Subject: Re: *this* would make a very nice NetBSD machine
To: Alex Pelts <alexp@broadcom.com>
From: Garrett D'Amore <garrett_damore@tadpole.com>
List: port-mips
Date: 08/14/2006 16:14:18
Alex Pelts wrote:
> I mostly agree with your argument except one thing - even though geode
> and alike can shrink their die size, mips will do so as well with
> moving to a new technology. It is all about the costs in embedded
> market. So if say mips will be 2mm on die and geode 4mm, mips still wins.
> As far as holding on to low end I think you are not absolutely correct
> here. I can give many examples - PIC being on extremely low end, 8051,
> Z80, etc. These processors were expensive and large at some point but
> now they are tiny and very wide spread. ARM is a low end solution and
> has been expanding its presence for a very long time. I am not sure
> why set tops not using ARM but somehow they are not at least not in
> big numbers.

The problem is that the folks who are spending big on new technology
investments aren't investing in MIPS.  AMD have sold off their MIPS
product line and are pushing Geode hard now.  Intel has XScale/ARM
products.  MIPS will still be used because the core is easily licensed
into other architectures, but this is true for ARM as well.

So I expect the embedded market to consolidate over the next few years
on ARM and x86 variants. The other players are just too much in the
fringe.  Plus, even though a lot of folks have experience with MIPS,
even more folks have x86 experience.  Right now the biggest hindrance to
x86 adoption in the embedded world is power consumption.  Products like
Geode are targetting this really hard.

Once you get below a certain threshold (5W?) the power consumption
differences don't make much difference.  Manufactures will simply go to
the platforms that offer them the best per-unit cost, and it will be
impossible to beat the economies of scale that the larger players can offer.


>
> MIPS64 is not very useful on low end, there is not much memory and you
> don't need much horse power there. It just takes extra die space
> without any measurable benefits. That is why you don't see it much. As
> far as multi core goes Broadcom has multi core chips that are in the
> routers and some other networking products.
>
> At the end it is all decided by price/performance ratio or !/$.

Yes.  But I like the MIPS ISA, and would like to see some truly
competitive hardware out there -- stuff for servers and power desktops. 
These need a good MIPS64 platform, and this is why OCTEON  (the x16
processor) and some of Broadcom's stuff is interesting.

These platforms also need a good OS, which is one of the reasons why I'm
intrigued by the idea of porting Solaris to the platform.

    -- Garrett
>
> Regards,
> Alex
>
>
>
> Garrett D'Amore wrote:
>> Alex Pelts wrote:
>>> Why do you think mips is dying architecture? It holds probably 50% of
>>> embedded market. I think most of the set top box market is mips, and
>>> probably 50% of home router/gateway.
>>
>> ARM/Xscale and now low power Geode/x86ish solutions seem to be
>> encroaching very heavily into this place.  I would be surprised if 50%
>> of _new_ designs were still MIPS.  Certainly there is still a lot of
>> movement of MIPS product, but I think it is almost exclusively at the
>> low-end, typically systems that are < 500MHz single core MIPS32.  I've
>> not seen any real movement in the MIPS64 space.
>>
>> This new product is a higher-end product, and shows some promise.
>>
>> (MIPS -- or any architecture -- can't hold onto just the low-end --
>> because eventually process improvements and cost-reductions push
>> formerly high-end products into the low-end.  So there has to be
>> innovation and getting into higher-end products is a good way to ensure
>> that innovation eventually trickles down into the lower end market
>> segments.)
>>
>>     -- Garrett
>>> Regards,
>>> Alex
>>>
>>>
>>> Garrett D'Amore wrote:
>>>> Timo Schoeler wrote:
>>>>> http://www.movidis.com/products/rev.asp
>>>>>
>>>>> 16 core MIPS CPU, ECC RAM, very low power design.
>>>>>
>>>>> or wait for PA Semi and their PowerPC-based PWFficient? :)
>>>>>
>>>>> cheers,
>>>>>
>>>> I want one. :-)
>>>>
>>>> Just when you think MIPS is a dying architecture, news like this comes
>>>> out...
>>>>
>>
>>


-- 
Garrett D'Amore, Principal Software Engineer
Tadpole Computer / Computing Technologies Division,
General Dynamics C4 Systems
http://www.tadpolecomputer.com/
Phone: 951 325-2134  Fax: 951 325-2191