Merchant Silicon and Vendor Software – The Hype in 2012

I’ve blogged a few times this year about the difference between merchant and custom silicon and had several discussions on Packet Pushers about the value of merchant silicon.

But how much merchant silicon is there TODAY. Check out this graphic from Broadcom who makes the Trident network processor at the heart of all these switches :

Broadcom merchant silicon 1

Yep, all of the major switch vendors are using the same switch chipset from Broadcom to make switches that are basically the same. All of them. Even Cisco, home of the “we invest in R&D and continue to make our own silicon”1 has a Broadcom Trident chipset in their Cisco Nexus 3000 product (because their own silicon is high latency and high cost).

Not Server / Intel monoculture

It’s possible that we might end up with a single dominant silicon vendor for networking in the same way that Intel x86 CPUs in the server market. At first, you might think this isn’t a bad thing. But if you are DELL, or HP, or IBM you might wish that you could build a product that offered better features than Intel can make. When your only option is to glue Intel chips to a board it’s hard to deliver new features that could give you an advantage over a whitebox vendor from East China/Taiwan (like Facebook does).

And Intel has been dominating that market by moving into the support chipsets too. Once, vendors would choose north bridge and south bridge chipsets that connected the CPU to memory, graphics, USB, keyboard peripherals, but today all chipsets come from Intel. Every motherboard is basically the same. That’s a big reason why we have blade servers – server vendors are desperate to make products that are actually different from each other.

Customers like the “idea” of a fast car, but most people by the family sedan. The

The EtherealMind View

Now the vendors have shown a lack of business capability to make their own silicon, their products will lack differentiation in the hardware, so we will have a LOT of marketing about software. Be ready for it.

Cisco is still attempting to develop their silicon, and might turn this into a competitive advantage (as it has done for the last ten years). On the other hand, Cisco is years late on delivering their latest product lines based on new silicon to market. This suggests that there are internal problems at Cisco in maintaining product development and profitable investments. Once upon a time, Cisco got everything right in their products and design, but recently Cisco has failed at so many products that it’s hard to be confident that they still have their mojo.

What’s interesting is that software driven networking is likely to be big this year in repsonse to the virtualisation dynamic. Whether the system is OpenFlow/SDN or NETCONF/XML or some other software platform isn’t yet clear. But I bet the network operations will look very different by the end of 2012 – and one reason will be that most vendors will have the same network hardware.

Never thought in 2002 that I would say that in 2012.


  1. John Chambers – in every keynote for the last three years at least. Seems to be standard boilerplate.
  • http://twitter.com/_AshishShah Ashish Shah

    Greg,

    It means that the basic switching technology has been commoditized and is good enough. It’s allowing new-comers to invade the incumbents’ turfs. However, I don’t think this takes away from developing specialized silicon. It’s just that basic switching is not that space. 

    Also, merchant silicon based switching solution is allowing adjacent product vendors to build networking class products. 

    Ashish

    • Ryan Malayter

      It’s not just “good enough”. Frankly, from a specs *and* price/performance standpoint the Trident+ kicks the crap out of all the other switching silicon available, no matter the vendor. Yes, even the collection of space-heaters in your Nexus 7k.

      The Trident+ has 64 ports of 10G, ultra-low latency, 32-way ECMP, with TRILL support, all on a single chip. Nothing else I’ve heard or read about being in actual production even comes close. There’s a reason so many vendors are building switches with Trindent+, and it is the same reason that so many server vendors ship Xeons – they just beat the pants off the rest of the market.

      • http://twitter.com/the_socialist Jon Hudson

        Here’s the thing I wonder. While right now you are right. The Trident+ is _way_ impressive, this is a leapfrog game. 

        Broadcom, Marvell and Fulcrum all have great teams busting out killer chips each a little ahead of the other each round. 

        Ah…but now Fulcrum isn’t Fulcrum. Fulcrum is Intel. INTEL.

        So what is to keep Intel from doing to this market what they did to the x86 market and say 5yrs from now all fixed port switches have “Intel Inside You Can’t Hide” stickers on them?

        • Ryan Malayter

          That might not be a bad thing. People forget how much AMD actually innovated with Opteron, forcing intel into copying and changing course. A lot of AMD boxes were sold for a few years, because they were simply better. Intel caught up and then flew past AMD. Nothing wrong with that – that’s the basis of a healthy marketplace.

          Intel has process advantages, but Broadcom can certainly lead or play leapfrog for a long while. And they seem to be a bit more savvy at scale production than AMD, and will probably figure out a way to make money in the process. They seem to compete just fine with intel in other areas.

  • http://www.crenacransconsulting.com/ Natosach Nua

    This is also reflected with the company behind both the ASR Quantumflow processor and the Juniper ISG processors among others. Tensilica Xtensa processors are used in both instances.

  • Jon Hudson

    I know we don’t have huge market share. And I’m not saying we use _all_ our own chips in _all_ of our products.

    However, we are an ASIC company. And we’re _damn_ good at it.

    Want kick a$$ hardware? buy Brocade =)

    • Jon Hudson

      My apologies for the unabashed self promotion…mostly ;-)

  • http://twitter.com/qxam Jason Braddy

    Just a minor point: the Arista 7050-series switches are based on the newer Trident+ chipset while at least some of the others are using the older Trident ASICs. I believe that some of the switches (the 7050S at least) also use a PHY-less design – while others don’t – which can shave another 100ns or so off the latency and reduce power consumption somewhat.

    That being said, I agree with your central thesis that software (and how that software allows the products to be leveraged as part of a larger solution) will definitely play a much larger role in differentiating the products in this space going forward.

    • http://twitter.com/thomas0002 Thomas Scheibe

      phy-less or integrated phy?

      • http://twitter.com/qxam Jason Braddy

        I’ve always assumed that “phy-less” was just shorthand for “phy-as-a-discrete-chip-less”…something still has to handle L1, right?

  • Will

    Wow.  Would love to see a sub-picture showing the price differences.

  • http://twitter.com/cloudtoad Derick Winkworth

    This depresses me for some reason.  The software will provide the differentiation but still you can’t overcome the fact that fundamentally you are dealing with the same hardware.  Instead of RFI/RFP, why don’t I spin a roulette wheel and make my decision?  NOS software sucks these days, sorry guys… they’re all riddled with bugs.

    • Anonymous

      Perhaps there will be an open-source switching operating system in the future. It would require network cards with many ports though. It’s probably possible today, but no one has tried it, as fast switching is reliant on logic embedded in the asic hardware.
      Imagine a commodity switch interface with the same command line options everywhere!

  • http://blogstu.wordpress.com stu

    Greg – we’ve seen the industry move to a majority of designs with merchant silicon (http://wikibon.org/blog/commoditization-of-ethernet-switches-how-value-is-flowing-into-silicon/), but as @twitter-24077995:disqus  points out, there are multiple vendors providing these chips. Intel’s acquisition of Fulcrum Micro was a big move, not only do IBM, Dell, HP and Cisco have big partnerships with Intel, but even Arista has old ties there from the management team. I agree with your point about software (especially marketing) being a big point for 2012 – but will this help or hurt the contenders for Cisco’s crown?

  • Birenpatel9275

    Now the vendors have shown a lack of business capability to make their
    own silicon, their products will lack differentiation in the hardware,
    so we will have a LOT of marketing about software. Be ready for it.

  • Pingback: Network, Interrupted

  • http://twitter.com/tristanbob Tristan Rhodes

    Just as a follow-up, does anyone remember if Chambers mentioned Cisco silicon during his Cisco 2012 keynote in San Diego?

  • Pingback: Is The OS Relevant Anymore? « The Data Center Overlords

  • Jz12316

    What is the part # for the Trident /+ on the Broadcom web site ? I see a lot of 10gbps Switch devices and am not sure which this is

  • Pingback: Public Clouds and Network as a Service

  • Pingback: IETF Draft - BGP-signaled end-system IP/VPNs

  • Pingback: Plexxi Capacity Planning Done Right

  • http://www.facebook.com/john.summers John Summers

    This topic is exactly why Mellanox went into the Ethernet switch silicon business a
    few years back. Users need a range of competitive solutions that provide them
    new and differentiated features that add value to their business, both on
    productivity and economics.