Should Apple create its own cluster?
I was wondering if it would be a good idea for Apple to create its
own cluster when they come out with the G5 server. And I'm
talking about a cluster that is huge compared to VA Tech cluster.
5,000 dual processor G5 servers hooked up together & optimized
with apple's Xgrid technology. It would be a showcase that they
could use to promote mac clusters to other businesses, colleges,
& etc.
Costwise it would be cheap. They can get the servers at cost, plus
they could rent it out to FX companies, colleges, the government,
or anyone else that needs that kind of processing power. The free
press to having one of the largest clusters on the west coast would
be priceless.
I can just picture it now. Pixar rents it to render their next picture in
a fifth of the time that it would have taken with their render farm.
The DVD has a featurette showing the render farm & how kickass it is.
own cluster when they come out with the G5 server. And I'm
talking about a cluster that is huge compared to VA Tech cluster.
5,000 dual processor G5 servers hooked up together & optimized
with apple's Xgrid technology. It would be a showcase that they
could use to promote mac clusters to other businesses, colleges,
& etc.
Costwise it would be cheap. They can get the servers at cost, plus
they could rent it out to FX companies, colleges, the government,
or anyone else that needs that kind of processing power. The free
press to having one of the largest clusters on the west coast would
be priceless.
I can just picture it now. Pixar rents it to render their next picture in
a fifth of the time that it would have taken with their render farm.
The DVD has a featurette showing the render farm & how kickass it is.
Comments
Fairly hefty showcase there. I think the Virginia Tech setup is the showcase (but not by Apple's expense). When the G5 Xserve does come out, Apple could use the VT setup and then claim it can be had for half (or a third or a quarter) of the space.
Screed
I just did the math. The G5 is 3044 cubic inches and the Xserve is 852. 3.57:1
Originally posted by mello
How much is an xserve at cost?
Probably not much less than a Dual MDD-PowerMac G4. Add in a few extra fans, LEDs, etc...my guess would be $1000-$1200.
Heh. Could you imagine going to store.apple.com and entering 1100 into the quantity field?
3 day FedEx shipping? Eleventy billion dollars.
Screed
Originally posted by sCreeD
Currently the dual G4 Xserve costs $3800, so yes it costs more than the dual G5. However, that includes a copy of OS X (Unlimited). I'm sure VT got a discount on such a huge order.
Heh. Could you imagine going to store.apple.com and entering 1100 into the quantity field?
3 day FedEx shipping? Eleventy billion dollars.
Screed
Configured with dual drives, a must:
Item: Xserve Dual 1.33GHz
Part No: Z09P
Est Ship: 3-5 bus.days
$ Each: $4,049.00
Qty: 999 (the most you can enter)
Total: $4,044,951.00
Yeah, baby!
Apple is not the company you normally think of when you hear the word "supercomputer", but nevertheless it can now count itself among the ranks of the supercomputing elite, courtesy of the Virginia Polytechnic Institute. Technicians and students at Virginia Tech have assembled a supercomputer using 1,100 64 bit G5 Apple Macs. In effect, they have linked together 2,200 IBM p5 chips and the result is a massive processing capability capable of 7.41 trillion operations per second.
As far as supercomputers go, that makes it at least the fourth fastest in the world. Virginia Tech is still taking measurements and is suggesting that the final rating may be significantly greater. The three faster machines (in order) are the Japanese Earth Simulator, a machine at the Los Alamos Laboratory and one at the Lawrence Livermore National Laboratories.
If you're thinking "so what?", the "so what" is that this computer was assembled in the space of about one month using a largely volunteer labour force and cost just a little over $5million. Compare that to the typical $100million to $250million and many months that it normally costs to assemble a world leading supercomputer. Supercomputers are normally put together slowly in a custom built manner that takes time and money - although the Lawrence Livermore system which has 2,304 Intel Xeon processors was also assembled on the cheap, with an estimated cost of somewhere between $10m to $15m. It is slightly faster than the Virginia Tech machine at 7.63 trillion operations per second.
The decision to assemble the computer seems to have been almost a spontaneous act. Scientists from Virginia Tech met with Apple in June this year just after it launched its new screamingly fast 64 bit desktop and Apple agreed to provide the college with some of the first machines off the line. It turns out to have been a clever move by Apple that will undoubtedly increase its credibility in many areas outside academia.
Screed
Originally posted by McCrab
With the development of xGrid, what makes you think that they don't already have a sizeable cluster?
Like someone else said, they do have showcases: VT is one. Out here, they told me yesterday that UCD has two: One ~100 (Bioinformatics) node cluster, and one ~10 node (Food Science). They are aggressively pushing clustering. They all but promised me that they would beat anyone on price/performance when we build our new cluster (yeah, I know, they all say that, but they're walkin' the walk right now).
If I get my promised tour of XGrid and existing clusters, I'll post my impressions.
altivec & etc?
Originally posted by Rhumgod
Configured with dual drives, a must:
Item: Xserve Dual 1.33GHz
Part No: Z09P
Est Ship: 3-5 bus.days
$ Each: $4,049.00
Qty: 999 (the most you can enter)
Total: $4,044,951.00
Yeah, baby!
Please turn off one-click shopping!
Apple expecting major surge in cluster computing sales: According to reliable sources in Cupertino, Apple is hiring new staff to coordinate large-scale purchases of G5 systems (currently only PowerMacs of course, but soon to include Xserves and possibly the rumored Xstation) for use in clustering systems - scientific, educational, business, and governmental applications primarily.
Documents recently acquired by rumors include a memo which states that "we now expect yearly systems sales directly attributable to clustering/grid computing applications to exceed 50,000 units in 2004 and 80,000 units in 2005." The memo goes on to state that indirect sales for systems that will distribute loads across multiple computers in less formal environments could eventually double those numbers. Xgrid, Apple's distributed computing technology suite, is under heavy development and Apple is hard at work on other aspects of its hardware and software lineup to take into account rapidly growing demand for these technologies -- not just industry-wide, but also very particularly in terms of consumer interest in Apple itself, following the much-publicized Virginia Tech G5-cluster supercomputer.
Originally posted by KANE
Apple should simply put 5 000 or more Dual Powermacs together in a big, big room and then simply install the Folding@Home client on them. Can you say "Team Apple Computer for #1"?
After how many minutes
Originally posted by mello
Does that Linpack testing take into account mac features like
altivec & etc?
The etc part, yes, but AlitVec, no. IFAIK. There might be som cool stuff in IBM's compilers for doing autovectorizing of some things in Linpack, but I think that the numbers we see from the VT cluster is mostly the double FPUs talkng.
Originally posted by Eugene
Apple needs to come up with server/workstation hardware that supports ECC RAM first...
Sorry, but I don't get this ... what's ECC ram, and if it's so important, why isn't it stopping Big Mac?
Mr. Curious
Originally posted by OverToasty
Sorry, but I don't get this ... what's ECC ram, and if it's so important, why isn't it stopping Big Mac?
Mr. Curious
Error Correcting Code...it really isn't that crucial nowadays...back in the OS 9 days it may have been a bonus (or Windows 9x days, for those Microsoft shops), but nowadays, it really isn't too important. Servers on the Windows side have had support for them for years.
How often does memory corrupt data?
But I can't see how it would hurt...