Now that Intel has secured Altera for $16.7 billion and has a deal with eASIC, Intel is firmly in the Field Programmable Grid Array (FPGA) business. Now what's it going to do with it?
For starters, eASIC remains a partner. CEO Brian Krzanich was asked about it during the press call announcing the Altera deal and he said there is no change. eASIC will offer benefit for certain workloads, and an Intel spokesmen told me eASIC is somewhere between an ASIC and an FPGA so it’s not like Intel has two FPGA companies under its wing, per se.
There are several reasons for the deal, according to analysts I've spoken with. For starters, it gives Intel more revenue, an obligation for a public company. The mobile business isn't bringing in any money; in fact, it's been bleeding billions of dollars.
"Every company gets to a point where they can only grow so much and start reaching out for acquisitions. Intel needs places to grow with high profit margins. There's very few of areas out there. The last remaining arena with high profit margin is FPGA," Jim McGregor of Tirias Research told me.
Custom accelerators like FPGAs are good in parallel processing scenarios when one task needs to be repeated over and over, like search or pattern matching. That's why search engines like Bing use FPGAs, and a company called Ryft has an FPGA-based Big Data machine it claims is up to 200 times faster than a Xeon.
Intel is very strong in the data center and having a rich FPGA offering will only cement that position further.
"Altera will enable some new growth in HPC where CPUs and FPGAs are both used, but being the dominant force in their primary strategic market, CPUs, it is difficult to continue growing CPU share in PCs and other compute platforms," said Tom Hackenberg, principal analyst for embedded processors at IHS.
With the PC market slowing and mobile going nowhere, Intel needs a new growth market. "A reasonable strategy then is to look to expanding other markets where they participate or finding new markets. This is where Altera comes in," said Hackenberg.
Another reason, McGregor surmises, is that Intel needs to fill out its fabs. Intel has either shuttered or stopped the expansion of fabs in recent years as sales have suffered, and that's very expensive equipment to leave idle. FPGAs have large dies and large margins. Intel had a manufacturing deal with Altera, but after it struggled with 14 nanometer, it went to TSMC.
Both McGregor and Hackenberg expect Intel to use its FPGA solutions to make a big push into industrial IoT where machinery and heavy equipment rather than wearables are involved.
"A high performance router or switch for example may have one or more high-performance CPUs, but to get to extreme bandwidths such as 100GbE platforms, much of the processor complex is facilitated by co-processors such as FPGAs or application specific network processing units," said Hackenberg. This will give Intel a one-two punch of selling processing units and accelerators.
"I expect them to go into Big data and Analytics," said McGregor. "There's huge interest from the market and you can't do everything with an x86 core. I would expect we're going to see a new breed of products to handle this. But it may take five years to integrate that."