Interesting article about Xilinx FPGAs in the new Cray

I'd imagine this is mostly the same as or is infact what they got from the Octiga Bay purchase awhile back.

The article it self is fluff, but it would be interesting to know what various customers are actually doing with these.

Anyone know the price?

I wonder though about some applications split between an FPGA and Opterons. If the app doesn't need too much of the Opterons abilities, probably better off using a couple of soft/PPC cpu cores right in the fabric. Now if FPGAs ever get embedded FPUs maybe 1 for every n muls or so, probably would'nt need the external cpu.

regards

johnjakson at usa dot com transputer2 at yahoo dot com

Apart from abusing the word 'leverage' as a verb, the article seems big on gas and small on substance. Which is essentially

"Cray machines are using Xilinx FPGA as computing engines".

BFD.

The Sanger Centre has used FPGA solutions to accelerate pattern-matching in DNA sequencing.

And that was not rocket science either, as this is essentially comparing a pattern with bits in a shift register.

Well, fpgajournal is more focused on advertisement than on science... and they are late: the XD1 was presented at SuperComputing'04.

Cray is not the first to see DNA comparison as a possible application for high performance computing, and the market is already well occupied by IBM, SUN, ..., life science departments. Sun Servers can include Timelogic FPGA accelerators for DNA applications, and they know what thay are talking about.

The real question, I think, is how to use XD1? Who will make the RTL? Who will split the algorithms on the cluster, and then, between SW and HW? "C to RTL" is far from ready, and still need a lot of human intervention. So what? Cray will sell the machine or the engineers going with?

I'd like to play with, anyway!