Slashdot Mirror

While this sort of machine is useful (I just built one for quantum Monte Carlo calculations 6 months ago) it is hardly news. NVIDIA has been pushing this sort of machine since the launch of the Tesla. In fact, they have had a parts list on their website for some time telling exactly what is needed to put together a computer with 4 C1060's. This is not even the first commercial offering of this nature, with companies like appro and microway having similar products for at least a year (see nvidia) for a more complete list.

You are talking about 'cheaper' in regards to supercomputers? Their super computer has 68 of these, which are so expensive they wont even give you a price tag without calling them for a quote.

It's a quad-opteron in a 4U chassis. I don't get it, what's so special about that?

I'd be much more impressed with a 1U quad opteron with 32GB of RAM via 16x2GB DDR400 and 1.5TB of storage via 3x500GB drives.

Oh wait. It's already been done. It's called Appro's 1142H, a 1U quad opteron server.

Big fucking deal, a 4-way 3u Opteron Server . . .

Appro has a Quad 1u unit. Makes for an awesome cluster . . .

I do not work for Appro, but I am a customer

http://www.appro.com/ do same damn fine boxes, including 1U (yes, 1U) quad (yes, yes, quad) operton boxes that take 32GB of RAM.

I only wish the company I work for could afford boxes like that :(
Oh, and there's that "need" thing I keep hearing about.

Or, I went to AMD's page here and clicked on one of the manufacturers listed. Where I found this dual opteron supporting 16GB ram. Took me all of 2 minutes.

Just to get my word in, an Athlon64 3000+ based system would be cheaper, and much faster for gaming. They must be pretty good chips, as even Intel is buying the AMD64 boxes.

"I'm looking forward to a full Intel^3 (cpu/chipset/board) solution for ultimate stability."

I've had too many problems with Intel boards to touch them ever again. If I were you, I'd definitely go with an Asus setup. The quality is better, and the tech support is excellent. Then again, I'm a reseller, so I don't know what the end user tech support is like. Anyway, don't think that just because Intel makes lots of ads that they're somehow super-duper stable by any measure. Just as any other tech company, they have plenty of problems with their products.

You can order amd64 systems from places like appro and Penguin Computing right now, with decent sized collections of 64-bit applications provided by popular distributions such as SuSE. Let's not forget that the amd64 CPU's can run ia32 binaries at speeds faster than many ia32 CPU's and on a system with an amd64 kernel allow for more aggregate address space consumption across processes and the ability to install tremendous amounts of physical memory for buffers and cache even if individual processes can only take advantage of a few gigabytes.

With other groups like the Debian project well underway in their amd64 porting efforts, you can expect thousands of popular applications built for the amd64 platform. There's tons of software available for amd64 already, and you can bet by the time that AMD releases their "Athlon64" or whatever they're targeting the low-end market with, there will be even more.

No, we just buy well designed servers.

We have one of these, and it kicks much ass, and strangely enough never bursts into flame. I guess we forgot the napalm addition.

In many clusters today, each node relies on a hard disk to hold a distribution. This is a real big problem because hard disks generate a lot of heat. This is especially true with 10-15K RPM SCSI drives.

To add to that, most cluster node manufacturers that I know of design their 1U cases so that the hard disks are mounted on the front of the case for hot-swapping (Which in itself should suggest that people have problems with drives failing). The fans lie somewhere behind the hard disks. For example, this is a picture of an Appro 1122H dual Opteron server. Some cool air (But not a lot) is drawn from the vents on the front-right of the case, but a lot of hot air gets sucked away from the hard disks and blown over the CPUs. A better solution is just to take out the hot swap bays, make sure there's no SCSI backplane in the way, and let the cool air be drawn in from all across the front.

There are some manufacturers who will put a cardboard heat tunnel around the intake vent, lead it over some blowers, and eventually have to run over the CPUs. That's good design for keeping the CPUs cool, but it basically leaves the hard disks and other components such as north bridge chips and memory modules to fend for themselves when it comes to fighting thermal death. Often times they fail.

I have yet to see a hard disk that was smart enough to spin down to a lower speed when it reached a certain temperature. Even if they were that smart, who wants an HDD performing poorly when you've already spent ungodly amounts of money for bleeding edge 15K ultra-320 drives? As far as I'm concerned, they're practically built to destroy themselves. This is why diskless cluster nodes are so appealing to me.

Sorry for my little rant there. I hope it helped to clarify what I meant when I was emphasizing the advantages of having fewer moving parts.

You could purchase 10 APPRO AMD MP servers. These machines are based off of the Tyan boards so have all the built-in high end features you could wish for, while still being relatively cheap and easily networkable given the builtin NICs. Plus SCSI and potential for RAID if you feel you need it. and all in 1U.

http://www.appro.com/

or for the AMD systems specifically http://www.appro.com/1124.html

Should give you enough information to put together a 10U Beowulf cluster...

You could purchase 10 APPRO AMD MP servers. These machines are based off of the Tyan boards so have all the built-in high end features you could wish for, while still being relatively cheap and easily networkable given the builtin NICs. Plus SCSI and potential for RAID if you feel you need it. and all in 1U.

http://www.appro.com/

or for the AMD systems specifically http://www.appro.com/1124.html

Should give you enough information to put together a 10U Beowulf cluster...

We just (yesterday at 4 PM) placed an order for one of the new APPRO1124 dual Athlon 1U rackmounts, with a 2 GB of DDR RAM and 2 1.2 TBirds, and a couple 10K SCSI drives.

I'm looking forward to playing with it when it gets here next week. :) The writeup on Anandtech about "the perfect server?" was one of our primary reasons for deciding to give this a shot.

For the curious, the machine will be running J2SE, JBoss, and Jakarta/Tomcat, for a custom app that connects to a (in this case, local) Postgresql database, which is why we have so much RAM in it.

The place I work now has four Cobalt RAQ's (2 Raq 1's, 1 Raq 2, and a Raq 4). However, before we got the Raq 4, I found the Dell PowerApp series. Personally, I was quite impressed with both the features and the price. They didn't buy it only because a promotion allowed them to buy a Raq 4 for a very nice price.

Anyway, the Dell PowerApp's come in two flavors really useful to Web hosting companies - a web hosting rack mount and a web caching rack mount. They have CD-ROM drives (unlike the Cobalts), run SCSI hard drives (unlike some Cobalts), support both RH Linux or Windows NT. It also features a web based GUI configuration system. While personally I wouldn't buy the NT version, I can easily see this as being useful to companies who want to maintain different OSes but the same hardware. They run Celeron's or Pentium III's and are priced slightly cheaper (at least when I looked at them) than their comparable Cobalts.

I really liked the floppy/CD-ROM that's built in, which makes it much easier to reinstall than the ethernet install required by the Cobalts. It also has a video, keyboard, and mouse port, so one can run X to config or their precious NT environment.

I've been noticing a lot of Cobalt-like boxes in Linux Journal, like Penguin Computing's Relion servers. Appro (never heard of them before) also offers some rack mounts. Apparantely Cobalt's popularity has really brought many competitors to the field, as they seem to be rather numerous in my Linux magazine.

Hope this helps.