back to article SmartNICs to make up 38% of network market by 2026

SmartNICs are expected to play a significant part in growing the Ethernet adapter market, which is forecast to reach $5 billion by 2026, according to research outfit Dell'Oro. Server connectivity is also expected to continue to migrate to higher speeds, with ports of 100Gbps and above accounting for nearly half of the …

  1. Duncan Macdonald Silver badge

    38 percent - maybe - it depends on how you count

    The smartNIC market could easily reach 38% by revenue - but smartNICs will not reach even 5% of the total number of NICs as the ultra low cost NICs on PC motherboards, tablets and phones vastly outnumber the number of server NICs.

    1. Roland6 Silver badge

      Re: 38 percent - maybe - it depends on how you count

      It has taken a long-time for the wheel to go full circle. Before the PC, smart NICs were the norm.

      1. Vometia has insomnia. Again. Bronze badge

        Re: 38 percent - maybe - it depends on how you count

        I'm reminded of the Motorola boxes we had. The MVME330 was the basic ethernet interface (10 Mbit AUI, this was in the late '80s) that most of them had installed, then along came the MVME374 which was all "new! Shiny! Now even faster, because it's intelligent, it has its own processor and everything!" (still 10 Mbit AUI, tho') and seemingly not that long afterwards was the amazing upgrade to the MVME376, "no longer limited by a puny on-board processor, the new interface can take full advantage of the raw power of the system CPU!" etc. I presume it wasn't just a rebadged MVME330 but nothing would surprise me.

        1. Roland6 Silver badge

          Re: 38 percent - maybe - it depends on how you count

          Funny that, some of the last smart NICs I saw and handled in the late 80's were from Motorola, forget the board numbers but they were for 802.3 and 802.4 and supported the MAP/TOP variant of OSI and fitted the Sun-3 pedestal.

          I think those last smart NICs had a faster CPU than Sun had used on the Sun-3 motherboard.

          1. Vometia has insomnia. Again. Bronze badge

            Re: 38 percent - maybe - it depends on how you count

            That may have been the MVME374... I remember some guys working on OSLAN interoperability though I wasn't even indirectly involved, I just remember all the swearing. :D Not sure if it's the same project that ultimately involved DeathNet Phase V, that may have been an entirely different source of swearing. 802.4 is new to me but probably as we didn't use it; I guess it might depend on which transition card was connected to the backplane (my assumption that it isn't AUI probably confirms that I never encountered it! I mean other than when I ended up with a 3174 terminal controller to connect a 3278 I never got round to acquiring to my Hercules System/370 emulator).

            I was slightly disingenuous with my reply in that the MVME374 was an upgrade to the MVME330 as used in the M680x0 systems (ours were mostly MVME141s with an M68030 at 25 and 33MHz) and the MVME376 was the no-longer-smart upgrade intended for the newer M88000 RISC boards: ours were MVME188 quads; the upgraded MVME197 had its own onboard LAN as with the other 1x7 boards but I dunno which type.

            That said, I wasn't entirely convinced that the M88K was as big a step-up from the M68K in terms of outright performance as some people claimed, though the fact that there were four of it certainly helped! The main drawback was software support, M68K being stuck on the by then very long-in-the-tooth SVR3 (i.e. no shared libs, no job control, no FFS/UFS) with M88K being required to upgrade to SVR4. Also memories of Motorola deciding it was too much trouble to write a driver for the MVME336, a contraption that connected to six 16-port departmental terminal servers over what was described to me in somewhat vague terms as high-speed X.25 links. "Described to me in vague terms" to set the scene for my manager asking me if I could write a driver myself. By the end of the week would be lovely. Bearing in mind I had zero experience of either X.25 or writing kernel drivers... that didn't stop him dumping another X.25 project in my lap, though the main problem wasn't my lack of knowledge but the vague and seemingly rather movable specification, such as it was. :|

            Of course the M88K boards were comparatively short-lived as they moved development to the PowerPC. I was at DEC by this time and suspect a significant part of the change in direction was the Alpha's appearance which completely pwned even the higher-spec M88Ks, or at least would've done if they weren't unobtainium: much of the stuff we used internally described as being Alpha-based to DEC partners tended to be overloaded Vax clusters...

            Anyway, I digress; if I will insist on posting at 3am... who needs sleep anyway?

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like