The company lands a big customer for its Power Mac G5: Virginia Tech, which will use 1,100 G5s as part of a cluster designed to make the list of the world’s largest supercomputers. In the meantime, Apple delays the shipping of its dual-processor Power Mac G5 to individual consumers, opting instead to send the first of the fast desktops to schools, Europe and selected customers.
woohoo im first!
Anyway I wonder why apple is continuing to diss the individual consumer by delaying shipments. People paid 2-3 grand for these machines come on.
Guys,
These G5s dont work the way they are supposed to……find a place to send them where they are least likely to complain untill we can get the bugs out of them.
I wonder exactly how much real super computing Virginia Tech is going to get out of that system since the G5 *do not* support ECC RAM?
Those bit errors are going to kill them :/
“Anyway I wonder why apple is continuing to diss the individual consumer by delaying shipments.”
Apple isn’t delaying shipments, they’re distributing existing merchandise stock to the individuals that need it most.
Makes sense to me.
“People paid 2-3 grand for these machines come on.”
So did the education institutions…
“These G5s dont work the way they are supposed to……find a place to send them where they are least likely to complain untill we can get the bugs out of them.”
I’ve been using a G5 the day they were first distributed. I have yet to find any bugs.
“the G5 *do not* support ECC RAM?”
I googled for any bit of information about this issue and couldn’t find anything. Where are you getting this?
Nevermind… i found it:
http://developer.apple.com/documentation/Hardware/Developer_Notes/M…
http://developer.apple.com/documentation/Hardware/Developer_Notes/M…
“Important
DIMMs with any of the following features are not supported in the Power Mac G5 computer: registers or buffers, PLLs, ECC, parity, or EDO RAM.”
This was from a search of “g5 ecc ram”
If you want things like ECC memory, built in Fibre Channel, etc., you really need to look at a higher-end workstation like a Sun Blade 2000. Sure, they are expensive as hell, but these components are not cheap. That’s why Sun charges a premium for them.
One thing you get with Sun that you’ll never see in an Apple: ECC across all data paths. That’s why they’re so expensive. People that do medical imaging and chip fabrication/design on these boxes want to know that a flipped bit in memory won’t translate into data corruption on the display, or that their data will be preserved intact on the hard disk after travelling from memory, across the PCI bus, through the Fibre channel adapter, and out to disk. The Sun workstation’s ECC insures that.
However, I must admit that I’ve never had a problem running a non-ECC machine 24×7 – so maybe it’s just the equivalent of an IT soother.
“the G5 *do not* support ECC RAM?”
I googled for any bit of information about this issue and couldn’t find anything. Where are you getting this?
As far as I know there never has been, or never in recent history has there been, a Mac that support ECC RAM. There hasn’t been a super computer cluster, either.
My question is whether these are going to be the same profile as the normal G5s. Wonder when there is going to be a G5 xServe. And of course I wonder when there is going to be an Apple video workstation marketed.
That is a lot of real-estate for a supercomputer. Just image 1100 24″x30″x8″ boxes compared to 1100 1U boxes.
… shut the mouths of all the people complaining about Apple releasing things like the ITMS in the US first and eventually Europe, Canada, and etc.
Wonder when there is going to be a G5 xServe.”
I’d wager that it will be when the 970 goes down to 90-nanometer manufacturing process. When this happens, the chip wilol be much smaller and cooler and will allow for much higher GHz. it will be at that time that you’ll see a G5 going in a power book.
Steve J. said that 3GHz will be had by year’s end. Coincidentally, IBM’s 970 roadmap shows 90 N.M. occuring shortly before the new year. (Apparently they already have chips in testing now). Some have speculated that this new chip will be called the 980… but who knows…
“Guys,
These G5s dont work the way they are supposed to……find a place to send them where they are least likely to complain untill we can get the bugs out of them.”
How do you know they don’t,,do you have pre-release model for testing? Do you EVEN own a Mac?
“I’d wager that it will be when the 970 goes down to 90-nanometer manufacturing process. When this happens, the chip wilol be much smaller and cooler and will allow for much higher GHz. it will be at that time that you’ll see a G5 going in a power book.”
I think that you need to rethink that statement. smaller transistors packed closer togeth generate *MORE* heat, not less. Something like a synthetic diamond substrate instead of silicon combined with say water cooling would be needed for a chip that runs slightly hotter than these upcoming puppies…
Cool. Its nice to see Apple get into markets beyond the creative fields. It shows that the G5 has the horsepower professional are looking for. VT could have chosen from a lot of other manufacturers so there must have been a compelling reason to go with G5.
You have no knowledge about chip design.
90nm process means a smaller junction and therefore you need
less voltage to power the chip and so the thermal power leakage is
lower. further on, the shorter signal lenth enables higher clock speeds without the need to raise the core vcc.
That’s why the 90nm process produces chips that consume less power than the 130nm chips at equal frequencies.
Owie! I think I may have just been struck with a clue-bat! I’ll read more on this, as it’s not an area with which I am intimately familliar. Anyone else have anything helpful to add? Always open to filling gaps in my knowledge…
Can’t get the Europe reference. Shipping of 1.6 and 1.8 is 2 weeks behind the US confirmed from Apple.
Lack of ECC could be a really serious issue depending on how the machine is used. In a machine with one or two processors and a few gigs of RAM, no problem.
However, on a machine with 1100 processors and potentially over a terabyte of RAM, the likelyhood of something going wrong gets rather significant. On the other hand, if these machines are truly running as a cluster rather than one large distributed machine, it may not be a big deal.
>>People paid 2-3 grand for these machines come on.
>So did the education institutions…
Nah, schools get a little break. IIRC, the prices my school was quoted were from about $1500 to $2500.
In any event, who would you ship to first: John Doe, who orders 1 computer, or a school that orders 500?
The reason I read either on here of slashdot was that schools needed to get the machines installed before kids returned from holiday.
You know what I think when I hear of a 1100 G5 cluster? What a waste of 1100 ATI 9600s. Sure the G5s might make a great cluster but please throw in a cheap video card, it’s not like you’ll be using it. I know I could use one or two 9600’s.
It’s ironic that the Powermac G5 is so fast that people must wait for them.
It’s ironic that the Powermac G5 is so fast that people must wait for them.
somewhat like watching a fan or a bicycle wheel spin at high RPMs. When looking at it – it actually appears to be moving very slow
It’s ironic that the Powermac G5 is so fast that people must wait for them.
somewhat like watching a fan or a bicycle wheel spin at high RPMs. When looking at it – it actually appears to be moving very slow … which I hope is not the case for OS X
ECC RAM is not just a mere IT soother. For mission critical applications it is a must. As a single bit error could literally mean life or death for someone. This is why all high end workstations and servers support ECC RAM.
If VT is doing high physics sims on that cluster, I for one would not like to be around if the results of those sims are ok’ed for real and practical use.
They could wind up teleporting half my body into the earth’s core or forming a black hole directly over New York city with the erroneous results a bit error would introduce into the calculations…
“90nm process means a smaller junction and therefore you need
less voltage to power the chip and so the thermal power leakage is
lower.”
Not 100% true as with shrinking process sizes one also get more constant leakage. It will become very problematic at process sizes below 65nm but is visible even at todays 130nm sizes…
“further on, the shorter signal lenth enables higher clock speeds without the need to raise the core vcc.”
Wire length have nothing to do with clock frequencys (as a properly superpipelined design allows to compensate for the propagation delays in the wires) and raising the core voltage increases transistor speed not wire speed.
I can accept the lack of ECC if there are algorithms such that not too much computational power is taken up to gain reliability. Or if the supercomputer’s used in apps where a bad bit here ‘n there isn’t fatal, like perhaps with some cellular automata.
If the cost of reliability doesn’t outweigh the costs saved by using G5’s, great. Otherwise, they did an exceptionally bad job of acquiring the machine since this is basic 101 stuff. I have a hard time believing they made such a mistake; after all, I’m sure supercomputers have been built without great hardware error-detection/correction.
They could wind up teleporting half my body into the earth’s core or forming a black hole directly over New York city with the erroneous results a bit error would introduce into the calculations…
If a black hole does form over New York City, there isn’t much anyone on earth – or outside it, can do (a blackhole is normally many many times the size of earth, would swallow up earth, all the planets in the system as well as the sun with its ernomous gravity).
And for teleporting half a body to the earth’s core… oh never mind, I get your point. VT probably could loose millions of dollars with the lack of ECC, but not to the extent of causing the extinction of mankind.
Megol!!!
What the hell are you talking about.
The command you made is so stupid that I wont’ even bother to correct you. It’s complete nonsense. Agrrrrr….
Ralf is absolutely RIGHT.
Do some more research about the chip design before you post anythig…
Ahhhh crackerjack, the G-4 is sitting right next to my trusty old Thinkpad……and it belongs to the Lady of the House. Cost her 5 grand Australian in January last year. The Apple service people used it more often than she did last year. Fascinating piece of technology…..comes with a ONE BUTTON mouse…..very advanced dont you think ??….built-in 56K modem was bad out of the box !!!
There is an old saying in computing…..garbage in = garbage out, but now Apple is elevating computing to a new level with their G5s assembled into a make-beleive “super computer” where the only output will be garbage. Hahahahahahaha….
A couple of months back it was “Adobe to Jobs…..Hit the road Mac”.
Now we find out today that the wizards at Virgina Tech do-not even do basic research on the equipment they are looking to buy for a major project……interesting computational problem for you Macaddicts…1,100 units X$3000 = ???? Its just to hard for an IBMer to do while my sides are split open from laughing.
Yes, you are all right… A school like Virginia Tech (notice the word tech) just jumped blindly into this project so they could be in the top 3 ranking for the fastest computer in the world. Yes, they did absolutely no research on any of the components and I am sure the school allowed someone who is no doubt a Mac fanatic to spend SIGNIFICANT amounts of $$$ on a zealot’s desire to spread the gospel of Steve Jobs. Some of you guys are just plain IDIOTS. Get the whole facts first before you go trolling and trying to act like you know everything when in reality you take a small bit of new info and mix it with what little knowledge you picked up from some linux forum and pose yourself as some almight hardware guru. Some of you should be petrified and turned into stone because your heads like fvckin rocks!
Oh yeah…did I tell you I am a Mac zealot at VT who has recently purchased “some” G5? :p
[i]Fascinating piece of technology…..comes with a ONE BUTTON mouse…..very advanced dont you think ??[i]
Will the one button mouse trolls ever give up. It’s not as if macs only work with a one button mouse. If that second button means so much to you, buy one! They really aren’t that expensive. Me, I’m happy to use the ctrl button on my keyboard.
As someone pointed out on Slashdot, with a cluster this large, chances are much greater that a flipped bit will occur due to a hard drive flaking out or problems with the networking equipment, rather than RAM. This would lead to two conclusions:
ECC is much more important when all critical data is on a single machine.
Since the people at VT are almost certainly not idiots, they’ll incorporate error correction routines in the software.
Cheers,
JT
How do you correct errors in software if you can’t detect them?
Even if there is a protocol layer over the data that they are trying to processes that can detect single bit errors, the overhead of doing CRCs and checksums would outweigh many of the performance benefits of the hardware.
So anyway you slice it not having ECC will hurt VT’s super cluster.
How do you correct errors in software if you can’t detect them?
Redundancy. As an extreme example, you could have three computers computing the same thing, voting on the results of each computation, with a 2/3 majority winning.
An error happens every so often, for a tiny interval of time. (Assuming it’s not some large system error.) The chances that two machines will generate the same error, at the exact same moment, are tiny. As long as there isn’t some environmental badness making them agree.
I think the interesting news is will MacOSX be used in the cluster or Linux? Will Apple have special clustering software?
“1,100 units X$3000 = ????”
Apple offers higher-ed discounts and this purchase is probably heavily discounted because of the volume so I don’t know where you got your 3K figure unless you know something about this deal that everyone else doesn’t.
looks like they will be using MacOSX for the clustering!