[Arm-netbook] The future of EOMA-68

[Christopher Havel]

On Sat, May 2, 2015 at 1:44 PM, Luke Kenneth Casson Leighton <lkcl at lkcl.net>
wrote:

> > due to the way the Crusoe works (it's not natively an x86 CPU but
> > soft-emulates one) it's basically the lowest rung on the ladder, bar
> none.
>
>  i really liked the transmeta idea.  much lower power (higher
> performance/watt ratio).  pity they couldn't continue.
>

*shudders violently*

The Crusoe makes a Geode GX500 look like the i5-540M in my Dell laptop.
(I've used a GX500 based system. It was a truly awful experience that I do
not intend to repeat.) The Crusoe is the Corvair of the CPU world, minus
the book by Ralph Nader -- if the Corvair in question is running on three
cylinders instead of six! Poorly thought out, to a fault, slower than a
dead sloth in quicksand -- and it shows. Transmeta's designs are actually
*worse* than Cyrix's trash -- in my mind, at least. (Cyrix would've been at
least decent if their in-house ALU design hadn't been nearly so pathetic.)

The problem, of course, is that it soft-emulates the entire processor
architecture it's supposed to be a part of -- so you get a minimum 50%
performance penalty just by way of operation -- and that estimation only
works if the instruction translation is perfect and takes exactly as long
as each x86 instruction it has to translate. IRL it's going to be
tremendously worse -- although I don't have exact figures. (Probably just
as well.)

Don't get me wrong, innovation is cool, and the idea of the Crusoe is
indeed innovative -- but innovation isn't everything, to put it mildly.

That one's an Edsel.


>  don't the latest loongson's have hard-emulation of the top 200 most
> common x86 instructions, meaning that they can emulate x86 code on a
> MIPS-based processor at 70% of the speed of an equivalent x86
> processor?


Afraid I can't speak to that. I don't follow Longsoon stuff -- or, really,
much of anything in the CPU/SoC realm, these days. Too much going on to
follow, really -- particularly since I either don't care about it or can't
afford it. That said, a ~30% performance penalty is going to be noticeable,
I would think.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.phcomp.co.uk/pipermail/arm-netbook/attachments/20150502/77d79a8a/attachment.html>