Welcome to the MacNN Forums.

Peter753 · Apr 19, 2004, 12:46 AM

I dunno if someone has pointed this out yet but http://www.macmall.com/macmall/families/powermac_g5/

Kinda interesting. You save 200 bucks and get an extra 512 mb of ram.

oh yeah AND A PEICE OF SUPERCOMPUTER HISTORY!

lol ok ill curb my enthusiasm

Lateralus · Apr 19, 2004, 02:13 AM

Yeah, its been pointed out before. But it has been a while.

Not much has changed since it was last pointed out. MacMall is still charging too much.

Mike656 · Apr 19, 2004, 11:17 AM

Why is VT selling systems from their cluster?

-Q- · Apr 19, 2004, 11:31 AM

Because they replaced them with G5 xserves.

JohnM15141 · Apr 19, 2004, 11:49 AM

Originally posted by -Q-:
Because they replaced them with G5 xserves.

How will the G5 xservers affect the performance of the cluster? Will it see a performance increase or just decrease the amount of space being used?

southtdi · Apr 19, 2004, 11:53 AM

I would think that with the space they are saving they will be able to add more nodes and make the cluster more powerful without adding more floor space.

Person Man · Apr 19, 2004, 12:19 PM

Originally posted by Peter753:
I dunno if someone has pointed this out yet but http://www.macmall.com/macmall/families/powermac_g5/

Kinda interesting. You save 200 bucks and get an extra 512 mb of ram.

oh yeah AND A PEICE OF SUPERCOMPUTER HISTORY!

lol ok ill curb my enthusiasm

You can buy a refurbished Dual 2.0 GHz Power Mac G5 for $2399 from Apple (whenever they're available). Why would you want to pay $2799 for one of those?

Ok, so they come with a full GB of RAM instead of 512 MB (but that still isn't worth the $400 difference in price. Those machines don't even have a modem in them (which you *do* get with a refurb bought directly from Apple).

stuffedmonkey · Apr 19, 2004, 02:32 PM

Any other VT grads here on Macnn?

-class of 99 here

-Q- · Apr 19, 2004, 02:42 PM

Originally posted by southtdi:
I would think that with the space they are saving they will be able to add more nodes and make the cluster more powerful without adding more floor space.

I think that's the primary reason I read. Plus, they'll save some money on heating and cooling b/c they don't have all that space taken up by the towers. I don't think they did it to realize any significant performance improvements, just some space and bottom-line cost savings.

Eug Wanker · Apr 19, 2004, 06:13 PM

Space etc. of course, but I'd say more importantly:

1) ECC memory: Having 1 memory error screw up your 3-day long run over 1100 machines is bound to make people unhappy.

2) Hardware monitoring. The G5 Power Macs don't have any real hardware monitoring, which is a major pain when you're trying to troubleshoot 1100 machines.

BTW, those G5s have been for sale for months. I'd rather get a cheaper refurb myself from the Apple Store though if I were in the market.

Spliffdaddy · 08:23 PM

I don't think ECC memory actually corrects the error - it merely acknowledges there was an error. The action it takes as a result of the error could be any number of things. It could simply light an indicator LED on the motherboard, like my old Xeon server.

While I'm not 100% certain, I don't think ECC memory actively corrects errors.

Maybe I'm confusing parity error checking and ECC. hell, just ignore me.

Eug Wanker · Apr 19, 2004, 09:13 PM

Originally posted by Spliffdaddy:
I don't think ECC memory actually corrects the error - it merely acknowledges there was an error. The action it takes as a result of the error could be any number of things. It could simply light an indicator LED on the motherboard, like my old Xeon server.

While I'm not 100% certain, I don't think ECC memory actively corrects errors.

Maybe I'm confusing parity error checking and ECC. hell, just ignore me.

Error Correcting Code memory

Single-bit ECC errors are corrected by the ECC circuitry and cause no loss of data or system availability. A summation of the state of all DIMMs is shown in the main status line for a given server.

Server Monitor will also detect multiple-bit errors, but a multiple-bit error will cause a Mac OS X kernel panic. Information will be stored in the Server Monitor logs after the server is restarted. If this happens, you should replace the failing DIMM.

Mike656 · Apr 19, 2004, 10:10 PM

Did they take all the orignal machines off the cluster or are they just replacing them a few (maybe 50-100) at a time? Seems like a big hassle (and cost) to me to build a whole new system to replace one installed less than a year ago, even with the advantages of the xserve G5.

Lateralus · Apr 19, 2004, 10:16 PM

AFAIK, they did it in one big sweep, replaced them all at once.

southtdi · Apr 20, 2004, 10:37 AM

My guess is they were, again, a factor in the shipping delays of the dual G5 xserves. Same thing happend with the shipping of the PowerMac G5's when they held up their shipping out.

I can understand why Apple does this since in the long run it is great enterpise publicity for them. Not to mention anyone else considering a G5 cluster.

As for VT, I wouldn't be surpised that they eventually double the nodes. Apple could very well just donate them too. Gives Apple the prestige by moving the VT supercomputer up in the ranking, as well as a good tax break. A faster cluster gives VT even more prestige on thier end, which importantly could mean more grant money. Especially if the govermant starts renting space.