Apple releases dual 90 nm (66 mm2 die size) 64-bit G5 2.0 GHz Xserve 1U with ECC support *Pix*

[Eug]

I don't know about you, but I think that if IBM can pull it off, niether Intel nor AMD will have anything close to the processing power of a 3GHz G5 by the same time frame.

intel's itanium2 has gobs of processing power if spec benchmarks are to be believed.

Yeah, I2 is a monster. Unfortunately, there is no true mainstream OS for I2. Windows runs on Opteron/Athlon 64 and on Prescott. Mac OS X runs on PPC 970FX. Linux doesn't count.

 

[jhu]

actually, windows does run on itanium

 

[MercenaryForHire]

Originally posted by: jhu
And I notice how you haven't said one peep about the Athlon FX and Opterons trashing your precious little baby G5.

Athlon 64 vs. Apple G5 Systems: Not Even Close

- M4H


uhm, those benchmarks came out a while ago and have been fairly well dissected by the nerds over at slashdot. two main complaints of the pc world test is that no one who uses a mac actually uses adobe premiere. they use final cut pro which is actually an os x app instead of an ox 9 app. second is the word benchmark. basically who cares about a word benchmark?

So basically the only benchmarks that matter are where Apple wins, and everything else is invalid because No One Uses That.

Yeah, no one uses Photoshop either.

- M4H

 

[klah]

Digital Photography Insights: G5 vs x86

x86 wins 58 out of 77 benchmarks.

 

[Sohcan]

Originally posted by: Eug

I don't know about you, but I think that if IBM can pull it off, niether Intel nor AMD will have anything close to the processing power of a 3GHz G5 by the same time frame.

intel's itanium2 has gobs of processing power if spec benchmarks are to be believed.

Yeah, I2 is a monster. Unfortunately, there is no true mainstream OS for I2.

Sure there is. Full production versions of Windows Server 2003, Enterprise and Datacenter versions, were released last April. The Standard edition is currently in beta mode. MS also has a special version of XP for Itanium that ships with HP's Itanium 2 workstations.

Then as you get less mainstream, IPF also runs Linux (with retail versions of Redhat and SUSE Enterprise), HP-UX 11i v2, OpenVMS, and NSK....and Solaris, though that's deeply hidden away somewhere at Sun.

 

[LethalWolfe]

Originally posted by: MercenaryForHire

Originally posted by: jhu
And I notice how you haven't said one peep about the Athlon FX and Opterons trashing your precious little baby G5.

Athlon 64 vs. Apple G5 Systems: Not Even Close

- M4H


uhm, those benchmarks came out a while ago and have been fairly well dissected by the nerds over at slashdot. two main complaints of the pc world test is that no one who uses a mac actually uses adobe premiere. they use final cut pro which is actually an os x app instead of an ox 9 app. second is the word benchmark. basically who cares about a word benchmark?

So basically the only benchmarks that matter are where Apple wins, and everything else is invalid because No One Uses That.

Yeah, no one uses Photoshop either.

- M4H

The Word and Premiere benches are flawed, IMO, and were the only ones that didn't make any sense as benchmarks. Word on the Mac is visibly slower than Word on the PC. It's not a great port. Much like iTunes on the PC is visibly slower than iTunes on the Mac. If I

The use of Premiere 6.5 (a three year old app) just makes no sense at all. Why not use a newer program that appears on both platfroms like Avid XpressDV? Or if all you want to do is measure encoding power why not use Cleaner? And the benchmarking, in these tests, just appeared sloppy. They gave no relevent details about the settings they used or the situtation they created. "In Adobe Premiere we timed the rendering of our workspace and timed the export of a 959-frame movie at 720 by 480 resolution and 30 frames per second into the QuickTime format"

What does "...timed the rending our or workspace..." mean? You can't render anything unless you manipulate the video (which of course they don't mention). "...timed the export... into the QuickTime format." That's about as useful as saying "timed the export into AVI format." There are dozens of codecs that can carry the .MOV extension and each of those codecs can have a multitude of settings applied to them.


I'm not saying the results are wrong, but the choice of a couple of the programs and lack of detail about how the benches were carried out makes the entire thing sloppy and, at the very least, questionable.


Lethal

 

[Eug]

Actually, what I meant by mainstream is a home desktop OS. That rules out Windows Server and Linux IMO. I was thinking more along the lines of a Power Mac vs. a Prescott or Athlon 64 home machine.

But if we're talking about blade servers, etc. then you are correct.

 

[GL]

Originally posted by: LethalWolfe

Originally posted by: MercenaryForHire

Originally posted by: jhu
And I notice how you haven't said one peep about the Athlon FX and Opterons trashing your precious little baby G5.

Athlon 64 vs. Apple G5 Systems: Not Even Close

- M4H


uhm, those benchmarks came out a while ago and have been fairly well dissected by the nerds over at slashdot. two main complaints of the pc world test is that no one who uses a mac actually uses adobe premiere. they use final cut pro which is actually an os x app instead of an ox 9 app. second is the word benchmark. basically who cares about a word benchmark?

So basically the only benchmarks that matter are where Apple wins, and everything else is invalid because No One Uses That.

Yeah, no one uses Photoshop either.

- M4H

The Word and Premiere benches are flawed, IMO, and were the only ones that didn't make any sense as benchmarks. Word on the Mac is visibly slower than Word on the PC. It's not a great port. Much like iTunes on the PC is visibly slower than iTunes on the Mac. If I

The use of Premiere 6.5 (a three year old app) just makes no sense at all. Why not use a newer program that appears on both platfroms like Avid XpressDV? Or if all you want to do is measure encoding power why not use Cleaner? And the benchmarking, in these tests, just appeared sloppy. They gave no relevent details about the settings they used or the situtation they created. "In Adobe Premiere we timed the rendering of our workspace and timed the export of a 959-frame movie at 720 by 480 resolution and 30 frames per second into the QuickTime format"

What does "...timed the rending our or workspace..." mean? You can't render anything unless you manipulate the video (which of course they don't mention). "...timed the export... into the QuickTime format." That's about as useful as saying "timed the export into AVI format." There are dozens of codecs that can carry the .MOV extension and each of those codecs can have a multitude of settings applied to them.


I'm not saying the results are wrong, but the choice of a couple of the programs and lack of detail about how the benches were carried out makes the entire thing sloppy and, at the very least, questionable.


Lethal

I agree with you. MS Office is vastly different on each platform. The code base isn't even the same. I wouldn't so much call it a port since the Mac Business Unit is responsible for MS Office for Mac and they don't just take MS Office for Windows and redo it for Mac. In fact, this would be like comparing a retail version of a Chrysler with the same model's NASCAR peer. The only thing they really share in common is their name.

The use of Premiere is laughable because it doesn't even run natively on OS X but has to be run in Classic. That would be like benchmarking an app on VMWare on the PC and drawing any meaningful performance conclusions.

Unfortunately, Macworld participated in these benchmarks so it is perceived to have given credibility to them.

One of the reasons I'm excited that Anand is now a Mac user is that there is a distinct lack of quality benchmarking procedures on the Mac. So it is very hard to draw any conclusions about their performance. Though I think it is safe to say that a dual G5 is in the same ballpark of performance as a high-end Pentium 4 or Athlon 64. I am sure the benchmarks go either way depending on how you slice it.

That being said, it will still be difficult to compare platforms. Is it fair to compare the performance of a videogame on the PC when it's code has been hand-rolled to perform best on that platform, to that videogame's port on the Mac which wasn't given as much optimization attention due to the labour & expense vs revenue trade-off? Some would say no. But there's an argument to be had for doing it anyways because in practical terms it will tell you which platform is better for that particular task.

But if we're going to slip into this gray area of benchmarking, all I ask is that everything be taken into consideration. For instance, it's not fair to compare the speed of travel within an urban center to that in a rural setting by just looking at road congestion. It seems to be an open and shut case. The urban center has roads that are congested whereas the rural setting is lucky to have half a dozen cars on any stretch of road any given time. But in many urban centers it's faster to walk or take the subway to your destination, and many attractive destinations are closer to your starting point. So what if I live out in farmland and can drive at 120 km/h without a car in sight if the nearest hospital is 200 km away? In an urban center, that hospital might be 10 minute cab ride, even if it is only 3 blocks away.

You must take into consideration all variables at hand. OS X has a lot of tricks up its sleeve that make working on it pretty efficient. For instance, why does it matter if Adobe Premiere on the PC can output a video file twice as fast as Final Cut Pro on an Apple if doing the editing work on the Apple takes 1/2 of the time or less? How do you quantify this? You can't very easily do so.

 

[Sohcan]

Originally posted by: Eug
Actually, what I meant by mainstream is a home desktop OS. That rules out Windows Server and Linux IMO. I was thinking more along the lines of a Power Mac vs. a Prescott or Athlon 64 home machine.

But if we're talking about blade servers, etc. then you are correct.

I would say Windows XP counts.

 

[Eug]

Originally posted by: Sohcan

Originally posted by: Eug
Actually, what I meant by mainstream is a home desktop OS. That rules out Windows Server and Linux IMO. I was thinking more along the lines of a Power Mac vs. a Prescott or Athlon 64 home machine.

But if we're talking about blade servers, etc. then you are correct.

I would say Windows XP counts.

Well, I stand corrected though. I didn't know Win XP 64-bit ran on I2.

How much of the current 32-bit apps run on it by the way? I would guess most of the drivers for a lot of stuff wouldn't, but how about just plain MS apps and 3rd party apps?

The reason I ask is somehow I just wrap my brain around using I2 for a mainstream box. Opteron yes, but I2 or I2-derivative no.

 

[jhu]

i would suspect that x86 programs would run at about the same speed as an equivalently clocked xeon (non-driver programs of course). there was recently an article comparing/contrasting various ways that x86 is emulated. the article studied fx!32 for alpha, transmeta's "code morphing", and intel's software emulator.

 

[Sohcan]

Originally posted by: Eug

Well, I stand corrected though. I didn't know Win XP 64-bit ran on I2.

How much of the current 32-bit apps run on it by the way? I would guess most of the drivers for a lot of stuff wouldn't, but how about just plain MS apps and 3rd party apps?

Assuming there aren't any incompatibilies with WOW64 (which would be due to Windows, not Itanium, see Anand's recent XP 64-bit article), all 32-bit apps run. The new emulation layer provides 50% of native performance for integer apps, and 30-40% for FP. It's hoped that the emulation layer can be improved to provide up to 70% of native performance in the future.

The reason I ask is somehow I just wrap my brain around using I2 for a mainstream box. Opteron yes, but I2 or I2-derivative no.

Oh, I'm not saying that anyone would use an IPF workstation for a home PC, I just wanted to point out that there is a version of XP 64-bit for it.

 

[Eug]

Thanks for the links. I know you're not claiming it has a huge market, but I don't see who would be buying HP Itanium 2 workstations with Win XP 64 at all, even though they do exist. What's the size of the market for this product? Or are they all running Linux anyway?

I guess the new Intel-64 is gonna be the chip of choice for Intel to go after this market, up against Opteron (and potentially 970FX), if we're talking 64-bit Linux.

 

[Sohcan]

Originally posted by: Eug
Thanks for the links. I know you're not claiming it has a huge market, but I don't see who would be buying HP Itanium 2 workstations with Win XP 64 at all, even though they do exist. What's the size of the market for this product? Or are they all running Linux anyway?

There have been a number of big-name engineering apps ported to Windows on IPF...the type of people who would use them are those that do mechanical CAD, CAE, visualization, EDA (electronic design automation), etc. Thus far HP has been the only supplier of IPF workstations, and given their custom chipset and system design, they have quite a price mark-up on the systems and components. SGI, Racksaver, and Celestica are releasing IPF workstations soon (the latter two based on Intel's 8870 chipset, I believe), so hopefully prices will go down.

 

[n0cmonkey]

http://www.ecommercetimes.com/perl/story/32837.html

Macs (like Sun's Ultrasparcs, Sunfires, and enterprise offerings; or just about anything from Cisco) are a platform. Not just a bunch of computer parts. Yes, other things are faster. Yes, others are cheaper. You don't like the platform? Ignore it. Go about your business.

 

[Eug]

IBM brags about the 970FX and its 90 nm process:

IBM today announced it has developed a new method of manufacturing low power, high performance microprocessors using an industry-first combination of silicon-on-insulator (SOI), strained silicon and copper wiring technologies.
IBM is putting the technique immediately to work in volume 90 nanometer production at its 300mm manufacturing facility. The company's award-winning 64-bit PowerPC 970FX microprocessor will be the first chip built using this trio of IBM technology breakthroughs.

 

[mrgoblin]

The Mac community may now Rejoice! While everyone else just shrugs and continues to use opterons. Oh well, theres always 65 nano

 

[Sunner]

Originally posted by: Sohcan

Originally posted by: Eug
Thanks for the links. I know you're not claiming it has a huge market, but I don't see who would be buying HP Itanium 2 workstations with Win XP 64 at all, even though they do exist. What's the size of the market for this product? Or are they all running Linux anyway?

There have been a number of big-name engineering apps ported to Windows on IPF...the type of people who would use them are those that do mechanical CAD, CAE, visualization, EDA (electronic design automation), etc. Thus far HP has been the only supplier of IPF workstations, and given their custom chipset and system design, they have quite a price mark-up on the systems and components. SGI, Racksaver, and Celestica are releasing IPF workstations soon (the latter two based on Intel's 8870 chipset, I believe), so hopefully prices will go down.

Sochan, since you're here, a little question(well more like a rant) for you :

One really annoying thing about I2 is the strange product lineup and the prices.
UltraSPARC-III's come in three flavours, 900 MHz, 1.05 GHz, and 1.2 GHz, all with 8 MB L2, etc, all the same CPU's only slightly different speed.
Same with most any other CPU out there.

I2 on the other hand comes in a varitey of versions.
1.5 GHz, 6 MB L3
1.4 GHz, 4 MB L3
1.3 GHz, 3 MB L3
1.4 GHz "DP" version, 1.5 MB L3

The high end ones are prohibitively expensive, buying three such CPU's from HP costs more than a fully loaded Sun Fire v440 with 4 1.3 GHz US-IIIi's, 16 GB of RAM, and some other goodies.
A full system with 4 CPU's and 8 GB of RAM costs upwards of $60.000.
The lower end ones aren't as expensive, but it's damn near impossible to find good performance information about the lower end ones since most benchmarks are with the 1.5 GHz version.
Now, if the 1.4 GHz one was just the same as the 1.5 GHz one, but slightly slower, it would be easy, subtract ~5-7 percent on the 1.5 GHz scores, but I have no idea how much of a difference those extra two megs of cache are going to make.

I just find this a strange and confusing aproach.

 

[Sohcan]

Originally posted by: Sunner


Sochan, since you're here, a little question(well more like a rant) for you :

One really annoying thing about I2 is the strange product lineup and the prices.
UltraSPARC-III's come in three flavours, 900 MHz, 1.05 GHz, and 1.2 GHz, all with 8 MB L2, etc, all the same CPU's only slightly different speed.
Same with most any other CPU out there.

I2 on the other hand comes in a varitey of versions.
1.5 GHz, 6 MB L3
1.4 GHz, 4 MB L3
1.3 GHz, 3 MB L3
1.4 GHz "DP" version, 1.5 MB L3

The high end ones are prohibitively expensive, buying three such CPU's from HP costs more than a fully loaded Sun Fire v440 with 4 1.3 GHz US-IIIi's, 16 GB of RAM, and some other goodies.
A full system with 4 CPU's and 8 GB of RAM costs upwards of $60.000.

You have to consider the systems, which is going to affect the markup quite a bit...switch from a v440 to a v480 (both capable up to 4-way), and a 4-way system with nearly identical specs (save the addition of an 8 MB external cache on the v480, but with the loss of 80 MHz in clockspeed and a faster integrated L2) and the price jumps from $26,000 to $43,000. The HP rx5670 has been superceded by the new slimmer and cheaper rx4640, which should be closer in price to the v480. The lower price systems are just starting to appear...the new HP rx1600 (released last week) starts at $2,700, and the SGI Altix 350 starts at $22,000 for a 4-way system.

The lower end ones aren't as expensive, but it's damn near impossible to find good performance information about the lower end ones since most benchmarks are with the 1.5 GHz version.
Now, if the 1.4 GHz one was just the same as the 1.5 GHz one, but slightly slower, it would be easy, subtract ~5-7 percent on the 1.5 GHz scores, but I have no idea how much of a difference those extra two megs of cache are going to make.

There's quite a bit of benchmarks on the 1.3 GHz and 1.4 GHz models at the SPEC website (see the Q4 2003 and Q1 2004, though take note that the submissions use a variety of compilers and chipsets), and SGI has some 1.3 GHz numbers here. The few numbers that I've seen using the 1 GHz / 1.5 MB L3 LV model place it close to the 1 GHz / 3MB L3 McKinley.

 

[Sunner]

Yes indeed, there are always SPEC benchmarks for everything out there, but once you move out of their site, it becomes harder to find benchmarks.
This is especially true once you move on to more "obscure" software where people don't take their time to benchmark on many configurations.

Granted, one could always blame the manufacturers/developers for not doing the benchmarking, but at least with US-III's, Xeons, POWER's, etc, it's easy to guesstimate performance based on the benchmarks that do exist.

And one thing to keep in mind with Sun is their ridiculously high list prices, you can often go between 20-40 percent below that, especially if you're willing to trade in some old hunk of junk for the new box.
For example, the v440 we're looking at has a list price of ~26.000 USD, we're trading in a old Ultra1E, and getting it for $18.500 instead.

 

[beachbreeze]

Originally posted by: mrgoblin
The Mac community may now Rejoice! While everyone else just shrugs and continues to use opterons. Oh well, theres always 65 nano

G5 970FX 2GHz : 24.5W (normal load)
Opteron 246 2.0: 84.7W to 89 W (different AMD sources) (Thermal Design Power)

G4 1.33: 45 W (normal load)
G5 2GHz: 55 W (normal load)
Athlon XP 3200+: 77 W (Thermal Design Power)
Intel's 3.2GHz P4: 82W (Thermal Design Power)
Itanium 2 1.5 6M: 107 W (Thermal Design Power)
Xeon 3.2: 110 W (Thermal Design Power)

Comments, please:
Laptops?
Scalability?

 

[Eug]

Originally posted by: beachbreeze

Originally posted by: mrgoblin
The Mac community may now Rejoice! While everyone else just shrugs and continues to use opterons. Oh well, theres always 65 nano

G5 970FX 2GHz : 24.5W (normal load)
Opteron 246 2.0: 84.7W to 89 W (different AMD sources) (Thermal Design Power)

G4 1.33: 45 W (normal load)
G5 2GHz: 55 W (normal load)
Athlon XP 3200+: 77 W (Thermal Design Power)
Intel's 3.2GHz P4: 82W (Thermal Design Power)
Itanium 2 1.5 6M: 107 W (Thermal Design Power)
Xeon 3.2: 110 W (Thermal Design Power)

Comments, please:
Laptops?
Scalability?

The G4 1.33 at 45 W and the G5 2 GHz at 55 W numbers are both max load, not normal load.

However, the G5 970FX 2 GHz isn't that fast anymore. I'd estimate Xeon 2.8 territory. The 2.5 GHz is VERY fast and is supposed to come out soon (being announced today), but what I've read is that its normal load usage is 52 Watts, which means probably means its max load is comparable to the Opteron. So:

G4 1.33: 45 W (max load)
G5 2 GHz: 55 W (max load)
Athlon XP 3200+: 77 W (Thermal Design Power)
Intel's 3.2 GHz P4: 82W (Thermal Design Power)
Opteron 246 2.0: 89 W (Thermal Design Power)
G5 2.5 GHz: ??? ~90 W (max load)
Itanium 2 1.5 6M: 107 W (Thermal Design Power)
Xeon 3.2: 110 W (Thermal Design Power)

 

[beachbreeze]

Originally posted by: Eug

Originally posted by: beachbreeze

Originally posted by: mrgoblin
The Mac community may now Rejoice! While everyone else just shrugs and continues to use opterons. Oh well, theres always 65 nano

G5 970FX 2GHz : 24.5W (normal load)
Opteron 246 2.0: 84.7W to 89 W (different AMD sources) (Thermal Design Power)

G4 1.33: 45 W (normal load)
G5 2GHz: 55 W (normal load)
Athlon XP 3200+: 77 W (Thermal Design Power)
Intel's 3.2GHz P4: 82W (Thermal Design Power)
Itanium 2 1.5 6M: 107 W (Thermal Design Power)
Xeon 3.2: 110 W (Thermal Design Power)

Comments, please:
Laptops?
Scalability?

The G4 1.33 at 45 W and the G5 2 GHz at 55 W numbers are both max load, not normal load.

However, the G5 970FX 2 GHz isn't that fast anymore. I'd estimate Xeon 2.8 territory. The 2.5 GHz is VERY fast and is supposed to come out soon (being announced today), but what I've read is that its normal load usage is 52 Watts, which means probably means its max load is comparable to the Opteron. So:

G4 1.33: 45 W (max load)
G5 2 GHz: 55 W (max load)
Athlon XP 3200+: 77 W (Thermal Design Power)
Intel's 3.2 GHz P4: 82W (Thermal Design Power)
Opteron 246 2.0: 89 W (Thermal Design Power)
G5 2.5 GHz: ??? ~90 W (max load)
Itanium 2 1.5 6M: 107 W (Thermal Design Power)
Xeon 3.2: 110 W (Thermal Design Power)

Interesting Eug... but do you really think there will be an almost doubling of power usage with the 25% increase in clock speed from G5 2.0GHz to 2.5GHz? Is that the kind of exponential relationship overclockers work to? I'm afraid I have no experience of overclocking. This PCMag benchmark reckons the G5s faster than you do:
Apple Power Mac G5: Neck-and-Neck with Intel PCs.
In another topic you wonder about prices. This is what they are today, similarly configured:

Prices TODAY 16th Feb 2004:
Dual 2GHz PowerPC G5 - $2,999.00 ($2,699 - education)
Dell Precision?Workstation 650 Dual Xeon 3.06GHz 512K cache - $3,896
Dell Precision?Workstation 650 Dual Xeon 2.80GHz 512K cache - $3,412

 

[Eug]

Originally posted by: beachbreeze
Interesting Eug... but do you really think there will be an almost doubling of power usage with the 25% increase in clock speed from G5 2.0GHz to 2.5GHz?

I found that surprising too, but this page (Japanese) does indeed say that the 2.5 will need 52 W, which is more than a doubling of (typical) power. However, we should get more info today, directly from the horses mouth. I suspect the 2.5 currently requires significantly higher voltage than the 2.0, but maybe we'll be pleasantly surprised.

This PCMag benchmark reckons the G5s faster than you do

Well, just about anyone here will tell you not to trust any mag like PCMag to give us reliable benchmarking results.

My 2.8 GHz Xeon estimate for a 2.0 GHz G5 is just my guesstimate based on various different benchmarks I've seen out there. However, in certain benchmarks, the G5 sucks @ss and gets destroyed by a 2.8 GHz Xeon. Conversely, in other benchmarks the G5 is absolutely killer. It really depends. Also it's important to consider if the app is integer heavy or floating point heavy. The G5 is excellent at FP, but merely OK for integer. Compiler choices also matter. ICC for the Xeon is super fast. The XL compilers from IBM for the G5 are also very fast, but they only came out recently for Mac OS X. Unfortunately for the G5, the XL compilers don't auto-vectorize like ICC does for x86. And gcc/g77, which is the default (free) compiler for Mac OS X, is slow.

 

[Acanthus]

Originally posted by: beachbreeze

Originally posted by: MercenaryForHire

Originally posted by: beachbreeze

Originally posted by: Derango

Originally posted by: beachbreeze

Originally posted by: MonkeyDriveExpress

The advanced G5 architecture provides an industry-leading front-side bus dedicated to each CPU as well as up to 8GB DDR SDRAM with EEC protection.

A bit of a whoopsie there...

No, it isn't a mistake. The new xServes have EEC RAM.

No, they have ECC ram. There was a typo in the page when he posted.
See, apple isn't perfect after all

Oh, my mistake. I thought this was a computer tech forum not a secretarial convention.

The point isn't that the typo existed or not - it's that you don't know what ECC RAM is and were merely regurgitating specifications that you don't understand.

- M4H

ECC - Error Correcting Code
Both parity and ECC (Error Correcting Code) are forms of error detection for memory modules. Parity is a simple form of error detection that adds an extra bit for every 8 bits on a memory module. This extra bit records whether there is an even or odd number of 1's registered in the 8 bits. If they don't match, then an error has been detected within the memory. ECC is a more advanced form of error detection that goes beyond the single parity bit and can actually handle error correction.

Careless assumption M4H

haha, gratz on google.