GDMS said: Intel [said] that they were an 800lb Gorilla that could not be trifled with.
I do not have any insider knowledge and this is pure speculation, but I suspect that it's the x86 division within Intel that's the 800lb Gorilla, and any up-start alternate architecture within Intel that the x86 division saw as a threat was stomped on, or else hamstrung so as not to compete with x86 -- which meant that it couldn't compete with anyone else's products either.
You see this a lot with big companies: the "cash cow" division sets the rules, and rule #1 is "no competition from elsewhere in the company". However, when the "cow" starts running dry, it means there's not a strong alternative to help out. I've read that there was a promising tablet group at Microsoft a decade or so ago, but the Windows fiefdom prevented it from getting anywhere, so Microsoft had nothing that could compete with iPad or Android.
You hit the nail on the head. This proc. (like the failed Settop box processors) is a distraction.
IXP is now owned by Neronome which uses Intel as its fab for its latest part. The new part acts as a co-processor to a Xeon using its ample PCIe bandwidth. And has max throughput of 400G. So why not just push this part ?
I still remember a meeting with Intel when they tried to strong arm me saying that they were an 800lb Gorilla that could not be trifled with. I pointed out that the Gorilla was an endangered species ! A lot of these mystifying decisions had hubris and arrogance behind them and hence do not stand up to rational scrutiny.
Interesting thoughts @GSMD...I never understood why Intel was aquiring new companies or technologies...or was quiting some products lines that were doing reasonably well...my theory is that if you are that big main product line and people running it just kills amd muddies everything else...right now the best move for Intel would be to start manufacturing ARM based devices for smart phone makers...you got to fill their very advanced and expensive fabs with something...just few comms processors will not do...Kris
1. Don't get me wrong, Intel will increase their market share. I just meant that this could have wiped out the competition in an earlier era. Wrt the IXP family, I still have no idea why they gave it up. I was one of the earliest design partners for that family and it was doing well. They gave it up for no rhyme or reason. Maybe because the core was ARM but they could have swapped that with a pentium M core back then.
2. I went thru the same thing with intel on Xscale annd their first attempt at being a merchant foundry (back in the mide 2000s). Could never figure out why Intel gave up on good ideas or promising businesses. Nobody I knew at Intel ever gave me a convincing answer.
3. I was just being a little facetious about giving up x86. But seriously they had at least 2 opportunities to launch a clean slate arch but never went that route. I am taking purely from a technical sense. Commercially it probably made sense to stick with x86. But then they had the i860 and i960 family. Go figure. The x86 decode penalty is not much these days but it does chew up dev time. But I am glad they stayed with x86, we made good money developing ref Xeon server MBs for them !
Seems just like yesterday that I used to develop 8085 systems. Time flies ...
These accelerators would have killed the market for other vendors a few years ago. I think a few things have changed now
1. Power matters and a Xeon plus accel. Combo is not very power efficient. Let us not even talk about PCB space issues. For comparison Freescale T4240 sips 30w typically, TDP is around 60-70w.
2. X86 compatibility is not really that big a deal in this space. May matter a bit, unlike mobile where it is irrelevant. But far less than Intel makes it out to be. Comm processor control processors also should not have deep pipeline stages. They are exception prone in the control plane and hence pipeline flushes can be common. So shorter pipelines are preferable.
3. Integrated comm. processors are much easier to work with, in terms of programming framework, ease of data sharing and work partitioning among components.
4. I guess in in scenarios where you have x86 servers and want to add security or comm functionality, this makes sense. Intel should never have sold the IXP line( or for that matter the Xscale line). Now they are recreating it. It would have made more sense to have gotten rid of the x86 product line !