Paul A. Clayton (paaronclayton.delete@this.gmail.com) on February 27, 2019 6:06 pm wrote:
> Doug S (foo.delete@this.bar.bar) on February 27, 2019 3:04 pm wrote:
> > Paul A. Clayton (paaronclayton.delete@this.gmail.com) on February 27, 2019 1:32 pm wrote:
> > > An arms dealer selling mercenary services as well as effective weapons and training would not
> > > be that much more an economic threat to a mercenary company than one that selling just weapons
> > > and training when anyone can get good recruits. For ARM's licensees, fab technology is not
> > > a distinguishing factor ("anyone can get good recruits"). If ARM's design offerings become
> > > "good enough" (and complete enough), ARM effectively becomes an indirect competitor.
> >
> >
> > But ARM doesn't sell a complete SoC, they sell a CPU and a GPU. Modern SoCs have so much more than that,
> > and you cannot be successful with just those elements. That's like your hypothetical arms dealer selling
> > only AK47s and cartridges, while your mercenary companies can not only supply well trained soldiers but
> > everything from land mines to rocket launchers to helicopters armed with depleted uranium rounds.
>
> First, I did include "complete enough". Second, if ARM can license design components to produce a server/personal
> computer processor (which is the hardware competition of concern in this thread, I think), ARM could
> probably acquire the right to sub-license such design components and optimized hardenings of them,
> which could then be provided to a company making server/personal computer processors. ARM seems to
> want to become a one-stop shop for design components for a substantial portion of the semiconductor
> market. ARM also seems to want to make using ARM design components as easy as possible with low time-to-market
> and low risk (Processor Optimization Packs). While it is a significant extrapolation to go from these
> tendencies to ARM providing "Chip Optimization Packs", it does not seem more unreasonable than the
> technical issues with ARM producing a processor chip for sale.
>
> The economic impact to companies making ARM server/personal computer processor chips of ARM
> providing a ChOP for such a chip would not be that different than ARM developing such and using
> the ChOP to produce its own chips for sale as well as licensing the ChOP to others.
>
> I doubt ARM will be selling hardware any time soon (it already has enough concerns even with the recent
> influx of capital), but if it did I tend to agree with Michael S that "most of their customers either wouldn't
> object ... or object silently with no visible effects". It seems the options are RISC-V, an internal ISA,
> or ARM. RISC-V is still immature (it may dominate the "microcontroller" market before ARM has a substantial
> server presence, but I suspect it will be a while before it is as broadly used as ARM is now), and RISC-V
> is a threat to ARM regardless of how much ARM competes in server/personal computer hardware. Apple might
> adopt an internal ISA, but the other ARM customers seem to recognize a benefit from a shared software ecosystem
> (even though as a rising tide it raises all boats, i.e., helps competitors).
>
> If ARM customers continue to fail to provide good developer systems, ARM might be pushed into
> selling hardware. (At least that seems to be the theory behind the suggestion.) I am skeptical
> that ARM would develop a good personal computer/entry-level server processor chip and motherboard
> reference designs (that would probably be enough for multiple companies to make and sell systems),
> but ARM has announced plans for higher performance cores and better "uncore" designs.

It is also not beyond the wit of Arm to fabricate a design and only supply it on boards that (while reasonably priced compared to x86 for what they are) are massively overkill for most people - e.g. a 32t (either 32c or 16c/32t design) CPU running at 3.5 to 4.5 GHz with 4 DDR4 channels and 64 PCIe lanes from the CPU, supplied on a motherboard exposing all CPU memory channels to DIMM slots and all PCIe lanes to expansion card slots (M.2, PCIe x16 etc) at around $1,200 in one of the ATX form factors.

You could even make it customer-friendly by offering your customers access to the full design of this board - from motherboard schematics and PCB layouts down to how Arm puts together the SoC that they used on the board.

This is a reasonable price compared to an equivalent Intel or AMD system (it's about the same as you'd pay for a Threadripper 2950X at comparable clocks with a motherboard to use it all), it's something you can build a workstation around to get you an ARM desktop setup, and it's not directly competing with any Arm customers - and it's something they can start from and evolve into a server design, or a client design, or something more suited to real users.