[X4U] Re: move to Intel

[David Ledger]

When the list gets busy like this it's hard to keep up.

>Date: Tue, 7 Jun 2005 10:56:35 -0700
>From: Robert Ameeti <Robert at Ameeti.net>
>Subject: Re: [X4U] Couple questions on Apples move to Intel
>
>At 10:39 AM -0700, 6/7/05, James Jones wrote:
>
>>Yes, they do. NewWorld Macs still have hardware-specific start-up
>>code in Boot ROM
>
>Oh? What chip on the G4 motherboard is the ROM?

All computers I have met have either front panel programming 
switches, a control processor (with its own ROM) or a real ROM (be 
that chip ROM, a diode matrix or hand-knitted core). How else does it 
start running?

- - - - - - - -

>From: Richard Gilmore <rgilmor at uwo.ca>
>I just got a G5 and I personally do
>not want to go back to a 32 bit machine.

64 bit machines are usually slower than their 32 bit counterparts, 
but they handle huge dataspaces. If you don't use > 4GB address space 
you don't gain anything, but all those extra bits have to 
sucked/squirted serially from and two the disc and other devices.

- - - - - - - -

>From: "William H. Magill" <magill at mcgillsociety.org>
>
>The original IA64 was a disaster for Intel. It never achieved any of
>its performance or power consumption goals. Intel bought the Alpha
>process engineering staff and technology from Compaq just before
>Compaq announced they were being bought out by HP.

Only due to the press. The original wasn't expected to be high 
performance, at least by HP. It was more of a sort of public proof of 
concept. Current versions are equaling or out-performing PA-Risc.

>The "next" version (2006-7) of the IA64 (I don't know its name) will
>be "Alpha-inside."

The Itanium cpu is entirely different to anything that has gone 
before except an experimental design about 25 years ago. When the 
company that developed that folded, Intel bought the hardware patents 
and HP bought the software ones.

Upper league cpus have multiple instruction pipelines. Pentium and 
the cut-down PPC that we use only have one (plus specialist pipelines 
like Altivec). A pipeline is like an assembly line, so it decodes and 
actions multiple instructions at once, but each instruction it works 
on is at a different stage of decode/action. Multiple pipelines means 
that more than one group of instructions can be worked on at once. 
PA-Risc has 4, I believe Sparc and high-end PPC have 4. Alpha has 5, 
which is why it has the edge. The downside is that the instructions 
have to be routed to the best pipeline depending on preceding and 
subsequent instructions so as to get the best throughput. This is 
done on the fly by extra logic on the chip. The amount of Silicon 
used for this increases with the square of the number of pipelines. 
On the Alpha, more Silicon is used to route instructions than to do 
the computing, so 5 is about the maximum practical. (Many designers 
say 4, regarding the Alpha as over the mark). Itanium takes the 
routing off the chip altogether and puts it in the compiler. That 
means that the number of pipelines can be increased until it's 
no-longer effective in increasing performance.

At an NDA session 10 years ago I was told that Itanium would 
recognise both Pentium and PA-Risc code, but only in a single 
pipeline as routing information would be missing. I've not heard 
anything since to say that's not still the case.

Apple are unlikely to go with Itanium as it is at the moment, partly 
because there is no free compiler (and the HP patent may still be 
active?) and partly because a bare chip costs more than an entire Mac.

>Now of course, what I'm talking about here is the same problem that
>IBM has with the Power chips -- they are destined for high-
>performance, high-end, enterprise class servers ... i.e. expensive.
>
>This WAS Intel's roadmap for the IA64 "before AMD" (proved that they
>could make a 64 bit chip cheaply).
>
>My guess is that the "x86" chips that Apple will be getting will be
>these new generation "Alpha-inside" IA64 chips.
>
>I suspect that they will outperform anything currently on the market.
>
>Remember -- 2006/7 is a FULL GENERATION (if not 2) away from today's
>technology.

Depends what you call a 'generation'.

David


-- 
David Ledger - Freelance Unix Sysadmin in the UK.
Chair of HPUX SysAdmin SIG of hpUG technical user group (www.hpug.org.uk)
Leader / Co-Chair of Unix SysAdmin SIG of Interex (www.interex.org)
david.ledger at ivdcs.co.uk
www.ivdcs.co.uk