Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
PowerMac 1.6GHz Benchmarks
- Thread starter MacRumors
- Start date
- Sort by reaction score
That would be why when my faithful B&W G3 finally died last month I went out and bought one of the last Dual 1.25 FW800 PowerMacs in my city...and not waiting on the G5...seeing that I got a helluva deal (last one CompUSSR had...talked them down to $1400)...dropped 2GB of CAS 2 PC2700...another 80GB Barracuda (and 2 more 40's in RAID 0 on the ATA66 chain)...PLUS with the money I saved over the Dual 1.8 (or 2.0) G5...let me justify a 23" Cinema HD https://forums.macrumors.com/images/smilies/biggrin.gif
...and as another sidebar...the G5 1.6 scores a 128 on XBench...and the Twin 2.0 scores 164...so I'm happy with the 146 mine rates...and really happy for NOT paying G5 prices (ok...I really like my Flat-Panel too)
NavyIntel007
macrumors 65816
More photos please...
... of the keyboard, mouse (I assume they're the new "keyboard" and "mouse"), and how about some shots of the packing materials?
What does "about this macintosh" show?
Enough bickering about the benchmarks already. Four pages worth, sheesh. MR ought to recycle the posts from all the old benchmark-bickering threads.
... of the keyboard, mouse (I assume they're the new "keyboard" and "mouse"), and how about some shots of the packing materials?
What does "about this macintosh" show?
Enough bickering about the benchmarks already. Four pages worth, sheesh. MR ought to recycle the posts from all the old benchmark-bickering threads.
i heard it comes with a sightly modified emac keyboard and mouse...so much for the bluetooth or metal finish keyboards...🙁
reminder for noobs
I keep seeing posts referencing the Xbench benchmarks...
this app is broken on the G5... hell the 1.6 is getting faster scores than the 2.0s in some cases, surely this can't be correct.
These benchmarks are totally unrepresentative of G5 performance. Being proud that your g4 has comparable xbench performance to the g5 is a moot point.
Jason
I keep seeing posts referencing the Xbench benchmarks...
this app is broken on the G5... hell the 1.6 is getting faster scores than the 2.0s in some cases, surely this can't be correct.
These benchmarks are totally unrepresentative of G5 performance. Being proud that your g4 has comparable xbench performance to the g5 is a moot point.
Jason
jayscheuerle
macrumors 68020
Big machine...
Interesting that on Apple's own page, they had to scale up a PC to appear to be as large as the G5. Even then, the PC still has 2 optical drives, a floppy/zip drive, space for 2 hard-drives...
Half of the G5's space appears to be taken up by fans and heat-sinks...
Interesting that on Apple's own page, they had to scale up a PC to appear to be as large as the G5. Even then, the PC still has 2 optical drives, a floppy/zip drive, space for 2 hard-drives...
Half of the G5's space appears to be taken up by fans and heat-sinks...
Keyboard and mouse?
What I really want to know is... what keyboard and mouse ship with the thing?
Are they bluetooth? are they the "non-pro" white versions that ship with the iMacs?
(Web boards are such a PITA to use, I apologise for not already searching through screens of comments for the answer)
-b
What I really want to know is... what keyboard and mouse ship with the thing?
Are they bluetooth? are they the "non-pro" white versions that ship with the iMacs?
(Web boards are such a PITA to use, I apologise for not already searching through screens of comments for the answer)
-b
Audio Guys
You know, as much as I want the spiffy Dual G5 2Ghz doing so would require that I get all new PCI cards and as an audio guy I don't want to have to reinvest into all my PCI hardware. So even though the Single 1.6 may not be the end all be all of computers it may be the right computer for me and others like me.
BTW, MOTU just annouced a trade in program for their 424 PCI cards, check out their site. Othewise if you have a UAD or a PCI Powercore or any PCI audio interface card (like myself) then my understanding is that they won't work on the new 3.3V PCI-X architecture, but will work just fine in the 1.6Ghz machine because it has the older architecture PCI slots.
Time for either firewire powercore or firewire something else I 'spose. I really do think the 1.6Ghz would be a great machine for audio with its increased bus speed and superior hard drive storage architecture and RAM. I want someone to post some info on how many virtual synths it can play and tracks it can record etc etc. Mostly the recording side is of interest to me.
peace,
James
You know, as much as I want the spiffy Dual G5 2Ghz doing so would require that I get all new PCI cards and as an audio guy I don't want to have to reinvest into all my PCI hardware. So even though the Single 1.6 may not be the end all be all of computers it may be the right computer for me and others like me.
BTW, MOTU just annouced a trade in program for their 424 PCI cards, check out their site. Othewise if you have a UAD or a PCI Powercore or any PCI audio interface card (like myself) then my understanding is that they won't work on the new 3.3V PCI-X architecture, but will work just fine in the 1.6Ghz machine because it has the older architecture PCI slots.
Time for either firewire powercore or firewire something else I 'spose. I really do think the 1.6Ghz would be a great machine for audio with its increased bus speed and superior hard drive storage architecture and RAM. I want someone to post some info on how many virtual synths it can play and tracks it can record etc etc. Mostly the recording side is of interest to me.
peace,
James
Re: reminder for noobs
What about Cinebench? Is everything "broken" on the G5? I'm guessing I'll just have to see for myself once I get my own machine. Hopefully it'll be as fast as all those bakeoffs showed it to be (and the Pixar Renderman test), but the 1.6 doesn't look too promising.
Anybody get a 1.8 yet to benchmark? Maybe the 1.6 is hobbled in a couple more areas than we though.
What about Cinebench? Is everything "broken" on the G5? I'm guessing I'll just have to see for myself once I get my own machine. Hopefully it'll be as fast as all those bakeoffs showed it to be (and the Pixar Renderman test), but the 1.6 doesn't look too promising.
Anybody get a 1.8 yet to benchmark? Maybe the 1.6 is hobbled in a couple more areas than we though.
Re: PowerMac 1.6GHz Benchmarks
Cinebench uses 64 Bit double precision floating point math which, once combined with their opinion that raytracing tends to not be vector friendly, sees to it that they make no use of SIMD extensions (Altivec/SSE/etc.), on either the PC or Mac platform.
That makes Cinebench a pure test of raw Floating Point performance using code compiled with the 7450 G4 as the target. Thus, looping back to caveat 1: this test is likely not making full/efficient use of the G5's FP units at this time.
This Ars Post from a Maxon developer for further reference
There's perhaps one other, or a corolary to caveat 1 at least:
Cinebench uses 64 Bit double precision floating point math which, once combined with their opinion that raytracing tends to not be vector friendly, sees to it that they make no use of SIMD extensions (Altivec/SSE/etc.), on either the PC or Mac platform.
That makes Cinebench a pure test of raw Floating Point performance using code compiled with the 7450 G4 as the target. Thus, looping back to caveat 1: this test is likely not making full/efficient use of the G5's FP units at this time.
This Ars Post from a Maxon developer for further reference
Maybe you missed the part about "at the same time".
This is entirely untrue. Non-real-time 3D rendering is not affected at all by the GPU.
Now if you are talking games, then yes of course the 9700 is much better than the 5200. But you can get a 9800 for the G5, and with that addition games will be much faster on the G5 than the Dual G4. Games make heavy use of the memory bus.
And these will only make a difference if you are CPU-limited, not bus-limited. How hard is it to saturate a 167 MHz bus?
As someone else pointed out, it is now. And more and more will be optimized with time.
Re: Re: That's A Big Box !!
Why do you think they made it that way? They can't afford people doing just that.
Why do you think they made it that way? They can't afford people doing just that.
Just a rather crude comparison but this should explain at least to some extent why the G5 won't significantly outpace a comparative G4 in some benchmarks:
Legend:
G4: Motorola 7455
K7: AMD Athlon
P4: Intel Pentium 4
G5: IBM PPC970
Process Technology:
G4: .18 micron, SOI
K7: .13 micron
P4: .13 micron
G5: .13 micron, SOI
Clockspeed:
G4: 1400 MHz
K7: 2200 MHz
P4: 3200 MHz
G5: 2000+ MHz
Nothing really new here, most of this is already known to all of us so I'll just leave this as it is.
Pipeline Stages:
G4: 7 stages
K7: 10 stages (?)
G5: 16 stages
P4: 20 stages
Branch History Table:
G4: 2048 entires
K7: 2048 entries
P4: 4096 entries
G5:16000 entries (!)
Average Hitrate:
G4: 93% (?)
K7: 93%
P4: 96%
G5: 97%+
Penalty Estimations:
G4: (.07*7= .49)
K7: (.07*10=.7)
P4: (.04*20)= .8)
G5: (.03*16)= .48) or with a 98% Hitrate (.02*16)=.32)
I didn't find any data regarding the G4's branch prediction although it'll probably have a hitrate similar to the Athlon. The G5 is going to have excellent branch prediction, 16000 entries is simply amazing, this is going to more than make up for it's greater branch mispredict penalty.
Integer Instruction Issue Rate (IPC):
G4: 3 instructions/cycle
K7: 3 instructions/cycle
P4: 4 instructions/cycle
G5: 2 instructions/cycle
FP Instruction Issue Rate (IPC):
G4: 1 instruction/cycle
K7: 3 instructions/cycle
P4: 1 instruction/cycle
G5: 2 instructions/cycle
Vector Instruction Issue Rate (IPC):
G4: 2-4 instructions/cycle (Altivec)
K7: 4 single instructions/cycle (SSE)
P4: 4 single/ 2 double instructions/cycle (SSE2)
G5: 2-4 instructions/cycle (Altivec)
Max instructions execute+retire (IPC):
G4: 3 instructions/cycle
K7: ??????????????? (if anyone knows, please post it)
P4: 3 micro-ops/cycle
G5: 4 instructions/cycle
OoO Window:
G4: 16 instructions
K7: 72 instructions
P4: 126 instructions
G5: 215 instructions
This is the important part, while the G5 will be able to extract more ILP out of code, and execute+retire more instructions/cycle, it's two ALUs limits it to only two integer instructions/cycle, this is perhaps the main reason why we might not see the G5 perform as impressively in certain benchmarks. Most likely, it will achieve parity to a similarly clocked G4 in typical integer code. However, the G5 should perform excellently in floating point code, largely due to it's dual double precision FPUs. Theres no doubt that the G5 will easily exceed the G4 in this aspect and probably the Athlon (triple FPU/extended precision:80 bit) as well due to it's ability to extract more ILP out of code (that contains more ILP and less load-stores). Vector performance should be up there with the G4 which is, to say the least, quite impressive.
L1 Cache (Instruction / Data):
G4: 32 KB / 32 KB
K7: 64 KB / 64 KB
P4: 12 KB / 08 KB
G5: 64 KB / 32 KB
L2 Cache:
G4: 256 KB (256 Bit)
K7: 512 KB (64 Bit)
P4: 512 KB (256 Bit)
G5: 512 KB (256 Bit)
Front Side Bus:
G4: 166 MHz
K7: 400 MHz
P4: 800 MHz
G5: 700-1000 MHz
Not much else to say here that isn't already known, it should be noted that the G4 also has a L3 cache that alleviates some of the bandwidth bottleneck, the fast fsb will definitely help the G5 in memory intensive apps. I haven't seen anything about the G5's cache latencies or hitrates so I can't comment on them yet.
The main point is, the G5 probably won't look as impressive in some benchmarks (heavy in integer operations) as it does in other benchmarks (heavy in floating point operations). Perhaps Cinebench employs rendering techniques (i.e raytracing) that are written in integer code. If thats the case, it should come as little surprise that the G5 doesn't perform as well in this benchmark relative to others we've seen so far.
Legend:
G4: Motorola 7455
K7: AMD Athlon
P4: Intel Pentium 4
G5: IBM PPC970
Process Technology:
G4: .18 micron, SOI
K7: .13 micron
P4: .13 micron
G5: .13 micron, SOI
Clockspeed:
G4: 1400 MHz
K7: 2200 MHz
P4: 3200 MHz
G5: 2000+ MHz
Nothing really new here, most of this is already known to all of us so I'll just leave this as it is.
Pipeline Stages:
G4: 7 stages
K7: 10 stages (?)
G5: 16 stages
P4: 20 stages
Branch History Table:
G4: 2048 entires
K7: 2048 entries
P4: 4096 entries
G5:16000 entries (!)
Average Hitrate:
G4: 93% (?)
K7: 93%
P4: 96%
G5: 97%+
Penalty Estimations:
G4: (.07*7= .49)
K7: (.07*10=.7)
P4: (.04*20)= .8)
G5: (.03*16)= .48) or with a 98% Hitrate (.02*16)=.32)
I didn't find any data regarding the G4's branch prediction although it'll probably have a hitrate similar to the Athlon. The G5 is going to have excellent branch prediction, 16000 entries is simply amazing, this is going to more than make up for it's greater branch mispredict penalty.
Integer Instruction Issue Rate (IPC):
G4: 3 instructions/cycle
K7: 3 instructions/cycle
P4: 4 instructions/cycle
G5: 2 instructions/cycle
FP Instruction Issue Rate (IPC):
G4: 1 instruction/cycle
K7: 3 instructions/cycle
P4: 1 instruction/cycle
G5: 2 instructions/cycle
Vector Instruction Issue Rate (IPC):
G4: 2-4 instructions/cycle (Altivec)
K7: 4 single instructions/cycle (SSE)
P4: 4 single/ 2 double instructions/cycle (SSE2)
G5: 2-4 instructions/cycle (Altivec)
Max instructions execute+retire (IPC):
G4: 3 instructions/cycle
K7: ??????????????? (if anyone knows, please post it)
P4: 3 micro-ops/cycle
G5: 4 instructions/cycle
OoO Window:
G4: 16 instructions
K7: 72 instructions
P4: 126 instructions
G5: 215 instructions
This is the important part, while the G5 will be able to extract more ILP out of code, and execute+retire more instructions/cycle, it's two ALUs limits it to only two integer instructions/cycle, this is perhaps the main reason why we might not see the G5 perform as impressively in certain benchmarks. Most likely, it will achieve parity to a similarly clocked G4 in typical integer code. However, the G5 should perform excellently in floating point code, largely due to it's dual double precision FPUs. Theres no doubt that the G5 will easily exceed the G4 in this aspect and probably the Athlon (triple FPU/extended precision:80 bit) as well due to it's ability to extract more ILP out of code (that contains more ILP and less load-stores). Vector performance should be up there with the G4 which is, to say the least, quite impressive.
L1 Cache (Instruction / Data):
G4: 32 KB / 32 KB
K7: 64 KB / 64 KB
P4: 12 KB / 08 KB
G5: 64 KB / 32 KB
L2 Cache:
G4: 256 KB (256 Bit)
K7: 512 KB (64 Bit)
P4: 512 KB (256 Bit)
G5: 512 KB (256 Bit)
Front Side Bus:
G4: 166 MHz
K7: 400 MHz
P4: 800 MHz
G5: 700-1000 MHz
Not much else to say here that isn't already known, it should be noted that the G4 also has a L3 cache that alleviates some of the bandwidth bottleneck, the fast fsb will definitely help the G5 in memory intensive apps. I haven't seen anything about the G5's cache latencies or hitrates so I can't comment on them yet.
The main point is, the G5 probably won't look as impressive in some benchmarks (heavy in integer operations) as it does in other benchmarks (heavy in floating point operations). Perhaps Cinebench employs rendering techniques (i.e raytracing) that are written in integer code. If thats the case, it should come as little surprise that the G5 doesn't perform as well in this benchmark relative to others we've seen so far.
With Audio work, you are talking about a lot of Math functions so it would be best to migrate over to native OS X apps when you can, since the G5 has much better math processing capability than any Intel/AMD processor-Audio benefits from a lot of vector math too. In the meantime, even though you have 1MB of RAM, make sure you allocated the max 512MB to the VPC session. Then after starting your Windows session, if you are running XP, turn off the Aqua look and go to the Classic look. For XP and win2k: Turn off anti-aliasing and Show Window when dragging. Turn off shadowing, animations and for the most part 16-bit color is more than enough. Under networking turn off the firewall if it is on and turn off File and print sharing in the protocol stack.
Whenever possible use Win2k over XP and Home over Pro. Win2k and XP are functionally identical at the core as far as any current user needs to be concerned. XP is basically v5.1 and Win2K is v5.0. But in MSFT's case 5.1 is more bloated and slower than 5.0.
I am running VPC with Win2Kpro on a G4/500MHz AGP with 512MB allocated to the VPC session and it is pretty snappy. It is actually faster in many areas to my Compaq 500Mhz with 25MB and just a little slower than my 1GHz Pentium III Dell with 384 MB. I use for VS Studio 2003 development, if that isn't a kick in the pants.
Re: Re: $1590 !
And has been stated before, some of us don't have this option. I will personally spend most of this summer/fall out of cell-phone range, let alone near any type of 120v AC source. In fact, I drove over 350 miles yesterday to come home, and will drive it again tomorrow. And I need Windows, as much as I may dislike MS products.
(tig)
And has been stated before, some of us don't have this option. I will personally spend most of this summer/fall out of cell-phone range, let alone near any type of 120v AC source. In fact, I drove over 350 miles yesterday to come home, and will drive it again tomorrow. And I need Windows, as much as I may dislike MS products.
(tig)
Photoshop will run a lot faster (with the G5 plugin) on the Dual 2.0 than it did on the Dual 1.42. Like, twice as fast. And yes, it will also easily beat the Xeon.
The same will be true of many other apps when they are recompiled for the G5.
Once again, the XBench numbers are worthless. Ignore them. I would also not worry too much about the Cinebench numbers until the app has been recompiled for the G5 (though I don't know how soon that will happen since apparently Maxon uses CodeWarrior, not GCC). The fact that it is beating the G4 by 10 to 20 percent per clock cycle in Cinebench is actually pretty impressive given that the code is compiled with the MPC 7450 as the target processor. Other apps will show much bigger gains, especially when they have been recompiled for the G5.
Everyone seems to be forgetting how the MPC 7450 was considered a dud at its release because it was performing 20% slower than the MPC 7400 core, clock for clock (see, for example, http://www.xlr8yourmac.com/G4ZONE/G4_733mhz_review/apple_G4_733_tests.html OR http://www.barefeats.com/G4733.htm). But today the 7450 core matches the 7400 core clock for clock (the 7400 is no longer produced, but it is literally the same core as the G3 750FX, so you can compare the iBook 900 to the PB 12" 867 on non-Altivec tasks to see what I am saying). We are also forgetting how the 1.5 Ghz Pentium 4 was considered an abject failure on release day because it could barely keep up with a 1 Ghz Athlon, but today the 3 Ghz P4 will spank a 2 Ghz Athlon (1.5/1 is the same ratio as 3/2, if that isn't clear).
Actually, you REALLY ought to take a look at the second link (http://www.barefeats.com/G4733.htm)! The Cinebench (Cinema 4D) benchmark in it "proves" that the new 733 Mhz G4 is actually SLOWER than the 533 Mhz G4 it replaced! Yes, that's right, the single G4/533 ran the test in 44 seconds vs. the single G4/733 time of 47 seconds. In other words, before Cinema 4D was recompiled for the MPC 7450, it actually ran 33% SLOWER clock for clock on the MPC 7450 than it did on the MPC 7400. But today, with Cinebench recompiled for the MPC 7450, it runs at the same speed clock for clock on an MPC 7450 and an MPC 7400 (in other words, there was a speed gain of almost 50% from recompiling for the 7450). So, like I said, it is really impressive that the non-recompiled Cinebench app is running 10-20% FASTER clock for clock on the G5 versus the G4.
The moral of the story is DON'T judge a new processor before the before binaries have been recompiled for it.
tazznb
macrumors regular
When I used Photoshop on a G5 at the expo it was a lot faster than my dual Xeon machine I use.
Example: I resized a.psd file from a 56Mb (can't remember the exact size right now) all the way to 2.5Gb I was able to see it finish in under a minute.
I went home to try it on my Dell Xeon (with UW SCSI 3),and ended up walking away so it can finish it.
allpar
macrumors 6502
Comparing the 1.6 G5 to a dual 1.25 G4 is very appropriate since they cost roughly the same amount. (The dual g4 is a bit cheaper.)
With an unoptomized system and not-G5-aware XBench, the G5 was slower than the dual G4 1.25 - faster on memory, slower on hard drive, slower on CPU. This isn't very surprising.
The G4's dual processor design also means it is sometimes very snappy where the G5 might not be.
It is not really fair at this point to compare a dual 2.0 to a dual 1.25 G4, since the dual 1.25 G4 is selling for $1,500 or so and the dual 2.0 for nearly $3,000. The current comparison should be single 1.6 to dual 1.25...where the G5 is only a couple of hundred dollars more expensive.
I'll add that the smaller cases on the G4s have room for two optical drives (but not cartridge DVD-RAM!!!) and four hard drives! And the current models are pretty darned quiet. Case design is STUPID though with no reset button, and the on button ONLY on the case (unless you have an Apple monitor) and no possibility of a cartridge optical drive or zip drive. My $50 PC case has four external bays and two internal bays - I have in it a catrdige drive ($8), a DVD-RAM I used to have in my B&W G3 which I got "new-used" for $80 some time ago, a CD-writer, and a CD-reader. Yup, using 'em all. I'd rather have the DVD-RAM in the Mac...but it's not going to happen. (The second optical slot will eventually go to a Superdrive when they reach $100.)
HOWEVER, of course, the G5 has a lot more room to grow - as more programs are optimized for the G5 it will become faster.
With an unoptomized system and not-G5-aware XBench, the G5 was slower than the dual G4 1.25 - faster on memory, slower on hard drive, slower on CPU. This isn't very surprising.
The G4's dual processor design also means it is sometimes very snappy where the G5 might not be.
It is not really fair at this point to compare a dual 2.0 to a dual 1.25 G4, since the dual 1.25 G4 is selling for $1,500 or so and the dual 2.0 for nearly $3,000. The current comparison should be single 1.6 to dual 1.25...where the G5 is only a couple of hundred dollars more expensive.
I'll add that the smaller cases on the G4s have room for two optical drives (but not cartridge DVD-RAM!!!) and four hard drives! And the current models are pretty darned quiet. Case design is STUPID though with no reset button, and the on button ONLY on the case (unless you have an Apple monitor) and no possibility of a cartridge optical drive or zip drive. My $50 PC case has four external bays and two internal bays - I have in it a catrdige drive ($8), a DVD-RAM I used to have in my B&W G3 which I got "new-used" for $80 some time ago, a CD-writer, and a CD-reader. Yup, using 'em all. I'd rather have the DVD-RAM in the Mac...but it's not going to happen. (The second optical slot will eventually go to a Superdrive when they reach $100.)
HOWEVER, of course, the G5 has a lot more room to grow - as more programs are optimized for the G5 it will become faster.
This was a really great explaination, but for those of us who've bought the G5 what does it mean? Will we have to wait awhile with underpar performance, waiting for optimized and recompiled apps? For those people who bought the 3Ghz P4 I'm sure life is good compared to their AMD brethren, but what about those on the 1.5 P4-did they get screwed (or did the new line of applications that push the 3Ghz P4 past the Athlon also help the 1.5P4 against the 1Ghz Athlon?)?
allpar
macrumors 6502
In the case of the 1.6 and 1.8, I believe these will provide out of the box performance somewhat exceeding the dual G4 (feel sorry for those who just spent $3,000 on a dual G4!).
In time, while the G4 seems to get slower, the G5 will remain current or get faster as 10.3 comes out, and other app makers recode. Examples - photoshop (already done!), Safari, VirtualPC.
The G5 is NOT slower than the dual G4 in real life, I'd guess, but the 1.6 is probably similar - and since the price is similar, that makes sense.
But there ARE advantages to dual processors!
In time, while the G4 seems to get slower, the G5 will remain current or get faster as 10.3 comes out, and other app makers recode. Examples - photoshop (already done!), Safari, VirtualPC.
The G5 is NOT slower than the dual G4 in real life, I'd guess, but the 1.6 is probably similar - and since the price is similar, that makes sense.
But there ARE advantages to dual processors!