2CPU.com » News » December 2005 » Supercomputing 2005 Hosts TYAN

SC|05, 11/11/2005 - At this year's Supercomputing 2005 event, Tyan Computer Corp. and Rocketcalc LLC both showcased the latest in server and clustering technology including a world first: the Personal Supercomputer. The Personal Supercomputer (PSC) project represents the latest in a series of initiatives from the research and IT community to support small science, a term applied to scientific applications powered by scaled down machines designed to deliver supercomputing performance at a fraction of the cost. Machines with enough power to run tsunami simulations, but small enough to fit on a desktop! Find out more by visiting Rocketcalc's website.

11/14/2005 03:39 PM: Cray to Build more Opteron Supercomputers by admin
Tech Report has linked to a press release by Cray regarding the creation of new Opteron-based supercomputers. AMD (NYSE:AMD) and Cray Inc., (Nasdaq NM:CRAY) today announced that they have signed an ag...

09/22/2005 02:12 PM: Apple-based supercomputer planned by duke
The Inquirer has a post up about a new supercomputer that is being planned by a European startup. It will utilize thousands of Apple Xserve servers. Omneta is building an Interoute internet-linked clu...

10/27/2004 01:56 PM: Itanium Powers NASA Supercomputer by duke
The Inquirer also has a post up about a new supercomputer that NASA has thrown together using Itanium2 processors. The $50 million machine is a cluster of 20 computers built by NASA's Ames Research Ce...

09/29/2004 02:01 PM: IBM Owns World's Fastest Supercomputer by duke
Slashdot has linked to an Associated Press article claiming that IBM's Blue Gene supercomputer is now the fastest in the world. IBM said its still-unfinished BlueGene/L System, named for its ability t...

07/27/2004 01:57 PM: US Navy buys IBM supercomputer by Jim_
The Register has some information about a recent purchase by the US Navy. How about a new IBM Supercomputer? The US Navy has bought a new supercomputer from IBM that will, at a theoretical peak perfor...

07/20/2004 02:42 PM: New SGI supercomputer to scale Linux to 1,024 CPUs by Jim_
Computer World down in Australia is telling us about a new SGI supercomputer that might just scale to 1024 CPUs. Silicon Graphics is building an Altix supercomputer for the National Center for Superco...

06/22/2004 02:49 AM: US Army builds another G5-based supercomputer by Jim_
Our Irish friend Tom was kind enough to let me know about a new G5-based supercomputer that the US Army has thrown together. They've labelled it MACH 5. COLSA today announced the purchase of 1566 dual...

05/25/2004 08:22 PM: Microsoft creating Windows for supercomputers by Jim_
Hooz let me know about this story over at C|Net. Microsoft has launched an effort to produce a version of Windows for supercomputers. Trying to gain share in a market currently dominated by open-sourc...

02/27/2004 02:35 PM: Cray buys into AMD supercomputing by Jim
ZDNet is running an article which discusses Cray's acquisition of OctigaBay, a manufacturer of "special-purpose computers" based on AMD's Opteron. OctigaBay, based in Vancouver, is working o...

02/11/2004 10:43 PM: IBM sells climate research supercomputer by Jim
C|Net is reporting that the University of California, Irvine has bought eight IBM machines to power an interconnected collection of global climate simulations. The overall system, which cost more than...

jamesmcn Registered User Posts: 64 Joined: 2005-11-29

#38488 Posted on: 12/08/2005 02:04 AM Given that it can run off of a standard 120v outlet, it must be populated with the (expensive!) low power chips. Cool stuff, but if you can afford a few more amps on your circuit, you could just cluster up 8 1U dual core servers. Is there really a market for these micro-clusters, or are these systems just neat PR pieces?

dafid Registered User Posts: 2 Joined: 2005-12-08

#38490 Posted on: 12/08/2005 04:41 PM Hmm - 8 * 100 W at 80% efficiency = 1000 Watts drawn 110 v * 15 amps = 1650 Watts; leaving a good chunk for memory and stuff. Not so expensive chips at 100W thermal output. The chips will be expensive I **expect** because the will use the 8-way ones with the extra hyper link... The big difference between your 8 1u servers is the memory bandwidth/latency between the cores. Those Hypertransport link between the CPUs add up to lots of GBytes/sec - worst case access memory being two HyperTransport latencys; ie high speed buss transactions versus the network access methodology of the 1Us..

dafid Registered User Posts: 2 Joined: 2005-12-08

#38491 Posted on: 12/08/2005 04:57 PM i just reread the site - 4 boards - networked with Gigabit networking with 20 microsecond latency (optional IPMI ) so there are 4 groups of (2 or 4 cores that can acess memory with NUMA speeds) where memory access between the groups is LAN speed. Not so attractive;- my app is not that distributable and 20 microsecs is a LONG time. But it does mean they can use the MUCH cheaper 2 way CPUs.

jamesmcn Registered User Posts: 64 Joined: 2005-11-29

#38493 Posted on: 12/08/2005 10:08 PM It is nice that you can get quasi-COTS NUMA gear. The HyperTransport links definitely make access to non-local memory faster, but what is the granularity of memory access? Suppose you have a hash table that all of your processors need to share. The table resides in the memory of core #0, and cores #1, #2, and #3 need to access it. Will mutexes grind the system to a halt, or will an application like this fly on a NUMA box?

2CPU.com » News » December 2005 » Supercomputing 2005 Hosts TYAN