Here is a collection of highlights, selected totally subjectively, from this week's HPC news stream as reported at insideHPC.com and HPCwire.

10 words and a link

Fujitsu plans for 10PFLOPS super in 2011

UK Met Office caught in green flap over super emissions

New Zealand buys big IBM super

AFRL dedicates new SGI super

Maui doubles computing power

OSC expands flagship IBM supercomputer to 75TF, 9500 cores

Student research that fundamentally changed computing

PGI tutorial on GPU programming during SC09

Masterworks sessions announced for SC09

AMD launches 40W Istanbul

Greening this old historic datacenter

MemoryScape 3.0 released

OSC expands flagship IBM super

The Ohio Supercomputing Center, Monday, announced a new expansion to its current flagship IBM supercomputer. The $4 million effort is billed to provide further computational support for the state's economic development aspirations. The expansion integrates additional hardware into the original, IBM Cluster 135, system that went operational in January 2008. Due to overwhelming demand, the original system reached capacity in three months.

"The expansion, which is dedicated to bioscience and additional research efforts targeted by the state, provides badly needed high performance computing resources for academic and industry researchers," said Stanley C. Ahalt, executive director of the Ohio Supercomputer Center.

"For every $1 the State in Ohio invests in OSC, $17 returns to the Ohio economy," Ahalt said. "With these critical, new supercomputing assets in place, researchers can further propel Ohio to the forefront of biosciences research, funding and job creation."

The new, combined, system provides 75TF of compute power and 24TB of memory via 9,532 cores. The specifics are as follows:

• 75 teraflops of theoretical peak performance
   
• 1,624 nodes/24 Tbytes of memory/9,532 cores
  • 877 dual socket, dual core nodes (2.6 GHz Opteron); 8 GBs memory/node
  • 86 quad socket, dual core nodes (2.6 GHz Opteron); 16 GBs memory/node
  • 2 dual socket, dual core nodes (2.6 GHz Opteron); 8 GBs memory (login nodes)
  • 650 dual socket, quad core nodes (2.5 GHz Shanghai); 24 GBs memory/node
  • 8 quad socket, quad core nodes (2.4 GHz Shanghai); 64 GBs memory/node
  • 1 e1350 Blade Center, with 4 dual cell-based QS20 blades
• All connected together by 10 Gbps Infiniband

For more info on the OSC expansion, read the release here.

Cray acquires PathScale Compiler Suite

Cray has announced that it has officially acquired the PathScale Compiler Suite from the recently defunct assets of SiCortex. Cray plans on leveraging pieces of the PathScale IP in order to enhance its own compiler offerings over time. Other parts of the suite will be contributed to the open source via an alliance with NetSyncro.com. NetSyncro is an organization of compiler engineers with good ties to the open source community. NetSyncro will continue developing the PathScale Compiler Suite, provide support for its current users and will rebrand the effort under the original PathScale name.

"Our main goal for this acquisition was to provide clear direction for those Cray customers who want to continue using the PathScale Compiler Suite on Cray supercomputers," said Peter Ungaro, Cray president and CEO. "We believe turning the PathScale compiler's future development and customer support over to a new and similar PathScale organization accomplishes our goal, as it provides a path forward for PathScale compiler users and helps ensure that the software will have a robust, open source HPC community around it. We also expect our own world-class compiler to benefit from some of the PathScale technology. This is another indicative example of our strategy to acquire the key technology components necessary to building a productive, high performance user environment on our supercomputers."

"I cannot express enough gratitude to Cray in helping rebuild PathScale and giving us this opportunity for the future," said Christopher Bergström, PathScale's new CTO. "Our vision ahead is bright and optimistic with a focus to continue our position as one of the highest performing HPC compilers in the industry. As a new member of the PathScale team I intend to lead the way in building a strong open source community for PathScale. We believe our solid commitment to open source will enable the community to thrive by facilitating collaboration, sharing of knowledge, innovation and research."

For those unfamiliar with PathScale, it started life as the Silicon Graphics MIPSPro compiler. When Silicon Graphics decided to divest itself of the compiler business, it sent the MIPSPro compiler to the open source. It then took on the new name of Open64. PathScale adapted much of the lower end semantics from Open64 [with its obvious secret sauce on top]. For more info, read the full release.

Dell and Brocade team up on application to unified computing club

It started with Cisco adding servers to its datacenter communications products on a project codenamed California and marketed as the Unified Computing System. Then in late July IBM announced that it was teaming up with Juniper to do the same thing, and HP already has its own line of switches and servers. This week Stacey Higginbotham at GigaOm is reporting that Dell and Brocade are teaming up, along with enterprise cloud software maker Scalent, to get in on the unified computing market.

According to the article Dell will put its name on Brocade's switches.

Dell's partnerships means that HP, which is combining its own servers with its Pro Curve networking gear, IBM, which has signed the partnership with Juniper and Brocade, and Cisco are all offering some type of networking and server product to manage some of the chaos caused by the ability to run multiple applications on one server. After Cisco launched its servers, a Dell executive told Om that its ability to offer its customers gear from multiple vendors, rather than a box from a single vendor would be its answer to the Cisco threat, and it's still pitching openness with this announcement. Only now the focus is on openness as a result of adhering to industry standards rather than grouping its servers with switches from any vendor.

I agree with Stacey's assessment that the partnership route here seems a little weak. Furthermore, it's not really clear yet that this strategy even makes sense for customers. It might have been smarter for Dell to wait and see if the market really responds to this offering and, if it does, to swoop in as a second mover fixing all the errors the first movers made and take over the market.

For those of you who might have been considering clusters from these vendors, does the unified story matter to you?

-----

John West is part of the team that summarizes the headlines in HPC news every day at insideHPC.com. You can contact him at john@insidehpc.com.