHz single G4 or a Dual 1.42GHz G4 (or both), in a controlled environment. I want to see the gains with a similar setup, the old high-end vs. the new low-end.

I don't know why, but so far these tests feel unreliable. :\

 

[macrumors12345]

Originally posted by Mav451
Then all the more power to you. If this is indeed the "weakest" scenario, then it is a good sign.

The true weakest scenario is focusing only on the filters that do not use Altivec at all (number 13, and numbers 16-20). Averaging the relative scores for these seven filters, the P4/2.8 is about 28% faster than the G5/1.6. In other words, a G5/1.6 is equivalent to a 2.2 Ghz P4 in the "worst case" scenario, at least in Photoshop (don't know if these filters have been at all optimized for the G5...I sort of doubt it since they use CW for Photoshop, and right now CodeWarrior doesn't even have an option to recompile for the PPC 970 core...so this might *really* be the worst case scenario for the G5, but I can't be certain). That is not bad...I was expecting the G5 to be roughly 50% faster than the P4 at a given clock speed in general, so 38% faster in a test that uses none of the G5's strengths is very respectable.

 

[macrumors12345]

Originally posted by mustang_dvs
Thanks Arn.

I don't know why, but so far these tests feel unreliable. :\

What, you mean because the scores are bouncing all over the place? (to date the PSBench tests have "proved" that a G4/1.25 is substantially faster than a G4/1.42 when running Photoshop) I'm sure that Rob over at BareFeats will indulge us as soon as he gets his hands on the Dual G5.

 

[Mav451]

8/27/03 -- The Dual G5 should easily beat the Dual Opteron 242 running Photoshop. In the GamePC article, the Dual 1.6Ghz Opteron ran Photoshop 29% faster than the Dual 2.4GHz Xeon. If you combine the results of our two Photoshop action files, the Dual 2Ghz G5 (without G5 plugin for Photoshop) is 44% faster than the Dual 2.4GHz Xeon. Ergo, the G5 should easily beat the Dual 1.6GHz Opteron. The Dual 2GHz Opteron might be another story. We'll be testing one in a few days.

It is funny that Barefeats based their comparison from an article that is nearly 3 months old (i.e 5/28/2003 vs. the newest review of the 244, on 8/26/2003). If you went to take a look at the more curent review (already posted this a few posts down):

http://www.gamepc.com/labs/view_content.asp?id=opteron244&page=6)

of the opteron (244), then you will see that the dual 1.8 opterons (244) gets 71.2 seconds in the Pshop7.0 filter test.

By Barefeats logic, the dual 1.8Ghz Opteron would be 47% faster than the dual 2.4ghz Xeon (xeon time / opeteron time):

104.8sec/71.2 = 1.4719

Or essentially 47%.

I bring this up b/c the 242 is now considerably "low-end" with the 246's introduction earlier this month.

Considering that the 1.8ghz Opteron is now the "mid" level opteron and that it is "slightly" faster than the fastest Apple model (dual 2ghz G5), we will have to wait and see how the optimizations play out.