dmcq (dmcq.delete@this.fano.co.uk) on February 27, 2019 4:49 am wrote:
> Michael S (already5chosen.delete@this.yahoo.com) on February 27, 2019 2:53 am wrote:
[snip]
>> If Arm returns to chip-making business and then to boards-making and computer-making business
>> then most of their customers either wouldn't object (Apple, Samsung, STMicro, NXP, probably
>> Hisilicon) or object silently with no visible effects (Qualcomm, Mediatek).
>> On the other hand, Arm's software partners would LOVE it.

[snip]
> It is true there was no recognizable IP model until a couple of years
> after the founding but no competition and being an IP company was built in from the start. Before the founding
> and for a while afterwards there were numerous talks with other companies about what sort of business they
> could make a success of. I'm surprised how helpful companies can be to each other in things like that. Perhaps
> that also contributed to their policy of forming an ecosystem and working with partners.

Competing with one's customers is difficult to manage, but it is not without benefits. As already mentioned, ARM already competes with those licensing its ISA. Currently, this competition seems to be limited in that ARM does not design special purpose processors (except GPUs, which are still high-ish volume) and has not provided high-end processor designs. The former seems likely to remain free from direct competition by ARM, but those licensing processor designs from ARM are likely pressing for higher performance designs. Making it easy to assemble a competitive product is in ARM's interest.

In some ways, Intel had a similar issue. Intel provided chipsets and reference motherboard designs which (combined with increased integration) shrank the x86 motherboard ecosystem. System vendors, already substantially focused on price, mostly benefited from this change.

In addition to RTL designs, ARM also offers Processor Optimization Packs which provide a hardened (and tuned) "implementation". With the importance of tighter design integration for optimizing power-performance-area (as seen in recent designs having less configurable cache capacities), designs that are close to being complete product designs seem likely in the near future (comparable to Intel's reference motherboard designs?).

Once a design offering is very close to a product design, it might not be inconceivable that a potential customer might want to add some small special purpose hardware to such a reference design and produce a low-ish volume product. By having ARM manufacture the product and sell parts with the extras fused off or at an extra cost that goes to the designer of the special purpose hardware, the potential customer could get a better price on the hardware.

This would not be the same as producing personal computer/server processors, but it hints that making hardware might not be inconceivable.

If (when?) anyone can trivially assemble ARM design components, ARM will be hurting the profit margins of those licensing its designs. When a MediaTek chip that just assembles ARM design components is nearly as good as a Qualcomm product, it would become difficult to justify extra design effort from Qualcomm and higher prices (to fund such effort and increase profit).

Standardization of interfaces (e.g., Server Base Boot Requirements) tends to commoditize products, which is bad for profit margins but good for end-users and can actually grow the market in volume even as diversity (in both bad and good forms) decreases. ARM seems philosophically linked to commoditization as a business model.

If ARM produced a good laptop/entry-level server chip (and reference motherboard designs), it might prevent companies like MediaTek from doing this (but MediaTek has many other ARM ISA-based markets and could still sell two- or four-core chips effectively based on the same four- or eight-core design used in the ARM hardware) and it might force serious personal computer/server ARM hardware companies to either adopt ARM hardware (adding value in areas outside the processor chip) or produce processor chips that add significant value (greater core count, better I/O integration). Having good "mid-level" processor chips would presumably help increase the size of the market, increasing revenue for bottom feeders and for the higher end and system vendors.

If ARM does not make it easy to produce such a chip, it will be violating its orientation toward making chip design easier and lower risk (and offering "good enough" designs). If it does make such easy (and the designs are "good enough"), the difference between MediaTek (or a joint venture) running off a few million chips and ARM itself doing so would not seem to be all that great (though such seems to assume semi-rational corporate behavior).

MIPS and Itanium both failed to succeed in the merchant chip with "internal" primary customer. MIPS suffered from the ease of producing an alternative ("anyone" could produce a comparable performance RISC and the NRE/volume tradeoffs did not seem prohibitive) and probably other factors (like litigation to wall-off the ISA). Itanium suffered from excessive hype, weak execution, problematic ISA choice, and excessive influence of the primary ("internal") customer. However, it is not clear that eating your own dog food is bad business practice even if you are the only or primary initial customer. (A veterinary chain might develop its own dog food, sell it at its own locations and still be able to sell the dog food to pet supply stores and other veterinarians. Pet supply stores have more diverse products (though they also often offer veterinary services), perhaps comparable to specialized design additions for licensers of ARM designs, and other veterinarians have location-based lock-in, perhaps comparable to channel and brand factors. This is not a perfect analogy, but it might hint that ARM competing with its customers might not be irrational business practice.)