Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Since they've been planning for a while, I really wouldn't be surprised if these systems were somehow customized to not have the video card and super drive and such. I mean, with an order that big, I'm sure they wouldn't mind making some alterations. I'm not saying they are doing that, I'm just saying I wouldn't be surprised if they did.
 
The computers are all base-models, except for 4GB of RAM in each box.

By base model do you mean 1.6 Ghz, or just not custom configured? I'd imagine 50 Dual 2.0's would outperform 100 1.6's.
 
me thinks this is added value

Originally posted by mainstreetmark
How much you think 1100 G5's will go for on eBay in 5 years?

One of the tenants of auto leasing, I suspect, is that you can get good stuff for reasonable lease rates because it is still very highly valued at the end of the lease.

A year from now, VT could off these to students and move up to the 980 based G-5's with nary a pause, and still look good financially. Try that with a commodity PC, and it won't be pretty.
 
Re: me thinks this is added value

Originally posted by TMay
One of the tenants of auto leasing, I suspect, is that you can get good stuff for reasonable lease rates because it is still very highly valued at the end of the lease.

A year from now, VT could off these to students and move up to the 980 based G-5's with nary a pause, and still look good financially. Try that with a commodity PC, and it won't be pretty.

I have been reading this thread with above average amusement.

I am not sure which post is best, porn, solitare, or 10.3 upgrade.

My opinion is that Macs hold their value so shockingly well to begin with, a dual G5 1000Mhz bus computer can only be better.

But what's more important, is about the first working month of this cluster will fully pay for it, and everythng after that will be gravy.

The fact it is so trivial for the first big Mac cluster to get into the top 5 or 10 or 20 or whatever is impressive. I wonder what will happen when a cluster that has NOT made the news on the rumours forums kicks its a$$?

Rocketman
 
Re: me thinks this is added value

Originally posted by TMay
A year from now, VT could off these to students and move up to the 980 based G-5's with nary a pause, and still look good financially. Try that with a commodity PC, and it won't be pretty.

<OT Rant>: There is no 980 as far as anyone knows....the 980 is something that was made up on the rumor sites....it is based in nothing.....
For the last 3 years everyone has been pissed off at apple/moto for not releasing the G5.... When it finally shows up, it isn't moto at all but IBM. we haven't seen the 15" laptop yet..something that everyone knows should be out. Why keep mentioning the 980, everyone is just going to get pissed off when it doesn't materialize. Even if Steve said that they will have 3Ghz chips by next year doesn't, by any stretch of the imagination mean it has to be the "980".....</OT Rant>

um....sorry about that......

Originally posted by Rocketman
But what's more important, is about the first working month of this cluster will fully pay for it, and everythng after that will be gravy.

i doubt that it will pay for itself in the first month....but it can within a few years. I am not sure if they can (being a publicly funded university) charge others to use the system. They should make savings by not having to run jobs elsewhere, and by not having to tie up lesser resources for long periods of time.
(and, i think the porn reference was the best one.....or maybe a quake server)

neilt
 
Re: Re: me thinks this is added value

Originally posted by neilt
i doubt that it will pay for itself in the first month....but it can within a few years. I am not sure if they can (being a publicly funded university) charge others to use the system. They should make savings by not having to run jobs elsewhere, and by not having to tie up lesser resources for long periods of time.
neilt

They can charge others for use, but the amount charged will be higher than what is charged to VT users because they have to factor in overhead costs like secretarial costs and building maintenance when people outside are charged.

Does anyone know who is actually paying for it? VT? NSF or other federal grant? A mixture? Many big installations of science equipment are a pure win situation for a university as they contribute very little money. So comparing it to a loss of scholarships can be misleading.
 
Re: Re: Hmmm...

Originally posted by holmesf
Um, if the work were distributed with 100% efficiency it would take around 10 seconds, but this is not possible.

Huh? :confused: Do the math...

8 * 3600 / 1100 / 5 / 2 = 2,62

So it's 2,62 seconds... and that is already pretty conservative, because it assumes that the G5 scales linear compared to the G4 (cycles per MHz).

Nice! :D

groovebuster
 
1,100 computers at roughly $2,000 or $2,500 apiece is less than $3M. That seems like small money in the research field. One 200kV TEM+EDS+EELS with site preparation costs that much. A synchrotron would probably be much more. The point is that this is not much money in the R&D field. This is especially small when one considers the return they expect. Having an extremely fast computer helps put them towards the front of the pack as a top university in physics, chemistry, biology, geology, meteorology and engineering.

Have you any idea what universities have spent to build research nuclear reactors, linear accelerators, large telescopes and the like? Even in the area of computation, expenses in the past would surprise you. I don't have the cost but I remember when the physics department of my university bought one megabyte of memory (about 1975). I'll bet it was several million dollars in today's dollars.

Practically everywhere you turn numerical solutions are displacing exact solutions to the problems facing scientists. Even when there are exact solutions scientists still have to deal with the giant data sets generated by todays automated experiments.
 
Originally posted by mymemory
Play solitarie.

I think is stupid to have a closter made of G5, at list they could wait for the rack version, can you imagine how much room that is going to take? 1000 superdrives they won't use, 1000 of VRam cards they won't use. All they need is one PCI slots, Ram and the procesors.

Do you know how many scolarships they could pay for that?

What a huge waist of money and resources.

Yup. Not waiting for the Xserve G5 just to make it into the top five this year is indeed strange. Being listed as one of the fastest computers in the world is no value whatsoever for research. I guess this is just a case of 'Use your budget till end of the fiscal year or return it'.
 
Not going to wait for xserve

Yup. Not waiting for the Xserve G5 just to make it into the top five this year is indeed strange. Being listed as one of the fastest computers in the world is no value whatsoever for research. I guess this is just a case of 'Use your budget till end of the fiscal year or return i

The choice was not Xserve versus G5, it was G5 versus Dell or HP. If they had not purchased the Apples, they would have gone with the other vendors. Apple is extremely lucky to get in on this deal at all.

Rather than "rushing" into a network of G5s, I suspect the opposite is true, with VaTech slowing down the process to accommodate Apple.
 
Originally posted by tduality
Yup. Not waiting for the Xserve G5 just to make it into the top five this year is indeed strange. Being listed as one of the fastest computers in the world is no value whatsoever for research. I guess this is just a case of 'Use your budget till end of the fiscal year or return it'.

The school year is starting - there is no G5 XServe. How much of the school year would you have them go without their supercomputer?

Being listed in the top 5 is no value for research. Having a top 5 supercomputer clearly is of tremendous value for research. Also, being listed in the top 5 has clear value for recruiting.
 
Originally posted by tduality
Yup. Not waiting for the Xserve G5 just to make it into the top five this year is indeed strange. Being listed as one of the fastest computers in the world is no value whatsoever for research. I guess this is just a case of 'Use your budget till end of the fiscal year or return it'.

I doubt we'll see a G5 Xserve any time soon. There's no reason to put a G5 in a 1U unit since the current systems are fast enough to do the job that 1U are designed to do. Consider that there are still Dell and Intel servers out there with P3s in them.

Also at a recent Apple Xserve meeting for the public, a gentlemen from Disney said that the limiting factor on the Xserve is not the processor, it's the network bandwidth into/out of the system.

theFly
www.flyonthemac.com
Rumors You Can Bet On
 
Originally posted by theFly
that there are still Dell and Intel servers out there with P3s in them.


And consider that there are also Intel 1U servers with dual 3.06GHz Xeons, as well as dual 64-bit Opteron 1U systems. I recently got sixteen of the HP ProLiant DL360 g3 systems - dual 3.06GHz, 4 GiB of RAM (supports 8 GiB), dual GigE NICs, hardware RAID, ... (http://h18004.www1.hp.com/products/servers/proliantdl360/index.html)

There's plenty of reason for a G5 - some people want fast 1U systems that hold a lot of memory!
 
I've got a friend down at Tech who has agreed to go to the information session and let me know of anything she finds that's interesting. I'll post here tomorrow once she gives me info.

--Cless
 
Originally posted by MacRETARD
My first reaction was yeah its a waste too, why not wait for the G5 xserve? But if they want them now for a new project for a new semester then they couldnt wait.

The money may come from a government grant, perhaps from the DOE or NSF. If that is the case, it is probably fiscal year 2003 money which technically expires at the end of September.
 
Very Inexpensive by Top 10 standards

The "all in" price of this machine is more like $5-6 million (on the order of $3 million for the Macs [have to figure in Ed discount and the additional RAM on each machine] and the rest for the interconnect hardware, routers, cooling equipment, racks, etc.). (Note: the software costs are never included in this "all in" price since the software is almost always a custom job.) This is very inexpensive for a Top 10 machine. The Top 10 machines are typically used only by major national labs or the intelligence community, and they usually have a price tag of over $20 million each, "all in". The currenly being designed/built future #1 (ASCI Purple) has a projected "all in" price tag of well over $75 million with some predicting the "all in" price to be closer to $100 million once it's up and running -- and it is supposed to be built around the POWER5 processor. (IBM won a bid for both ASCI Purple and Blue Gene/L for about $260 million, and Blue Gene/L is the follow-on/more expensive of the two.) To give you a further idea of the typical cost of these machines, the current budget for ASCI purple is $280 million including purchace, installation and maintenance for five years (but not including custom application software as mentioned above). If the Mac cluster even costs an unlikely $2 million a year, the $16 million or less five year cost will be less than six percent of the cost of ASCI Purple.

If indeed the new Mac G5 based system comes in at #3 then when Purple is operational it will drop to #4. Lets see... #1 at $100 million and #4 at $6 million. That's EXTREMELY cost effective. If this holds true, I would suspect many other major universities will be doing some version of this same thing.

In the late 70s up through late 80s many major corporations (including Apple) bought supers to do design simulations and heavy number crunching analyses. By the early 90s the cost of a high end machine became prohibitive for all but a few major corporations. This new capability (if the software is not TOO onerous) may allow many more major corporations back into the super computer family.

This is an interesting concept: Apple has not been able to consistently break into the larger enterprise desktop market. Now Apple may have a shot at supplying machines for the super market. This is a very limited market, and Apple will not be able to make a lot of money on it. However, it will definitely change the perception in the industry of what Apple machines are capable of doing.

IF (huge IF) the next generation box ("980"/"G6" or whatever it really is) can be a plug in replacement for the G5 box this will be a very, very interesting evolution to watch. This may be possible with the Mac G5 having the 133 MHz PCI-X slot in the current machines. Inter-processor bandwidth is almost always the biggest constraint on these machines. Upgrading the Top 10 machines has almost always been a complete replacement kind of job. (Typically the machines like ASCI Red, ASCI White, ASCI Blue have not had huge upgrades -- small ones to be sure, but not huge ones. They are just replaced by a new generation.) IF (again, huge IF) the follow-on to the 970 based boxes [G5+?, G6?] can be box for box replacement for the current G5 boxes and get even a factor of 2 speed improvement then this is a significant departure in the way Top 10 supers are done.
 
Re: You sure this isn't another CNET hoax :)

Originally posted by sososowhat
I enjoyed the last one. You just don't know who to trust these days!
What was the last one?
 
Re: Re: Re: Hmmm...

Originally posted by groovebuster
Huh? :confused: Do the math...

8 * 3600 / 1100 / 5 / 2 = 2,62

So it's 2,62 seconds... and that is already pretty conservative, because it assumes that the G5 scales linear compared to the G4 (cycles per MHz).

Nice! :D

groovebuster

u have a photo of a gibson head and you can do math!!!???
I knew i should've never go fender strat, hehhehehhehehe
 
Originally posted by AidenShaw
And consider that there are also Intel 1U servers with dual 3.06GHz Xeons, as well as dual 64-bit Opteron 1U systems. I recently got sixteen of the HP ProLiant DL360 g3 systems - dual 3.06GHz, 4 GiB of RAM (supports 8 GiB), dual GigE NICs, hardware RAID, ... (http://h18004.www1.hp.com/products/servers/proliantdl360/index.html)

There's plenty of reason for a G5 - some people want fast 1U systems that hold a lot of memory!

Just a side note for others, the configuration AidenShaw is talking about would run about $7000 per machine without OS. vs. a PowerMac G5 it would have a faster sustained disk throughput (15k rpm vs 7.2k (though you can buy 10k rpm SATA for around $200)) though smaller hard drive, less memory and front side bandwidth (1Ghz PTP vs 533Mhz (shared?)), and worse graphics (not that this matters at all) + the usual dual 3.06 P4 Xeon vs. 2x2 Mac comparison. Though the RAM is upgradeable to 8 GBs, that is with 2GB DIMM boards. I would hazard a guess, with those boards, the Mac would be upgradeable to 16GB (or upgradeable to 4 or 8GB at a fraction of the price). The cost of the OS, if it isn't Linux, would also be significant.

You could buy a base system from HP (or IBM x335 or Opteron) and then go to other vendors to beef it up, and save a bundle. Of course, then you might as well buy from RackSaver.

(I'll mention here that the HP probably has some memory correction like ECC or ChipKill (I couldn't check their site to figure out which), while the PowerMac and XServe do not. I wonder how VirginiaTech will get around this issue: 1000 machines with probably 4GB RAM = very high likelihood of a memory error!)

Which is a long way of saying that this isn't AidenShaw's point at all. His point is that there is a need for powerful dense (and ultradense) rack servers. Sure, if all you are doing is serving static web pages and running an FTP site, P3 class machines are more than powerful enough, but that doesn't mean there isn't a demand for a G5 XServe. Heck, stick a scripting enterprise web application in there (PHP, ASP.NET, Java) and you'll need the cycles and the RAM.

Not that we have to wait for Apple. If we're willing to forgo Mac OS X Server, we can wait only as long as Q1 2004 for a 970 IBM blade for their BladeCenter (and probably 1U rackmounts). Who knows, perhaps I'll get one to stick in our BladeCenter at work just to benchmark. *shrug*.

Oh yeah, besides VirginiaTech, there are a number (3) Opteron machines shooting for the top 10. Particularly relevant to the discussion at hand is the JAIST 1058 node cluster being built by IBM since the eServer 325 is reasonably priced at over $3000 (well, if you remember to price the second CPU and RAM (it has 6 bays I believe) then it will run closer to $7000) and thus would cost on the same order as the PowerMac G5 cluster with similar performance. I believe Cray is building a Opteron cluster that might give ASCIPurple a run for its money (no idea on the cost, though).

Also I noticed in that discussion, the $3 million estimate was based on the price of the machines only. We forgot to include the power costs, rack mounts, and the Infiband networking--the last of which alone adds millions to the price. Still it's a pretty good deal and I think future supercomputers will be built this way: as renderfarms for special effects and animation already do.

Finally, in the top100 list the cluster would appear as "Apple" as the vendor, not IBM. If you look down the list, you'll notice a few Itanium and P4 Xeon clusters (actually, you only have to look down to #3), yet they are listed with "HP" or "IBM" (etc) as the vendor.

Take care,

terry
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.