No 3GHz Soon and No G5 Laptop

[AidenShaw]

THERE IS NO VATECH XSERVE CLUSTER

You were wrong - as usual. In less than 2 years, VA Tech has built TWO supercomputers from stock G5s and xServes.

The VA Tech cluster is still offline for the Xserve upgrade. It hasn't completed yet!!! Read the VA Tech page that I posted.

The first cluster got some numbers for the Top500 fall list last year, but it proved to be too unstable for production due to the lack of ECC and was torn down.

AT THIS POINT VA TECH HAS PUT ZERO SUPERCOMPUTERS INTO PRODUCTION. NONE. NADA. ZIP.


Apologies to everyone else for shouting, but jragosta seems to be very hard-of-reading lately.

 

[jiggie2g]

If we are only talking about chips

If we are only talking about chips the Opteron/64 FX-5x series is the better Processor clock 4 clock.

Dual G5's are needed just to keep up with 1 AMD chip and that's a shame , that is what the Mac Zealots don't realize and what the PC guy's have been trying to tell u all along.

a Dual Opteron 2xx Config will completely destroy any G5 setup PERIOD.

P.S. i don't want to hear crap about it not being a fair match cuz the Opteron is a Server/Workstation Chip cuz so is the G5 970FX on XServes and newer IBM Blades.

 

[yuphorix]

Just WTF are you talking about?

How the heck is a RAID system supposed to prevent system memory errors (which is what the ECC would do)? The two concepts are entirely unrelated.

And as far as the old cluster working, they simply had to insert some SOFTWARE error checking since the hardware wasn't there. And it was STILL #3 in the world.

Umm... RAID not only improves performance but it also allows for parity checking and improved hamming distance calculations (although the specifics I'm not sure of)... this allows for reconstruction of errored data that may occur with high speed and high volume writing. I don't think the data is specifically held in the memory and most research data is written and archived immediately onto the harddrive. Thus, the HD performance and reliability is crucial. If you know anything about error control, you'd know that memory error can occur at multiple stages and with such a massive system, you need not only detection but also prevention for each stage.

But oh, sorry... my school is well informed about such technology... that's why we're ranked #4 in engineering...

 

[maverick13]

If we are only talking about chips the Opteron/64 FX-5x series is the better Processor clock 4 clock.

Dual G5's are needed just to keep up with 1 AMD chip and that's a shame , that is what the Mac Zealots don't realize and what the PC guy's have been trying to tell u all along.

a Dual Opteron 2xx Config will completely destroy any G5 setup PERIOD.

P.S. i don't want to hear crap about it not being a fair match cuz the Opteron is a Server/Workstation Chip cuz so is the G5 970FX on XServes and newer IBM Blades.

The Opteron looses to G5 in floating point performance.
And what are you man? Why are you so much trying to conveice yourself PCs are better than Macs? Do you have psychological problems with Macs?

 

[nuckinfutz]

If we are only talking about chips the Opteron/64 FX-5x series is the better Processor clock 4 clock.

Dual G5's are needed just to keep up with 1 AMD chip and that's a shame , that is what the Mac Zealots don't realize and what the PC guy's have been trying to tell u all along.

a Dual Opteron 2xx Config will completely destroy any G5 setup PERIOD.

P.S. i don't want to hear crap about it not being a fair match cuz the Opteron is a Server/Workstation Chip cuz so is the G5 970FX on XServes and newer IBM Blades.

LOL

What happened here
http://www.barefeats.com/g5op.html

Even in a dog of an application like After Effects the Opty is ok
http://www.mediaworkstation.com/articles/viewarticle.jsp?id=25633-1

Opteron loses again
http://www.geocities.com/sw_perf/PSBench.html

Although it does do well in the 3D apps. But then again most of the 3D apps are optimized for windows and better graphics cards.

Here's the honest truth about the Opteron versus the G5. They are very close which means really it's the software optimizations that matter. The ondie controller helps the opteron pull ahead in many cases where latency is a factor. However i'm loving this fanboy reaction I hear where they try to convince me that the opteron is ungodly fast.

For the final laugh you'd have to see the chip prices. I bet a G5 @ 2Ghz is at least %60 of what a 2Ghz opteron costs. I guess Apples crying all the way to the bank huh?

 

[rdowns]

I don't know. Why are you attacking Jobs on the basis of an honest miscalculation?

Attacking Jobs? I don't know where you get that. People were posting about what he said at last year's WWDC, I simply posted the quote. Are you one of those Apple and Jobs can do no wrong types?

 

[BakedBeans]

If we are only talking about chips the Opteron/64 FX-5x series is the better Processor clock 4 clock.

Dual G5's are needed just to keep up with 1 AMD chip and that's a shame , that is what the Mac Zealots don't realize and what the PC guy's have been trying to tell u all along.

a Dual Opteron 2xx Config will completely destroy any G5 setup PERIOD.

P.S. i don't want to hear crap about it not being a fair match cuz the Opteron is a Server/Workstation Chip cuz so is the G5 970FX on XServes and newer IBM Blades.

dont get me wrong the opteron is a very very good chip but as for the 2xx
beaing any g5 setup.....woops a daisy......dual 2.5 would run away with it
in a fair test

 

[jragosta]

Umm... RAID not only improves performance but it also allows for parity checking and improved hamming distance calculations (although the specifics I'm not sure of)... this allows for reconstruction of errored data that may occur with high speed and high volume writing. I don't think the data is specifically held in the memory and most research data is written and archived immediately onto the harddrive. Thus, the HD performance and reliability is crucial. If you know anything about error control, you'd know that memory error can occur at multiple stages and with such a massive system, you need not only detection but also prevention for each stage.

But oh, sorry... my school is well informed about such technology... that's why we're ranked #4 in engineering...

I couldn't care less where your school is rated - you don't know anything about it.

You cited lack of ECC as a critical failing of the VA Tech cluster and said that they needed RAID to fix that particular problem.

While RAID is quite useful, it does nothing to solve the problems caused by lack of ECC - no matter how bright OTHER people at your university are, you don't know what you're talking about.

 

[jragosta]

Attacking Jobs? I don't know where you get that. People were posting about what he said at last year's WWDC, I simply posted the quote. Are you one of those Apple and Jobs can do no wrong types?

Nope. Jobs and Apple have done lots of things wrong.

But you keep insisting that he was lying by saying they were going to be at 3.0 this year.

All I did was point out that it was a reasonable estimate at the time and the whiners who throw their 'Jobs is a liar and can't be trusted because they couldn't deliver a 3.0 system today' tantrums are fools.

 

[jragosta]

The VA Tech cluster is still offline for the Xserve upgrade. It hasn't completed yet!!! Read the VA Tech page that I posted.

The first cluster got some numbers for the Top500 fall list last year, but it proved to be too unstable for production due to the lack of ECC and was torn down.

AT THIS POINT VA TECH HAS PUT ZERO SUPERCOMPUTERS INTO PRODUCTION. NONE. NADA. ZIP.


Apologies to everyone else for shouting, but jragosta seems to be very hard-of-reading lately.

That's true. VA Tech built a supercomputer, tore it down, and has nearly finished a faster one in less time that it took anyone else to order, obtain, install, and deplay a single supercomputer.

What part of that don't you understand?

 

[Bigheadache]

Get real guys, a dual Opteron 250 would be quicker.

The catch is of course...

1. try finding Opteron 250s, I haven't seen any shipping.
2. A dual opteron 250 system is probably going to cost alot more than $2999. Don't forget Opterons use ECC/Registered RAM which costs an ungodly amount. And choose an NForce3 Pro 250gb board with the AMD 8131 PCI-X tunnel, not one of those elcheapo VIA chipset boards

Having said that, a G5 2.5 is still quick, and the IBM roadmap looks good.

The AMD roadmap basically has more of the same, with dual core in 05/06 (?)
IBM has POWER5 core to come, on die memory controller (?), and dual core at some stage.

 

[rdowns]

Nope. Jobs and Apple have done lots of things wrong.

But you keep insisting that he was lying by saying they were going to be at 3.0 this year.

All I did was point out that it was a reasonable estimate at the time and the whiners who throw their 'Jobs is a liar and can't be trusted because they couldn't deliver a 3.0 system today' tantrums are fools.

Take this for what it's worth. Lighten up dude.

You come into this forum and post prolifically with a chip on your shoulder. I never said Jobs lied, go back and read the threads. I simply posted his quote.

I'd be careful about throwing the term fool around.

 

[Bigheadache]

Umm... RAID not only improves performance but it also allows for parity checking and improved hamming distance calculations (although the specifics I'm not sure of)... this allows for reconstruction of errored data that may occur with high speed and high volume writing. I don't think the data is specifically held in the memory and most research data is written and archived immediately onto the harddrive. Thus, the HD performance and reliability is crucial. If you know anything about error control, you'd know that memory error can occur at multiple stages and with such a massive system, you need not only detection but also prevention for each stage.

But oh, sorry... my school is well informed about such technology... that's why we're ranked #4 in engineering...

You don't need RAID to "verify" the integrity of data being saved onto a hard disk. I assume you use a relatively modern hard disk.. because if it is LBA, "write verify" and "read verify" are already built in. RAID is there principally to provide data redundancy and speed. The parity allows you to rebuild an array should a hard disk fail.

Just think, your school could have been number #3 without...

 

[yuphorix]

I couldn't care less where your school is rated - you don't know anything about it.

You cited lack of ECC as a critical failing of the VA Tech cluster and said that they needed RAID to fix that particular problem.

While RAID is quite useful, it does nothing to solve the problems caused by lack of ECC - no matter how bright OTHER people at your university are, you don't know what you're talking about.

I never said that RAID would fix the problem... I just asked whether or not they set it up. I think you guys are assuming too much in people's statement by categorizing people as anti-mac or mac-fanatic.

 

[toontra]

Steve has messed up big time here. Didn't his parents ever tell him you shouldn't make promises you can't keep? Apple will take heavy losses over this; people like me who need all the processing power they can get who won't buy until the 3G target is achieved, and many G5 owners who won't now upgrade to rev.2 machines.

Bad for reputation and disasterous for sales - a PR double whammy!

 

[Uragon]

Steve has messed up big time here. Didn't his parents ever tell him you shouldn't make promises you can't keep? Apple will take heavy losses over this; people like me who need all the processing power they can get who won't buy until the 3G target is achieved, and many G5 owners who won't now upgrade to rev.2 machines.

Bad for reputation and disasterous for sales - a PR double whammy!

Not necessarily. Some analyst already pointed out that this is not Apple's fault and even commended IBM for cranking up the chip with 25% increase compared to, lets say, Intel and AMD.

To be honest, I believe it was not possible for IBM to come up with 50% increase, but did hope and hope and hope that Apple will come up with a dual 3.0GHz and that hope were becoming reality specially when the rumors were sprouting out of nowhere just prior to the announcement.

I frustrated myself and need not blame anyone, but it's always nice to read the rumors though...

 

[yuphorix]

You don't need RAID to "verify" the integrity of data being saved onto a hard disk. I assume you use a relatively modern hard disk.. because if it is LBA, "write verify" and "read verify" are already built in. RAID is there principally to provide data redundancy and speed. The parity allows you to rebuild an array should a hard disk fail.

Just think, your school could have been number #3 without...

Read/Write verify cannot prevent against data corruption and even errors that can occur with the disk head effected non-written areas. A properly setup RAID not only provides parity to rebuild an array, it can provide a more accurate hamming distance calculation to fix single bit errors. I'm assuming that a supercomputer of this caliber is dealing with more than a couple of HD's. In most cases, a standard HD can prevent against just errors, but we aren't talking about a regular computer or even a medium sized cluster. Google replaces tens of HD's a day just to keep up with pure disk failure and corruption, I'm pretty sure it's no different for Virginia Tech. In any case, ECC is more necessary than RAID, but with all major server/clusters, both should be implimented for reliability. And remember, VirginiaTech didn't forget RAID, they forgot ECC.

Oh, and my school would be #3 if we didn't have to rely heavily on corporate funding and donations (State doesn't give us ****).

 

[iChan]

Excellent. So I can feel confident/safe about buying a g4 pb in the next 2 weeks.

hot dang, i think buying a g4 PB is more than safe for a long while, even when a g5PB does come out, you'd still ahve one kick-ass machine. esp is you get 4x superdrive, 1.5Ghz, 128MB VRAM. woah! what a machine that would be.

 

[jiggie2g]

Get real guys, a dual Opteron 250 would be quicker.

The catch is of course...

1. try finding Opteron 250s, I haven't seen any shipping.
2. A dual opteron 250 system is probably going to cost alot more than $2999. Don't forget Opterons use ECC/Registered RAM which costs an ungodly amount. And choose an NForce3 Pro 250gb board with the AMD 8131 PCI-X tunnel, not one of those elcheapo VIA chipset boards

Having said that, a G5 2.5 is still quick, and the IBM roadmap looks good.

The AMD roadmap basically has more of the same, with dual core in 05/06 (?)
IBM has POWER5 core to come, on die memory controller (?), and dual core at some stage.

You can get any Opteron Chip at NewEgg.com for a resonable price and set-up a DP 250 system running at 2.4 ghz for about the same price asthe Dual 2.5 G5's .

 

[nuckinfutz]

A PC at the same price of a Powermac but only slightly faster is a waste. People keep forgetting the appeal of a Macintosh is not the hardware for most. It's the software.

So unless a PC is demolishing a Mac we don't care. We don't want to run windows and the only way many Mac users will run windows is if it corresponds to a large increase in speed to overcome the slower productive environment of Windows.

I thought that was clear with peecee sheeple I guess not.

 

[dr.Zoidberg]

wont buy?

Steve has messed up big time here. Didn't his parents ever tell him you shouldn't make promises you can't keep? Apple will take heavy losses over this; people like me who need all the processing power they can get who won't buy until the 3G target is achieved, and many G5 owners who won't now upgrade to rev.2 machines.

Bad for reputation and disasterous for sales - a PR double whammy!

If you wont buy till Apple hits 3Ghz you wont be getting any work done for awhile,why not get as much power now, the performance increase from a 2.0 to 2.5 has got to be MORE power than before....sure they HAVENT gotten to 3 YET.....if of course you "need all the processing power [you] can get"......i think you are just whining for the sake of whining here...

 

[johnnowak]

How can someone say the G5 is faster than the Opteron, and vica versa? It depends on the applications being used.

 

[mewzrh33]

Processors

This may be old news but I just received confirmation from an Apple tech that only the 2.5GHz proc is a 970fx (90nm), the 1.8 and 2.0 are both 970s (130nm). While I don't have a problem with the upgrades, I don't appreciate the misleading marketing at all. It seems like there is a surplus of 130nm chips and/or Apple doesn't want to take chips away from the XServe 2.0....

 

[dontmatter]

Wow, I just realized how pathetic these updates really are. Even apple thinks they're pathetic-they don't even get the main spot on the web-page, that's left to airport express. They had to make a new product announcment before the PM updates so they could hide those shameful updates behind something new, but not totally look like they knew the updates sucked, by not putting them up at all.

Sad.

God I hope WWDC comes up with something huge.


Edit: that was weird-I returned to find that the G5's were in the main spot, and was going to hit back until I got to the origional and take a screen shot. But then, I checked again in another window, and it was back to airtunes. Hmmm.

 

[nagromme]

VT's Big Mac / System X is a huge success

VT didn't "forget" ECC. The guy behind it discussed that issue early on, including the fact that certain calculations would need to be run twice and that the speed of Big Mac was STILL better per dollar. One could argue that the Top500 rules should demand ECC, but they don't. And that Top500 deadline was all about funding so don't blame VT for doing what they had to to build a Top500 machine in time for the deadline. Apple and the G5 made it possible. The considered and rejected Dell and many other options that they'd been focussing on long before they heard of the G5. Yes, it's a shame that NSF and Top500 deadlines resulted in a system being built twice, but VT played their cards wisely and it's paying off.

Now they're doing it over with ECC, lower cooling requirements and less space. Doing it "right" instead of "fast." They HAD to do it fast for the deadline--Xserve G5s simply did not exist--and it's impressive that they were able to pull it off. And now it's great for them AND for Apple that Big Mac is getting the big Xserve upgrade. The "down" time is scheduling, timing, and funding reality, with an excellent outcome. It's not some failing. It's a longer time because of IBM's 90 nm problems. And in the end, VT is getting an even BETTER supercomputer for their money than the first version was. They'll have a much better research system than anything they could have built for the money last year and not rebuilt. At a cost nothing else has yet touched, and with space to grow even bigger and faster.

And with a kit they're providing to other schools to do the same...