ARM as a primary architecture
przemek.klosowski at nist.gov
Thu Mar 22 13:11:12 UTC 2012
On 03/22/2012 02:31 AM, Kevin Kofler wrote:
> Tom Lane wrote:
>> That opinion is flat out ridiculous. Or maybe it makes sense if you
>> think consumer desktops are the be-all and end-all; but they are not.
> Consumer desktops and notebooks. The things we normally call "computers".
> Those have always been and should remain our primary target.
Check out the numbers from The Economist:
The number of desktops has been flat for last 7 years. The growth in the
smartphone/tablet area dwarfs 'what we normally call computers'.
The whole article is at http://www.economist.com/node/21531109
> Wow, you seem to really hate x86. But please accept the fact that, no matter
> how much you hate x86, it is THE relevant architecture right now.
ARM claims an installed base of 25 billion CPUS, and current deployment
rate of 6 billion new CPUs per year. Granted, a large fraction of that
are microcontrollers (single purpose, no virtual memory etc), but still,
the total number of x86 machines is about 500 million.
> Nonsense. As long as we have healthy secondary architectures, we can promote
> them to primary when it really makes sense, i.e. when/if the x86 apocalypse
> happens. (By the way, don't count on it happening at all. The end of x86 has
> been proclaimed so many times, yet each of the prophecies has been proven
> wrong so far. Not even the move to 64-bit and Intel putting its weight
> behind a non-x86 64-bit architecture were able to kill x86. What was killed
> instead was the Itanium, dubbed "Itanic" by many. Don't be too quick to
> write off x86!) There is no need whatsoever to make the move to primary now.
I had an interesting discussion with VIA---asked them why don't they
compete with Intel's inexpensive Atom platforms. They weren't interested
in the low end; they claimed that there's plenty of demand for
traditional 'few hundred bucks' price point. I think they are wrong
about how sustainable that demand is---ARM has low-end CPUs for >1$ (ST,
NXP, TI), and the OS-capable ones for under 10$.
Now, does it mean that we need to rush the ARM primary architecture? Of
course not--as others have said, one gets the job because one can do the
job, but we need to figure out the details of how to get there:
- toolchain/build environment speed
- better and/or standard installation mechanism
- QA techniques
More information about the devel