Intel in Talks to buy Altera

Oups, here's another link about the same news that doesn't require to log in:

--------------------------------------- Posted through

Hmm. While I have tons of respect for Intel as a company that makes stuff that people will buy, I'm old enough to have seen more than one generation of Intel embedded processors go by the wayside when the PC market picked up.

So I don't trust Intel's attention span vis-a-vis whatever they happen to think their core business is. If they kept Altera as an easily-spun-off business unit, and kept it supported, then I could see them spinning it off again when the PC market did pick up, or by some miracle they managed to make cell phone processors that actually worked, or something.

Not just embedded processors...

It wouldn't be the first time Intel was in the programmable logic business.

Didn't Altera buy Intel's PLD business back in the '90s???

Sorry. I don't see how this could be a good thing for Altera.

Rob.

Yeah, I'm concerned too. I'm hoping that Altera is big enough that Intel won't want to mess with them and destroy the company.

I think those data sheets are from the days when dinosaurs roamed the FPGA earth and was Intel's own attempt to enter the market. I have no idea why they actually bailed. I can only assume the competition was stiff then with a number of startups including Neocad providing the place and route software for a number of these companies. Xilinx has said they spend more on software development than they do developing the hardware. Several of these companies dropped their in house software development due to the huge cost. Maybe Intel dropped the product line because of it. But much more recently they were working with a company to produce some much more advanced product which I believe may still be operating using Intel's fab technology, or has it also gone belly up? I don't recall the name.

Looks like Intel likes the Altera approach and want to keep it, literally.

Yes, more advanced - ten times faster and/or lower power. Asynchronous, so no clock tree, but hard to get technical detail:

Jan Coombs

Well, Intel were Altera's 'second source' (back when that mattered) and (if I'm not mistaken) their original fab back in about 1983 - which is where the "Intel FPGAs" mentioned in another post came from.

So there's quite a long association there.

-- Brian

FCCM predicted this in 2011...

--Mike

Wasn't there also some kind of Intel Atom + Altera FPGA on the same chip? Ah yes, the Stellarton.

I guess those weren't huge sellers... But last summer Intel announced some upcoming Xeons with FPGA logic on board. No doubt targeting the server markets where FPGA coprocessors have made appearances recently.

I thought that was two die in the same package, no? Reading your article the term, "system-in-package" indicates multiple die. It think the advantage is in reducing the system size and having a very direct connection between the two chips. Putting them on one die would likely make a faster connection possible, but would be much more difficult to pair in various combinations. That's the big reason why FPGA makers have resisted for so long incorporating CPUs on the FPGA chip until recently.

More likely Intel wants to head off the market which has ARM processor(s) plus custom glue logic. Target: highly-integrated low-power server farms.

Intel acquired Wind River back in '09 but they've left them alone as a separate entity, so we can only hope they do the same with Altera.

I agree with the separate die part.

Uh, resisted? Xilinx had the Virtex 2 and 4 Pros with integrated PowerPC cores around a decade ago and I'm pretty sure that was single die. I was involved in such a project in 2006, Virtex 4 Pro was new then and Virtex

Altera had a similar thing, Excalibur I think? I have to assume those weren't big sellers either since they pretty much disappeared.

I have to assume they weren't resisting to offer these. My guess would be they couldn't offer them at competitive prices. Maybe the new generation is doing better in that regard.

Yeah, they had those and Altera had an ARM and Atmel had a... I think maybe an AVR on the die with the FPGA. But they all let them die rather than continue the part in the next generation. When asked Xilinx people in particular (who used to be very vocal in this group... in fact, too vocal which is why they aren't here anymore) said the problem was the many combinations of CPU, RAM and FPGA that would be required, not to mention the packaging. Seems they don't play the same game the MCU makers do who regularly have dozens if not hundreds of combinations of any given processor.

Also, this was the era when all FPGAs were power hungry and expensive. So yeah, they resisted the rest of the market who would have loved

I think the difference is that their market is determined by pretty much one industry, comms. When they say they would like to see a part, A and X reply, "Will Tuesday be soon enough?" I expect *that* market reached a point where the integrated ARM was powerful enough and needed tightly coupled enough that it became a worthwhile part.

I know the FPGA companies have all been chasing the high volume markets like automotive and even phones and tablets. But all of these guys have volumes that just don't make sense for FPGAs 99% of the time. Comms is where the FPGA profit is and will be for some time to come.

I was on a project that gave those parts a serious look. We ended up using a plain old FPGA talking to a processor. Partially this was because we had a ton of boards that were all using that same processor but without the FPGA, but partially this was because we didn't see a great big advantage in terms of project hours to the embedded processor.

I have thought for a long time that instead of hugely complicated, specialized peripherals, a chip that has a processor with an FPGA mapped to the peripheral space could have some use -- it seems like you never have the peripherals you'd really like, unless you have a whole bunch of peripherals just sitting there sleeping. There's probably a ton of practical reasons why it's a dumb idea.

There are two popular products on both ends of the spectrum that have been around for years: microcontrollers with programmable logic peripherals (for example, Cypress PSOC) and a microcontroller core implemented on an FPGA (for example, Altera NIOS).

I've used both example products with great success. As you said, it's real convenient to roll your own peripherals with impunity. It saved me hours of coding effort when you can smartly implement the peripheral of your dreams with a little HW design.

JJS

Calling the PSOC circuitry "programmable logic" is a bit of a stretch. Yeah, I guess technically it is logic and it is "programmable", but it is not at all general purpose. In some devices it has been dumbed down to the point of being configurable serial devices that can be a UART, SPI or similar devices, sort of a super USART... not really much like FPGAs or even PLDs. Even the parts that have some programmable elements are not terribly flexible.

The part that gets me about the newer versions of this theme is that they are large, pricey FPGAs and incorporate fairly high end CPUs which are typically programmed under Linux... a very far cry from the efficient solution I would like to see. There are few engineers who can even design the entire system on that chip spanning logic design and system programming.

Agreed. We're looking hard at both Zynq and the Cyclone V SOC, both of which have big monster Cortex A9s meant to run Linux with a mess of DRAM and etc. Which, I mean we can make work. But if I could get a 10-20KLUT FPGA with a dual or quad Cortex M4 instead? Nice and light with every intention of running bare metal with 10-20K of code? I'd take it in a heartbeat.

Is there any info on the price range for the latest Xilinx UltraScale+ parts?

Learn Forth and make that 4 to 8 KB of code. Which brings us to the possible reason they don't make the lower end CPUs in an FPGA... it competes against soft cores which can do the job you describe very well. There are fast, efficient CPU cores which run at 100 MIPS and use under 1000 LUTs, not so much in your 10-20 kLUT device, eh? Very probably cheaper in the long run than a chip with hard cores.

If you want to run bare-metal quick and easy, check out the J1 Forth CPU by James Bowman. I've rolled my own similar CPUs and I really like his design.

Just as a point of comparison, the GA144 is a bit like an FPGA but uses tiny 18 bit processors as logic elements, 144 of them. It pretty well sucks in many regards... well, "sucks" is a poor choice of words I guess. But they had a cool idea of the tiny processors and connecting them in a net, but then totally forgot to consider any applications when designing a chip around it.