Anyone have experience with Acer netbooks ?

[Lennart Sorensen]

On Tue, Oct 20, 2009 at 10:57:30PM -0400, D. Hugh Redelmeier wrote:
> The AMD 64-bit architecture supports the old-style FP.  The Linux and
> MS Windows ABIs have chosen not to do so.

And AMD recommended that too.  They have even said in the spec that
future x86-64 CPUs may drop the support if they drop 32bit support.
The x86-64 instruction set has x87 instructions listed as optional.
They are only required for x86 support.

> | This makes floating point much much faster, and avoids the awfullness
> | that is the x87 stack based FPU.
> 
> I've not seen why x87 stack-based FPU implementations had to be slow.

Because the stack means memory access which means slower than registers.
The CPUs try to fake it but you just can't do it as efficiently as if
the compiler has direct control over what goes in what register.

> Perhaps it is the NaN (signaling and non-signalling) and denorm
> support is expensive.  Is that gone from SSE?

Well dropping the stupid 80but floating point support is a bonus too.
Using the pretty much standard 64bit that everyone else has is just
better.  Fits in a more sensible memory size too.

> Actually, too many computer systems built by big-name manufacturers
> have top out at 4G just because 32-bit WinXP and Vista don't even
> manage to use all of that.  In other words, the requirements of a
> stupidly constrained OS encourage limited hardware.

That too.  4 or 6GB ram costs almost nothing these days.

> Luckily MS has declared 64-bit to be required for Windows 7 branding.
> They have held us back but are now kicking the industry forward.
>
> (I've been running x86-64 for 4 or 5 years.  Linux was ready rather
> quickly for the transition.)

Yes, about time too.  Linux users have been doing 64bit since forever.

> I think that MS has been a little better this time than for the 16->32
> bit transition.  The hacks for addressability that DOS and Windows
> users had to endure were horrible (users didn't seem to blame the
> perpetrator).

Well much to my surprise, there is still windows software out there using
16bit code which can't even run on 64bit windows (although it apparently
should run in XP mode on 64bit windows 7).

> PAE is standard practice, repeated each time memory grows beyond
> address space.  I remember bank switching on the PDP-8 when memory
> grew beyond 4K 12-bit words of core memory.  Unix as we know it grew
> for years on the PDP-11 with 16-bit addresses but a wider address bus.

That doesn't mean it isn't a bad solution compared to natively using a
larger address space.

> In fact PAE was designed long before most of us needed it which made
> the transition orderly.  It came out with the Pentium Pro!

Yes it did, and it still doesn't perform very well.

> Carefully crafted applications can use more than 2GB of memory on a
> PAE system.  They can use system calls to switch stuff in and out of
> the address space.  Some applications (eg. DB systems, OSes) are so
> hungry for memory that they will work hard to get it.

True, but at the same time on a 64bit OS they don't have to do that
making for simpler and faster code.

> | You are better off loosing 800MB on a 32bit OS out of 4GB than to turn
> | on PAE.  Much better.
> 
> Really?  The few real world reports of the performance hit for PAE
> haven't been extreme.

As far as I recall from reading about it, 10 to 15% overhead for PAE is
not unusual.

> Technical issues are one thing.  Business factors are key to
> understanding this.  It is important to realize that Intel has very
> good technical people and if they do something dumb technically, it is
> often for business reasons (not always).

Well they did make the pentium 4....  Just because marketing thinks it
is a good idea, doesn't mean it is.

> The following is "reverse engineered" logic: it may not be true, but
> it makes sense to me.
> 
> Intel wanted to push everyone into ia64 (Itanium) which they would own
> rather than x86 where they had to compete with other suppliers.  What
> better way than slow starvation of being limited to 32-bits in x86.

Of course, although even intel by now realizes that isn't going to
ever happen.

> There were strong rumours that Intel had designed a 64-bit x86 for the
> P4 generation.  But they suppressed it.  Further rumours suggested
> that this architecture was very close to the AMD64.

They were working on one, but microsoft told them that there would be
only one x86 64bit windows, and it would be the AMD design.

> AMD had nothing to lose and much to win with a 64-bit x86
> architecture.  They would have won even more if Microsoft had followed
> through with their promise to support the new architecture well and
> promptly.

Well 64bit xp betas were in use by many 3D graphics places very soon
after the 64bit AMDs came out.  It took a long time to get a final
release out though.

> The "bullet point" of supporting AMD 64 did cause a few sales.  And so
> did Linux support.  But all that was insignificant in the sea of PC
> sales.  They did make real inroads into the server world, but even
> that was slow because that world is conservative (for good reasons and
> bad).
> 
> Back to Atom:
> 
> Intel and Microsoft both feared the Atom might cannibalize their
> mainstream sales (they apparently actually have).  Each company
> crippled Atom systems in different ways to prevent this.
> 
> - Microsoft limits the sales of WinXP to netbooks with certain severe
>   limitations (screen size, RAM, and processor, I think).

I believe 10" screen and 1GB ram.  I don't remember a CPU limitation.

> - Intel will not sell bare Atoms to anyone who will not sign a license
>   that restricts on their use.  I don't remember all the limititations.
> 
> If anyone gets close to competing with Atom, the Atom will suddenly get more
> powerful or the agreements will be more liberal.
> 
> We already saw how Linux on netbooks forced MS to rethink the rules
> they imposed on their customers (the computer manufacturers).

If intel would try to be a bit more inovative rather than reactive,
they might actually do better.  There are people out there that want
efficient low power chips with good features.  Intel just doesn't seem
to want to make those because well they could cut into sales of high
end chips, except the high end chips don't fit the need.

To some extend the intel PXA and IXP lines are better CPUs for low power
use than the atom, except of course they are not x86 and intel hates
anything that isn't x86 (unless it is itanium).  They hate stuff that
they didn't invent.

-- 
Len Sorensen
--
The Toronto Linux Users Group.      Meetings: http://gtalug.org/
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://gtalug.org/wiki/Mailing_lists