Specific HPC product news we found of interest:

- Cisco, Reuters and Sun announced an end-to-end low-latency solution for automated and algorithmic trading that is said to reduce overall system latency by 38%.

- Microsoft and GigaSpaces announced that they have been integrating Excel 2007 with GigaSpaces’ in-memory data grid solution at individual customer sites and will productize a joint solution in 2008. This will help eliminate data latency in such scenarios as research analysts making research available to traders and spreadsheet-based risk management. Each Gigaspaces unit can handle data services such as validate, check/match and execute order at a pace of 1,000 transactions per second and according to Geva Perry, chief marketing officer, and the units are infinitely scalable – two units will process 2,000 transactions per second and so forth.

Excel is ready to run on Intel’s and AMD’s forthcoming quad-core chips, Stevan Vidich said. “Out of the gate, Excel 2007 will run four times faster on quad core for most Wall Street workloads,” he said.

- Teradici showed a demonstration of their remote computing offering and told us that IBM and ClearCube as well as Verari have agreed to resell it and units are in production. That means this interesting solution designed for Wall Street trading floors will be widely available next year. It puts a small device on the trader’s desk that accepts images from a blade or PC in another location (such as a data center) – each unit contains a special chip that compresses and decompresses images for rapid delivery. The desktop device, which was envisioned a few months ago as the size and shape of a hockey puck, is in its current incarnation more like a miniature PC tower, about two inches wide and five inches high. This allows for cleaner workstations that consume less power and simplified IT support, reassignment of PCs, and “waterfalling” – giving the best computers to the most-valued employees and the junk to those least valued. While earlier forms of remote workstations or thin clients were criticized for being limited in the amount of distance they can handle, one customer is testing the product with five kilometers between devices and PCs; according to a Teradici spokesperson the blades could easily reside at a data center in New Jersey and feed devices on Wall Street trading floors or at traders’ homes so they could work from home, particularly helpful in an emergency.

- IBM updated us on their Cell chips. “There are five million chips out there,” said Kevin Pleiter, director of the global financial services sector at IBM. “Hundreds of millions of dollars have gone into the development of the Cell.” The Cell chip has been packaged in traditional PC and server blades and is already in its second generation in this format. Its best used on Wall Street to handle core math code, Pleiter says. The third generation will come out in the second quarter of 2008. “Already, the Cell is typically five to six times faster than a quad-core processor,” Pleiter said.

Today IBM announced that the Cell processor can run Red Hat’s Linux operating system. IBM has also been working on a development environment that will allow for rapid development of applications to run on the Cell, this will be generally available next year.

“Most major global banks are working on significant proofs of concept of this, using it to help them process exotic derivatives,” Pleiter said.

- Nvidia showed their new Tesla graphics processor unit, which comes in a desktop PC unit for trader workstations and a server blade for the data center. "This GPU is 50 times faster than a CPU," said Gerald Hanweck, Jr., principal at Hanweck Associates, a consultancy that has deployed the Nvidia graphics chip at several investment banks and hedge funds. Which applications are best suited to this type of specialized processor? "Anything that usees Monte Carlo style simulations, trees or lattices," he said. "This chip is good at pricing large amounts of data, it's useful for options pricing." Some clients are achieving a 100X speedup over single core processing, he said.