By: Paul A. Clayton (paaronclayton.delete@this.gmail.com), August 10, 2014 9:23 pm
Room: Moderated Discussions
Ronald Maas (rmaas.delete@this.wiwo.nl) on August 10, 2014 7:12 pm wrote:
[snip]
> But most revenue is not made in the high end. So in order to successfully compete against
> Intel, ARMv8 systems needs to be able to provide comparable performance against Xeons
> in the USD 500-1000 segment at a lower TCO. And that, I believe, is achievable.
¤1 trillion in revenue in highly competitive lower performance parts of the processor market may be less helpful in supporting processor design effort than ¤100 billion in high margin, high performance processors. This is prominent in binning but presumably also applies to cross-vendor competition. Revenue from ARM microcontrollers, e.g., is unlikely to help much in supporting R&D for servers. (I suspect microcontrollers are a highly competitive market segment with substantial revenue even though the per unit revenue is tiny. The revenue from higher end, e.g., smart phone, processors may be higher.)
There is also a marketing benefit to having the "best" product (part of the importance of having the highest benchmark scores even if they are only modestly better and have much worse performance per monetary unit).
The diversity of ARM seems to make difficult the estimation of how much R&D can be funded, especially when comparing to Intel (with its integrated design and manufacturing). The inclusion of diversely-sourced design components in ARM SoCs also increases the difficulty of estimating how much of the revenue can be used for funding processor development. (Sadly, in my opinion, much of the R&D effort will be duplicated. I would not be surprised if ARM, Ltd., discourages or even prohibits resale of custom designs, and I suspect few companies are inclined to share even hardened (process-specific) designs with potential competitors and few companies would embrace funding a potential competitor. Obviously, setting up a design licensing operation would not be free, so there is that additional barrier.)
As I wrote before, starvation from inadequate farmland from a divided inheritance seems more likely than death from a thousand paper cuts.
[snip]
> But most revenue is not made in the high end. So in order to successfully compete against
> Intel, ARMv8 systems needs to be able to provide comparable performance against Xeons
> in the USD 500-1000 segment at a lower TCO. And that, I believe, is achievable.
¤1 trillion in revenue in highly competitive lower performance parts of the processor market may be less helpful in supporting processor design effort than ¤100 billion in high margin, high performance processors. This is prominent in binning but presumably also applies to cross-vendor competition. Revenue from ARM microcontrollers, e.g., is unlikely to help much in supporting R&D for servers. (I suspect microcontrollers are a highly competitive market segment with substantial revenue even though the per unit revenue is tiny. The revenue from higher end, e.g., smart phone, processors may be higher.)
There is also a marketing benefit to having the "best" product (part of the importance of having the highest benchmark scores even if they are only modestly better and have much worse performance per monetary unit).
The diversity of ARM seems to make difficult the estimation of how much R&D can be funded, especially when comparing to Intel (with its integrated design and manufacturing). The inclusion of diversely-sourced design components in ARM SoCs also increases the difficulty of estimating how much of the revenue can be used for funding processor development. (Sadly, in my opinion, much of the R&D effort will be duplicated. I would not be surprised if ARM, Ltd., discourages or even prohibits resale of custom designs, and I suspect few companies are inclined to share even hardened (process-specific) designs with potential competitors and few companies would embrace funding a potential competitor. Obviously, setting up a design licensing operation would not be free, so there is that additional barrier.)
As I wrote before, starvation from inadequate farmland from a divided inheritance seems more likely than death from a thousand paper cuts.
