[Eug Wanker]

See here:

[Severed Hand of Skywalker]

Benchmarks. This is what people do when they have no real work to do but are sure they need the fastest computer on the plant to brag to other geeks on the internet.

[MindFad]

I ran it on my dual 800. The G5 kicked its ass.

[Eug Wanker]

Benchmarks. This is what people do when they have no real work to do but are sure they need the fastest computer on the plant to brag to other geeks on the internet.

And I wonder what it means if you have time to post in threads on topics you supposedly have no interest in...

I ran it on my dual 800. The G5 kicked its ass.

I believe Xbench is normalized against a dual 800 or something similar. (Normal = 100 for each test.)

[Severed Hand of Skywalker]

Originally posted by Eug Wanker:
And I wonder what it means if you have time to post in threads on topics you supposedly have no interest in...

I wanted to see what the next argument would be all about

I NEED 2x 2GHz G5's for surfing the net and encoding an MP3!

[Eug Wanker]

Interesting that the Altivec performance is crap and the FP performance is monstrously high.

Most people are guessing that FP performance WILL be very high, but not that high. OTOH, most people are guessing that Altivec performance will not be stellar on a MHz for MHz basis, but it definitely won't be as low as this bench suggests. People are guessing that on a MHz for MHz basis, Altivec on a G5 will be similar to Altivec on a G4. If this bench were reflective of that belief, the Altivec score would be 200+, not 52.

It just goes to show you the limitations of synthetic benches when comparing CPU different designs (and their potential unusability to predice performance for specific software applications).

[Severed Hand of Skywalker]

Originally posted by Eug Wanker:
It just goes to show you the limitations of synthetic benches.

Bingo, real world is all that matters.

[Eug Wanker]

Originally posted by Severed Hand of Skywalker:
Bingo, real world is all that matters.

Captain Obvious strikes again...

[Severed Hand of Skywalker]

Originally posted by Eug Wanker:
Captain Obvious strikes again...

No that is another user.

Understood Admiral pointless topic?

[Eug Wanker]

Yeah, no offence to the real Captain Obvious...

[Severed Hand of Skywalker]

Originally posted by Eug Wanker:
Yeah, no offence to the real Captain Obvious...

Ya, he's a war hero!

Benchmarks onward!

[DrBoar]

Just to show what bench marks can do
Comparing a dual 1.25 GHz G4 with the G5 scores above
CPU GCD Recursion Mops/s 5.30--6.14
Floating point Gflops/s 1.32--0.50
Altivec Basic Gflops/s 2.84--8.57

Memory test System 183--149
Stream 296--147

These reusuls would suggest that a 1.25 GHz G5 beats the 2 GHz IBM 970 in several tests and that Motorlas 167 MHz bus is not far behind IBMs 1000 MHz bus

An other interpretation is that the Xbench ony works to compare very closely rellated CPUs like various version of G4 but that the IBM 970 throws it off the track completly.

Soon enough will will see if the 970 will sink or swim. I am sure that when IBM release their linux servers that they will be extensively tested

[Big Mac]

AFAIK, the author of xBench says the low scores are due to a known issue with the app itself. It's really wonderful to get all worked up over nothing.

[tikki]

Originally posted by Severed Hand of Skywalker:
I NEED 2x 2GHz G5's for surfing the net and encoding an MP3!

What's wrong with wanting to encode mp3s that fast! I plan on it

[Eug Wanker]

Originally posted by tikki:
What's wrong with wanting to encode mp3s that fast! I plan on it

Yeah, that's just icing on the cake, but it will be depend on the CD-ripping speed of the optical drive. But it would sure be nice to be able to to a full encode of a CD at 40X - in 1.5 minutes.

But more importantly, any video stuff will absolutely fly (regardless of what Xbench says the Altivec speeds are).

[Superchicken]

imagin in a few years when we see the direction of pixlet, the iFlicks movie store, using mpeg 5 with Apple's special ultra compression utility that requires a 2Ghz PPC 970 to decode

[BurpetheadX]

Originally posted by Big Mac:
AFAIK, the author of xBench says the low scores are due to a known issue with the app itself. It's really wonderful to get all worked up over nothing.

Welp, that makes this forum officially worthless.

[Vash]

Wow even if that bench mark is messed up it's still about double the total score of my dual 1ghz quicksilver That is one fast machine.

[nvaughan3]

yep looks like a nice workstation.

[Immortal K-Mart Employee]

Originally posted by BurpetheadX:
Welp, that makes this forum officially worthless.

Exactly, why even bother with the stupid app if it gives wonkey results.

[MindFad]

Some Cinebench results would be nice.

[rambo47]

It's gonna be so sweet doing digital video on a Dual-G5! Oh, baby - we've got WOOD !

[Spliffdaddy]

Originally posted by MindFad:
Some Cinebench results would be nice.

ditto

been waiting for any machine that will beat last year's dual Athlons.

[ReggieX]

Originally posted by Severed Hand of Skywalker:
I NEED 2x 2GHz G5's for surfing the net and encoding an MP3!

I'd use it to re-encode my DIVX movies to DVD so I can lend them out.

[MindFad]

Originally posted by Spliffdaddy:
ditto

been waiting for any machine that will beat last year's dual Athlons.

What are the scores?

[Spliffdaddy]

Under Cinebench2000 the dual 1GHz PowerMacs scored 22 on (multiple CPU) raytracing while the fastest dual Athlons returned scores in the low 40s.

I'm certain there have been many faster PC platforms released since then, but I thought it was curious just how utterly dominating the Athlons used to be at that task.

I'd like to see that sort of domination happen again. It promotes competition. Doesn't matter who dominates...may as well be IBM this time.

[theory]

Any one up for some photoshop tests

[chrisutley]

I have it on good authority that the G5's will be fast, faster than G4's and about as fast as the equivalents in the PC world. Don't tell anybody or Steve will fire me!

[MindFad]

Originally posted by chrisutley:
I have it on good authority that the G5's will be fast, faster than G4's and about as fast as the equivalents in the PC world. Don't tell anybody or Steve will fire me!

Well, I hope so, if Steve and company were really demoing and dual 3GHz Xeon with a Dual 2GHz G5 on that conference stage. Or was that mostly crap, too? Everybody's bitching about the numbers, but actually seeing it in action was pretty impressive. In action according to Apple, at least.

[Eug Wanker]

Some Cinebench results would be nice.

Here ya go (for the 1.6). Another Xbench in there too.

Too bad the G5 sucks so bad in these tests.

AFAIK, the author of xBench says the low scores are due to a known issue with the app itself.

Well I hope so. That, and the lack of optimization.

[Eug Wanker]

So basically, current Cinebench 2003 benchmarks on the G5 are pretty much useless. The developers say they have yet to optimize for it. (They didn't get an advance G5 prototype G5 to test with.) See the messages below:

---

I really wonder about what versions of C4D and Cinebench we are talking about?

Here some first hand clarifications about the optimizations done in Cinebench 2003 and C4D V8...

For the Mac OS the apps are compiled using CodeWarrior 8.3 with all speed optimizations switched on and scheduling is set for G4 7450...

For Windows the apps are compiled with the latest Intel-Compilers and optimization is done for the P4...

The source code (C++) for both platforms is exactly the same, except for a very small OS compatibility layer and some Apple specific GL optimizations (AppleVertexArryas and AppleFence), those have been implemented on request from Apple...

AltiVec, MMX, ISSE or ISSE2 is not used at all. There are two main reasons for that:

1) The time consuming calculations (intersections and color) have to be done in 64 bit, there is no way to do those in production quality with 32 bits only...

2) All those Vector Units are optimized for linear data access, always one value directly after the other, guess why they are called Vector Units. Raytracing is by nature exactly the opposite of this. Data is scattered all over the memory, the ray is simply not predictable, if it would be, the entire raytracing problem would be solved...

Simple test: Take an iBook G3 900 MHz and a Powerbook G4 867 Mhz and run on both machines some renderings using C4D, LW, Maya or EI. Why is the G4 not really faster? Because no one is able to use the current VE for production quality rendering...

SIMD (Single Instrucion Multiple Data) is great for streaming data (Music, Posteffects, Movies), this is what is was made for, but using it as a general FPU is not working...

Since Version 8 the apps are optimized for Hypherthreading, especially the GI (just try V7 against V8 on a HT enabled machine)...

Maxon is directly working together with Apple, Intel and AMD to get the best render speed out of the hardware. Depending on how early we get our hands on new hardware, we do what we can to optimize for it. Most times we have been the first app utilizing new hardware beaus of that...

Regarding the G5, it does not make much sense at all to discuss the current results, right now there are a few prototype machines with new GFX cards (probably beta drivers) and a patched OS running, what can one expect from a configuration like that...


Cheers, Richard (one of the Maxon developers)

---

Nathan,

an example: There is an algorithm that needs to access memory and of course does some calculations. Now if you always use 64 bit numbers you have to move twice as much memory compared to 32 bit numbers. Now there are some calculations to do, but for those 32 bit does not give enough precision...

Can you see the picture?

As you might understand I can not show you the details of our render engine

Since more than 8 years we are working, tweaking and optimizing on those routines, they have been developed and tested on the 68000, 68020, 68030, 68040 and even the 68060, 80386, 80486, P1, P2, P3, P4, AMD Athlon, PPC 601, PPC 604, G3, G4 and as soon as I get my hands on the G5, they will get optimized for this one as well...

Cheers, Richard

[clarkgoble]

I believe that for altivec code to perform optimally on the G5 a couple altivec library function calls have to be eliminated. This was discussed over on Ars a couple of days ago.

[Cadaver]

Originally posted by Severed Hand of Skywalker:

I NEED 2x 2GHz G5's for surfing the net and encoding an MP3!

Amen, brother!

He who dies with the fastest computer, wins!

[Eug]

New posting from the Cinebench developers:

With the current CineBench a single G5 1.8GHz scores at about 188, the optimized version will maybe score at about 238...

A hypothetical single G5 2.0GHz could score at about 210 on the old CB, optimized could be 265...

A dual G5 2.0 could maybe score at about 480 with the optimized version of CB....

Depending on the new compilers and our findings (thanks a lot to Apple for being extremely helpful and cooperative) we might even crack the 500 score for the dual G5 2GHz...

For reference, a single Xeon/P4 3 GHz scores a little over 300. With HT it's about 350. A dual Xeon (no HT) scores 568. A dual Xeon (with HT) scores 650.

Thus, after optimization, if the 500 number is achieved, then in Cinebench a G5 2.0 may score in the range of an extrapolated P4 2.66 without HT. Interestingly, this is the same as the performance of Dr. Hunter's Jet3D on the G5 2.0, with the unoptimized scalar code. (He expects a significant speedup with optimization.)

[Eug Wanker]

Barefeats has some unoptimized G5 numbers up:



I wonder why they don't just post the scores.

Predictions are that the G5 will easily surpass the Dual Xeon once the G5 optimizations are in place.

Uhhh... no. Word is that the dual G5 will be competitive with the dual Xeons, but the fastest Xeons will still be faster.

[ckohler]

Does anybody have a compairison between a single processor G5 verses a single processor G4? Or better yet, a dual processor G5 (2ghz) verses a single processor G4? I'd like to get an idea of just what kind of performance boost to expect if I upgrade.

[Eug Wanker]

Originally posted by ckohler:
Does anybody have a compairison between a single processor G5 verses a single processor G4? Or better yet, a dual processor G5 (2ghz) verses a single processor G4? I'd like to get an idea of just what kind of performance boost to expect if I upgrade.

For what apps?

[ckohler]

Quicktime. Mostly video encoding. For example, my 800mhz G4 iMac takes about 3.5-4 hours to encode 30min of DV footage into DiVX using the DiVX Quicktime codec. It would be great to know how long it takes a new G5.

Also, we all know it takes about 2x the length of a DV clip for a G4 and iDVD to transcode to MPEG2. What about the new G5s? That kind of stuff.

[kupan787]

Originally posted by Eug:
New posting from the Cinebench developers:
For reference, a single Xeon/P4 3 GHz scores a little over 300. With HT it's about 350. A dual Xeon (no HT) scores 568. A dual Xeon (with HT) scores 650.

Thus, after optimization, if the 500 number is achieved, then in Cinebench a G5 2.0 may score in the range of an extrapolated P4 2.66 without HT. Interestingly, this is the same as the performance of Dr. Hunter's Jet3D on the G5 2.0, with the unoptimized scalar code. (He expects a significant speedup with optimization.)

I am confused. If a Dual 2.0GHz scores close to 500, and a P4 3GHz scores 300 (actually I have only seen 281 on a few websites, but whatever), how would a P4 2.66 be faster than the G5 (when the G5 already outscores the P4 3.0 by nearly 200 points)?

[Spliffdaddy]

He was referring to a single 2Ghz G5 against a single 2.66 Xeon.


I just checked my non-HT P4 @ 2888MHz, and it scored 283.

[Eug Wanker]

Originally posted by kupan787:
I am confused. If a Dual 2.0GHz scores close to 500, and a P4 3GHz scores 300 (actually I have only seen 281 on a few websites, but whatever), how would a P4 2.66 be faster than the G5 (when the G5 already outscores the P4 3.0 by nearly 200 points)?

Dual 2.0 is estimated to score 480+.

A single P4 3 GHz scores around 300. (I've seen up to about 305, and over 350 with HT active.)

A dual Xeon 3 GHz scores as high as 565 (or well over 600 with HT).

A dual Xeon 2.66 would get about 500. (Not a single P4 2.66.)

I'm was just saying that with the estimated optimized Cinebenches and the unoptimized Jet3D benches, a single G5 2.0 performs similarly to a single P4/Xeon 2.66 without HT.

ie. With an app not limited by bandwidth (where the G5 would shine) and where the compiler performance isn't hugely different (where the P4 may shine) the single G5 2.0 would perform somewhere in the range between a 2.4 and 3 GHz P4.

The caveat is that the Jet3D bench might be much faster than this after optimization, according the author of the bench.

[Spliffdaddy]

Alright. I couldn't resist.

Cranked the 2.53 Northwood P4 up to 3.0GHz (19 x 158fsb = 3002) in an effort to see the big 300 in Cinebench. Keep in mind that the RAM in this machine is being underclocked from its norm of 333MHz DDR to just over 300MHz. So there's a performance penalty for that. A Xeon P4 would perform better, in other words. hell, any P4 of the same clockspeed with its RAM operating at its rated frequency would do better.

Just my luck...

89.6 seconds with a score of 294

be right back. she's gonna need more MHz to clear 300.

edit:

yay.

306 86.1 seconds

3077MHz

[kupan787]

Originally posted by ckohler:
Quicktime. Mostly video encoding. For example, my 800mhz G4 iMac takes about 3.5-4 hours to encode 30min of DV footage into DiVX using the DiVX Quicktime codec. It would be great to know how long it takes a new G5.

Also, we all know it takes about 2x the length of a DV clip for a G4 and iDVD to transcode to MPEG2. What about the new G5s? That kind of stuff.

This is what I want to know. How fast will a Single 1.8 GHz (what my brother got, so why I ask) encode mpeg 2 video (running bitvice for example). How about rendering in AE or FinalCut Pro? Ripping a DVD to DIVX?

[Eug Wanker]

Spliffdaddy, I think the 305 I saw was with a 3.06 GHz P4. Your numbers are pretty much bang on.

What I'm most interested in is iMovie, iDVD, and Final Cut performance. I've only been dabbling in this stuff (and have only tinkered with FCE), but will be getting a MiniDV camcorder so I want a computer to support this. My TiBook 1 GHz is not Snappy™ enough for this purpose.

Also, one thing that irritates me is the simple thumbnail autogeneration feature in the Finder, of directories full of images in the icon view mode. I'm starting to think it would take a 2 GHz G5 just to make this feature tolerably Snappy™.

A rev. B single 2.2 GHz G5 Power Mac might just do the trick for all of this...

[Spliffdaddy]

Oddly enough, I suspect that my dual Athlons from last year would still wreak havoc on this 'modern' P4 - and probably cost less while doing it. I can't afford dual Xeons to play with, but I really hate to lose. Life sucks sometimes.

I wonder what MP Athlons at over 2GHz would score on the newer version of Cinebench. I remember seeing (devastating!) numbers in the low 40's under the 'old' Cinebench...I'd guess that corresponds to about 400+ in the 2003 version of Cinebench.

[Eug Wanker]

Originally posted by Spliffdaddy:
I wonder what MP Athlons at over 2GHz would score on the newer version of Cinebench. I remember seeing (devastating!) numbers in the low 40's under the 'old' Cinebench...I'd guess that corresponds to about 400+ in the 2003 version of Cinebench.

CINEBENCH 2003 v1
*************************************************
Tester : Ken
Processor : Athlon 2600MP
MHz : 2.13
Number of CPUs : 2
Operating System : Win2k SP1
Graphics Card : GF3 Ti200
Resolution : 1200 x 960
Color Depth : 32-bit
*************************************************
Rendering (Single CPU): 263 CB-CPU
Rendering (Multiple CPU): 487 CB-CPU
Multiprocessor Speedup: 1.86
Shading (CINEMA 4D) : 257 CB-GFX
Shading (OpenGL Software Lighting) : 1048 CB-GFX
Shading (OpenGL Hardware Lighting) : 1162 CB-GFX
OpenGL Speedup: 4.52
*************************************************

CINEBENCH 2003 v1
*************************************************
Tester : sw
Processor : Xeon 3.06
MHz : 3.06 GHz
Number of CPUs : 4
Operating System : Windows XP Pro SP1
Graphics Card : ATI FireGL X1
Resolution : 1600 x 1200
Color Depth : 32 bit
*************************************************
Rendering (Single CPU): 306 CB-CPU
Rendering (Multiple CPU): 650 CB-CPU
Multiprocessor Speedup: 2.12
Shading (CINEMA 4D) : 301 CB-GFX
Shading (OpenGL Software Lighting) : 1267 CB-GFX
Shading (OpenGL Hardware Lighting) : 2084 CB-GFX
OpenGL Speedup: 6.92
*************************************************

N.B. The latter is with HT turned on, hence the "4" processors.

[Eug Wanker]

Originally posted by Big Mac:
AFAIK, the author of xBench says the low scores are due to a known issue with the app itself. It's really wonderful to get all worked up over nothing.

"Xbench 1.0 uses the vec_dst instruction, in the altivec test, and the memory stream tests, which causes a big performance hit on the G5. I talked to some Apple engineers at WWDC about the issue and am working on a 1.1 release that avoids using vec_dst when running on the G5.
The new release should be out real soon.
Best, Ladd"

[Bill Harrison]

Hmmmmmmmmmm..... My slightly Underclocked system pulled 468 on the cinebench 2k3. Thats with the fsb running at 128mhz.... (i need to remove the processors and reset the multiplier a little lower, but have not gotten around to it.)

[michaelb]

Originally posted by Eug Wanker:
CINEBENCH 2003 v1
*************************************************
Rendering (Single CPU): 306 CB-CPU
Rendering (Multiple CPU): 650 CB-CPU
Multiprocessor Speedup: 2.12

I'm not disputing the accuracy of the cut & paste, but how does it work that having an extra processor MORE than doubles the performance? It seems to defy commonsense!

HyperThreading I suppose ("4" processors), but then the single CPU would benefit from that too ("2" processors still should be at least half of "4" shouldn't they?).

Or maybe there is system overhead attached with using one CPU that is overcompensated for when the extra processor kicks in.

In that case a properly G5 optimized Cinema4D renderer could also start screaming on a dual-2.0 GHz G5 as well. Go Maxon!

[Eug Wanker]

Single CPU is 1 CPU without HT.

Multiple CPU is 2 CPUs with HT = "4" CPUs

BTW, a single CPU with HT turned on is about 350... 2x350 > 650.

CINEBENCH 2003 v1
*************************************************
Tester : Huigie
Processor : 3GHz P4/HT
MHz : 3260
Number of CPUs : 2
Operating System : XP
Graphics Card : 9700Pro
Resolution : <1024x768>
Color Depth : <32bits>
*************************************************
Rendering (Single CPU): 319 CB-CPU
Rendering (Multiple CPU): 383 CB-CPU
Multiprocessor Speedup: 1.20
Shading (CINEMA 4D) : 327 CB-GFX
Shading (OpenGL Software Lighting) : 1491 CB-GFX
Shading (OpenGL Hardware Lighting) : 2544 CB-GFX
OpenGL Speedup: 7.78
*************************************************