Intel linked with HPC boost buy
To FPGA or not to FPGA
Comment With Intel sending its "Larrabee" graphics co-processor out to pasture late last year - before it even reached the market - it is natural to assume that the chip maker is looking for something to boost the performance of high performance compute clusters and the supercomputer workloads they run. Nvidia has its Tesla co-processors and its CUDA environment. Advanced Micro Devices has its FireStream co-processors and the OpenCL environment it has helped create. And Intel has been relegated to a secondary role.
FPGAs (field-programmable gate arrays) are one way to boost the performance of HPC iron, and one that Cray, Silicon Graphics, SGI's LinuxNetworx, and a number of other HPC labs and universities have been monkeying around with for the 15 years or so. By their nature, FPGAs are more malleable than other kinds of co-processors, such as math co-processors or video cards that are rejiggered to act like them, and they can be programmed to run very complex algorithms rather than just doing single-precision or double-precision floating point math quickly.
For instance, Netezza uses FPGAs to run chunks of its database sorting algorithms in its clustered database appliances. XtremeData sells FPGAs from Altera, one of the key FPG suppliers, that plug into Opteron sockets and now also sells appliances based on FPGAs that have been tuned for analytic routines for SQL databases. Intel has dabbled with FPGAs too, providing an abstraction layer called QuickAssist to let X64 processors and their applications see FPGAs and take advantage of them as co-processors. (Intel, like many chip makers, also uses FPGAs to simulate its processors).
As reported by the EETimes, a JP Morgan Securities analyst is telling the world that Intel may acquire an FPGA supplier such as Xilinx or Altera, but he believes that Intel will pass on smaller FPGAs players, such as Actel and Lattice Semiconductor. In the same story, Raymond James Equity Research chip analyst Hans Mosesmann said the speculation that Intel might buy Xilinx or Altera was nonsense. (Mosesmann reminded everyone that Intel sold its FPGA business to Altera back in 1994 and added the requisite "they've been there, they've done that" comment).
Whatever Intel might or might not be thinking, buying either Xilinx or Altera is a very expensive proposition. When Intel shelled out $884m last June to buy embedded operating system maker Wind River, it coughed up a 40 per cent premium because the company wants to make sure embedded OSes are tuned tightly to its Atom processors so it can push these chips into more devices. This made a certain amount of sense, and more importantly, Intel could afford the $884m. But acquiring FPGA makers Xilinx or Altera is another matter entirely.
Altera had a market capitalization of $6.32bn. The company has not yet reported its fourth quarter 2009 financial results, but in the prior twelve months, Altera booked $1.14bn in sales, $231.1m in net income, and $1.36bn in cash and equivalents. With a 40 per cent premium like Intel paid for Wind River, the chip maker would have to pay something like $8.8bn for the FPGA maker - and that is if you ignore the cash hoard.
Add that in, and you are talking north of $10bn to do a deal. A Xilinx deal would not be any less expensive. Xilinx, which has about a 50 per cent market share of the $3bn FPGA business, has a market capitalization of $6.51bn and just last week raised its guidance a bit for the fourth quarter of 2009. In the twelve months ended September 2009, Xilinx had $1.64bn in revenues, $412.7m in net income, and $1.52bn in cash and equivalents. So a Xilinx deal might cost a little bit more than an Altera deal.
Intel might have $13.9bn in cash and equivalents through the end of 2009, but blowing most of that on an FPGA player makes no sense unless Intel suddenly thinks FPGAs are going to explode in the market, growing by an order of magnitude. It is far more likely that Intel might take an equity stake in one or both of these companies to bring them into the Core, Xeon, and Atom folds and partner tightly with them.
Despite what the Wall Street analysts are saying, if Intel wanted to get back into FPGAs, buying Lattice Semiconductor (market cap of $300m, annual sales of around $225m, and losses) might make sense and Intel could do it for pocket change. Ditto for Actel, which has a market cap of $312m and just shy of $200m in annual sales. If Intel is coming back into the FPGA fold after a 15-year hiatus, buying one of these loss-making companies - or perhaps both - would be a lot less expensive than trying to take over Xilinx or Altera. And much less likely to antagonize antitrust authorities in the United States and Europe too. ®
"software programmers are cheaper and easier to find "
Windows wannabees might be cheap, but they're unreliable (they may not deliver, or what they deliver may not work). FPGAs aren't windows wannabee territory.
Real programmers/engineers, especially the kind that (a) know how to decompose algorithms in a manner where parallelisation might help (b) understand the ins and outs of the FPGA world, are as rare as rocking horse poo and need TLC (and dosh) accordingly.
Exhibit 1: the chap above who thinks putting a processor on an FPGA and running gcc output on it is in some way relevant. Complete muppet.
Actel make antifuse or true flash based FPGA's.
Xilinx and Altera are SRAM based that must be configured using external flash etc. (Some have flash built in but same principle).
You can not use gcc to create an FPGA image, as you needs back end tools from the vendor for place and route.
agree with stan there. Most importantly the people who write HPC code don't know how to do it, and the people who have to support apps want no part of it - maybe when Fortran '27 is released....