Software stability and availability have a major impact on processor and networking decisions. There are probably billions of lines of code written for x86 for the data center. This is all running nicely on the Intel CPUs. Software and CPU architecture are huge. It's not just a process node issue on what wins and gets used.
It's a similar issue with software stacks for networking. The Broadcom software stack is extensive and completely debugged and working. Millions of lines of code running BGP under all complex scenarios. Networking is about the software too. All the L2 L3 switching and routing protocols.
Even though everyone lies about the process nodes. I think the industry clearly recognizes that Intel still has the lead in process technology. The TSMC 16 FF is closer to a real 20 nm solution.
Intel won't even hire you for a process R&D job unless you have a Ph.D. I think they have and will continue to have the clear lead in process technology. You really think they will let TSMC beat them in process R&D? I think they have a little pride over there.
It's quite interesting to note that Mellanox themselves use Intel x86 CPUs to drive the Spectrum 32 port Ethernet boxes. Why don't you put your money where your mouth is and move away from Intel x86 CPUs to power your Ethernet switch products?
I'm not sure the purpose of your post. You are in charge of marketing at mellanox so I assume there is some mellanox angle to it.
What is your market share in Ethernet switching? Is it even 0.1 % of shipping 25g ports? I dare you to tell the truth about that here.
The board of directorsof mellanox should hold Eyal accountable for this failed effort to get into Ethernet switching and compete with Broadcom. He has spent several years and millions of dollars to try this and made zero progress.
You make some loud claims about how superior your solution is but it never translates to any sales. I think Cavium xpliant has better luck than you with a big win at Arista..
Eyal has done one of the dumbest and most arrogant things ever to sabotage his own success. He makes, markets, and sells the complete switch product. Who in their right mind would want to use a switch silicon from someone who also makes the full box to compete against you?
In any case, it's getting harder, not easier to take on incumbents. The cost of 10 and then 7 nm products will reduce number of players that can even afford a tapeout.
I think semi industry is consolidating rapidly to just a few viable players who can succeed and make money. Intel for data center CPUs and Broadcom for Ethernet switching silicon. Niether one are standing still.
Tomahawk II has 64 ports of 100g switching muscle. That is huge.
Sranje: Thanks for the pointer to Linley analysis - they always do thorough work. Indeed I'm not sure who will be the leader in the next generation process nodes and cited the SemiWiki's opinion here.
What I am sure of is that Moore's Law is stretching out ... BK even confirmed this in an earnings call a few quarters back. So now you are seeing the tick-tock-tick model moving to a tick-tock-tock-tick model ...
So as the periods between Moore generations stetches out it will narrow the gap. So no matter who is first to a new process node, there will be a fast follower! And architecutres and ecosystems will become more important than ever.
Well I didn't make any predictions about Mellanox or our market share at all - but rather some big things I see that can impact the industry. But your predictions are certainly at odds with how I believe things will play out. As I said predictions are dangerous things as they make you accountable. It will be interesting to look at how reality plays out in 2017. It is always easies/safest to predict that the dominant players will remain dominant.
But I remember in the 2000 time frame thinking that the Microsoft operating system, browser, and office suite were so dominant in the enterprise that there was nothing that would derail them in my lifetime. But I certaintly didn't see the DotCom crash coming, the rise of Google, and then Apple, and Facebook, and the Amazon Cloud, the decline of the PC and rise of the mobile platform, and finally the re-incarnation of Microsoft's growth as a Cloud company. So I've become a bit more humble about my ability as a TechSeer than I was in my youth. There are things just beyond the horizon that I can't envision. The only thing I'm very confident of is that the status quo won't remain, and there will be dramatic changes in the future. If only I had your confidence and knew the future as precisely as you appear to!
Mellanox will continue to go completely unused as an Ethernet switch vendor. Broadcom Tomahawk II will continue with the complete dominance in Ethernet swtiching. I'm sure there is a 56G PAM4 switch coming soon. Nobody wants to compete against their switch chip vendor for the box sale too.
I thought this Open Ethernet had died. The truth is that Broadcom provides full robust software stack for these "open" switches.
Intel will increase the dominance in the data center, not reduce. Intel has over 95% market share in data center processor. That's not because of process technology alone. A huge part of that is simply x86 software stability and robustness. The data center can't crash hourly.
Open Power, really? I'll believe it when I see it. Power9 is probably going to be the last Power processor ever by IBM. The CEO of IBM wants to get rid of all hardware. That team has to be max expensive. Nobody used Power8 and nobody will use Power9. It costs quite a bit to fund a processor team for a processor that nobody uses.
ARM? I think we just saw how much traction that is getting. Paramesh Gopi is looking for a new job because of his big bet on ARM servers.
And lastly, Mellanox will get bought. They are a little guy and little guys are getting gobbled up daily. That market cap of 2B puts the MLNX right in the cross hairs of a major Tier 1 Semi company.