Should Apple start making servers?

By: Alberto (, October 7, 2018 6:07 am
Room: Moderated Discussions
dmcq ( on October 7, 2018 3:33 am wrote:
> Gabriele Svelto ( on October 7, 2018 1:46 am wrote:
> > Maynard Handley ( on October 6, 2018 6:58 pm wrote:
> > > The two articles feature the same sorts of graphs. I would
> > > assume the same program and methodology was used.
> > > But of course, at the end of the day WTF does it matter? Are we STILL going
> > > to insist on this grasping at straws "of course Apple can't match Intel,
> > > therefore [insert bullshit excuse] must somehow explain the results"?
> > >
> > > The points that matter are thing like
> > >
> > > - Apple's core used to be clearly inferior to INTC, then became comparable, then became
> > > clearly superior (certainly at power and IPC, PERHAPS even in peak performance if given
> > > the power and thermals; though people can still insist, if they want to, that somehow
> > > THAT is the magic piece that Apple couldn't implement as soon as they want to)
> > >
> > > - Apple's uncore started off good and just kept growing, to the point that
> > > really the only way in which was inferior to Intel was DRAM latency.
> > > This release feels somewhat like the A8 which was another release where, in retrospect after we'd
> > > had time to look at everything, it felt like THE stellar work was actually in the uncore (and really
> > > underplayed by Apple, perhaps because that's very difficult to explain to the public).
> > > And so that "good uncore except DRAM latency" now seems to
> > > be closer to "superb uncore, including DRAM latency".
> > >
> > > Of course there are aspects of a server SoC where Apple has clearly shipped nothing
> > > relevant, or we don't know the performance of what they have shipped.
> > > Their NoC clearly has many clients, but does it do well with large amounts of unstructured traffic?
> > > How well do their cores handle locks and inter-core messaging?
> > > How expensive are interrupts?
> > > Hell, they haven't actually even yet shipped a design with more than one memory controller. (While
> > > the A8X clearly [IMHO] threw in three cores as an experiment/proof of concept, to validate all the
> > > various uncore improvements [we all know that the hardest improvement is 1 to 2 cores, but 2 to more
> > > than 2 cores is also not trivial because the world becomes more complex than "me and that other guy"],
> > > that same A8X and the high end iPads all appear to have a WIDER path to DRAM [128 bits rather than
> > > 64], but beyond that the same single memory controller, not dual memory controllers.
> > >
> > > It's reasonable to state that these pieces have not yet shipped. Going beyond that to suggest that it's
> > > impossible/difficult/economically nonsensical for Apple to create them is downright moronic. It's likewise
> > > more an indicator of one's emotional state of mind than of one's rational faculties to make a big deal
> > > of 5% issues here and there. If a latency (or bandwidth, or SPEC value) is 5% off here or there as a
> > > result of slightly different testing methodology, WTF CARES? Why is that important?
> > > It's like being told that you won a million dollars in the lottery and quibbling about whether that's
> > > a million 2018 dollars or a million dollars deflated over 10 years --- it shows a certain blindness to
> > > what's IMPORTANT as opposed to technical details that are simply not relevant to the big picture.
> >
> > I made a very specific claim, that server cores usually have lower single-threaded integer latency than
> > their client counterparts which is easy to prove and rather obvious to anyone with the slightest bit of
> > technical knowledge. Scaling has its costs, it's not magical. Since the original poster directly compared
> > to the score of a Xeon I pointed that out. And that's the only thing I said, nothing else. I never made
> > the claim that Apple could not design a server processor. So I don't know what's happening in your head
> > (and in that of some other notorious posters that tend to have knee-jerk reactions that are even worse
> > than yours) but I find it rather childish that you call me a moron for a claim I never made.
> >
> > Oh, and for the matter, with the money they have now Apple could design
> > a server processor, a car, a fighter plane or a rocket if they wished to.
> I'd have thought the main question would be whether they'd make a lot a money from it. I
> don't think it would conflict badly with their main markets and there would be some internal
> gains. The ARM server market hasn't taken off yet but I think all the bits are now in place
> so maybe they will do something. It is possible they intend to only do large integrated systems
> rather than supplying components and that would require quite a bit of setup..

A good large integrated system is something Apple can not even touch, they have not staff dedicated to this. They have to fight a good number of companies with decades of experience
and thousands of very complex patents about the topic.

They can license a great core to others, only this or they can purchase a company with the right knowledge.
But honestly speaking the phone market give more billions and market capitalization of any other business. Who cares about a relatively small revenue from an hardcore segment when is so easy sell expensive pieces of Aluminum to joung illetterates??

Never change mission, the share could tank.
< Previous Post in ThreadNext Post in Thread >
TopicPosted ByDate
Should Apple start making servers?Doug S2018/10/05 11:48 PM
  Should Apple start making servers?Adrian2018/10/06 12:16 AM
    Should Apple start making servers?juanrga2018/10/06 12:37 PM
      Should Apple start making servers?Alberto2018/10/07 09:52 AM
  Should Apple start making servers?Andrei Frumusanu2018/10/06 12:23 AM
    Should Apple start making servers?Doug S2018/10/06 09:25 AM
      Should Apple start making servers?Maynard Handley2018/10/06 09:50 AM
        Good point (NT)Gabriele Svelto2018/10/06 02:57 PM
          Good pointBlaine2018/10/07 11:14 AM
        Probably notDavid Kanter2018/10/07 10:44 PM
          Probably notMichael S2018/10/08 12:58 AM
          Probably notMaynard Handley2018/10/08 03:52 PM
  Should Apple start making servers?Gabriele Svelto2018/10/06 05:44 AM
    Should Apple start making servers?Beastian2018/10/06 08:39 AM
    Should Apple start making servers?Doug S2018/10/06 09:29 AM
      Should Apple start making servers?Travis Downs2018/10/06 10:23 PM
        Should Apple start making servers?Doug S2018/10/08 03:08 PM
    Should Apple start making servers?Maynard Handley2018/10/06 10:04 AM
      Should Apple start making servers?Gabriele Svelto2018/10/06 02:56 PM
        Should Apple start making servers?Maynard Handley2018/10/06 06:58 PM
          Should Apple start making servers?Gabriele Svelto2018/10/07 01:46 AM
            Should Apple start making servers?dmcq2018/10/07 03:33 AM
              Should Apple start making servers?Alberto2018/10/07 06:07 AM
                Should Apple start making servers?Doug S2018/10/08 03:19 PM
              Should Apple start making servers?Doug S2018/10/08 03:17 PM
    Should Apple start making servers?juanrga2018/10/06 12:33 PM
  Should Apple start making servers?wumpus2018/10/06 08:27 AM
    Should Apple start making servers?Aaron Spink2018/10/06 11:17 PM
      Should Apple start making servers?Doug S2018/10/08 03:33 PM
  Should Apple start making servers?Alberto2018/10/07 10:36 AM
  Should Apple start making servers?Michael S2018/10/08 04:05 AM
    Should Apple start making servers?anon2018/10/08 06:35 AM
  Apple's winning strategy in serversAM2018/10/08 07:17 AM
Reply to this Topic
Body: No Text
How do you spell purple?