Intel in talks to buy Altera

[Idontcare]

FPGAs are increasingly recognized as very helpful in select niches within the Data Center. They tend to be high margin products because they bring value to customers by reducing the costs to process a target workload. Intel has a reasonable share of the Data Center market and has previously spoken of it's desire to expand into adjacent markets. The synergy of using a shared sales forces for marketing results in cost savings as does the the reduction of duplicated administrative costs. Customer's preference for a single source of supply that insures that "stuff" works well together may aid in marketing efforts. Intel has some semiconductor design experience that may be helpful with FPGAs. Finally, by bringing FPGAs in house, Intel has guaranteed supply and competitor actions will never deny Intel's access to this resource. On it's face, the possible acquisition appears straightforward and forward looking.

What you write makes sense, but I have never heard Intel talk about niche opportunities this way. Is this changing with BK at the helm?

 

[Exophase]

Seems like a much better acquisition than McAfee was at $7.68 billion, at least.

 

[schneiderguy]

There are a lot of really cool stuff I am imagining you can do with the Altera tech. This allows for reconfigurable tech (to an extent), right? Would it, in theory, be possible to say have it act as an IGPU (or IGPU extension) one moment, then say fixed-hardware encoder/decoder, etc the next?

It takes a while (probably hundreds of thousands of CPU cycles) to re-configure current FPGAs.

I'm imagining a situation where we get a generic, reconfigurable accelerator that can be repurposed depending on what you are doing. How reasonable is that, or does the nature of the chip pretty much destroy perf/watt?

Performance/watt is good in highly parallel applications. Similar to a GPU, you can get much greater performance/watt than a general purpose CPU in some workloads. FPGAs can beat GPUs in performance/watt in some applications.

Tool support is probably the biggest hurdle to Intel sticking an FPGA on a mainstream CPU. If they could develop a compiler that is smart enough to figure out which parts of your code could be sped up with the FPGA resources, convert your code to the RTL equivalent, do timing closure, and automatically connect the CPU and FPGA resources properly that would be interesting. With the current state of FPGA tools that's probably 10+ years away though.

 

[podspi]

It takes a while (probably hundreds of thousands of CPU cycles) to re-configure current FPGAs.



Performance/watt is good in highly parallel applications. Similar to a GPU, you can get much greater performance/watt than a general purpose CPU in some workloads. FPGAs can beat GPUs in performance/watt in some applications.

Tool support is probably the biggest hurdle to Intel sticking an FPGA on a mainstream CPU. If they could develop a compiler that is smart enough to figure out which parts of your code could be sped up with the FPGA resources, convert your code to the RTL equivalent, do timing closure, and automatically connect the CPU and FPGA resources properly that would be interesting. With the current state of FPGA tools that's probably 10+ years away though.

Interesting. I was thinking more along the lines of a future version of say, Cinebench (why not?) re-configuring at launch - or a video game re-configuring that portion of the CPU as a sort of physx accelerator. There are some issues of course (there would have to be some system that acted to make sure two processes didn't interfere with each other), but the possibilities look ... interesting.


Doing it automagically at runtime would be amazing, but I wasn't thinking about that - though it would be awesome. I was more thinking that it could be programmable via some API.

 

[AllDayBreakfast]

What I am looking to understand is the growth picture that is implied in any investment effort on Intel's behalf. Does Altera really have a >60% GM trajectory with 8-10 years sustained business opportunities?

Historically Altera's GM has been 60-70% (closer to 65-70%) after their foundry partners have charged them for their purchases, which would suggest that if they merged, Altera products manufactured at Intel would be a good deal above Intel's corp average GM.

 

[Phynaz]

Deal is dead:

http://www.eetimes.com/document.asp?doc_id=1326322

 

[witeken]

Deal is dead:

http://www.eetimes.com/document.asp?doc_id=1326322

With such an offer ($54 vs $42), people are waging that the acquisition will continue anyway .

 

[liahos1]

With such an offer ($54 vs $42), people are waging that the acquisition will continue anyway .

people are wagering that an activist is going to push management to accept 54. A $60+ ask is ridiculous.

 

[ShintaiDK]

The sellers must be very confident in a near future higher stock price.

 

[Phynaz]

Word is this deal is still alive.

 

[arandomguy]

Looks to be agreed to now at the same $54 a share

http://uk.reuters.com/article/2015/06/01/us-altera-m-a-intel-idUKM1KBN0OH2E020150601

 

[ShintaiDK]

Nice to see it settled.

 

[sm625]

Both Intel and Altera spiked hard at the end of march. Clearly there was some sort of inside information that got acted upon back then.

 

[ikachu]

Yeah there is definitely some shady stuff going on with all these mergers... for example the Broadcom / Avago merger leaked Wednesday afternoon and Broadcom stock jumped 20%. Someone made a lot of money there...

 

[witeken]

Arachnotronic listened to the conference call, what does he think?

 

[witeken]

Anyway, here's the presentation PDF: http://intelacquiresaltera.transactionannouncement.com/wp-content/uploads/2015/06/Investor-Deck.pdf.

 

[Arachnotronic]

Arachnotronic listened to the conference call, what does he think?

I like the idea of Intel integrating FPGAs with Xeon processors; this is something that will be very difficult for competitors trying to enter the server market to match. If there is a market for such integrated chips (Krzanich indicated that customers have been asking for these kinds of chips), then Intel should be in a really good position longer-term

The bummer is that such parts are quite a ways off given the kinds of design cycles we're talking about here.

 

[ShintaiDK]

Well we should have known shouldnt we.

Skylake+FPGA.

 

[Arachnotronic]

Well we should have known shouldnt we.

Skylake+FPGA.

Skylake + FPGA will be co-packaged, AFAIK, so whether Intel owned Altera or not wouldn't affect this.

My guess is that Icelake Server will offer on-die FPGAs.

 

[Fjodor2001]

I wonder what Intel's intentions are with this deal. What kind of use do they envision having of the Altera FPGA tech in Intel's products?

 

[mrmt]

I wonder what Intel's intentions are with this deal. What kind of use do they envision having of the Altera FPGA tech in Intel's products?

Accelerators built tailor-made to certain workloads. Even more efficient than GPGPU.

 

[ShintaiDK]

I wonder what Intel's intentions are with this deal. What kind of use do they envision having of the Altera FPGA tech in Intel's products?

Its listed in the businessplan presentation.

 

[Fjodor2001]

Accelerators built tailor-made to certain workloads. Even more efficient than GPGPU.

Sure, but do they have to buy the complete company? They could probably had licensed the technology instead. Just like they should have licensed the iGPU tech from AMD or nVidia instead.

$17B is a helluva lot of money. I wonder if they really get get ROI on that. Seems like kind of a niche market to me only suitable for some specific workloads.

 

[ShintaiDK]

Its a market in huge growth. The outlook is 1/3rd of all cloud will use it in 2020. Plus with the manufactoring advance, they will also gain vs the competition.

 

[mrmt]

Sure, but do they have to buy the complete company? They could probably had licensed the technology instead. Just like they should have licensed the iGPU tech from AMD or nVidia instead.

$17B is a helluva lot of money. I wonder if they really get get ROI on that. Seems like kind of a niche market to me only suitable for some specific workloads.

These companies would never license this technology to a competitor, at least not in economically viable terms, so a kind of M&A deal is necessary (although it is highly questionable to cash out the old shareholders).