LFS on PII 233 with no fan in heatsink

Ian Molton spyro at f2s.com
Wed Feb 5 15:52:04 PST 2003


On Wed, 5 Feb 2003 23:37:34 +0000 (UTC)
alangrimes at starpower.net (Alan Grimes) wrote:

> Ian Molton wrote:
> > > An Ideal chip would be a PC version of the 21364! It could be >>
> > > produced
> 
> > Nonesense. theres no way a 21364 could be adapted to run X86 code
> > natively without stupendous design compromises. The advantage would
> > be negative.
> 
> Who the hell wants to run x86 code?

You said you wanted a PC version. the term PC these days means X86, like
it or not.

> I thought we were trying to move into the Open Source age. 

X86 is an open architecture. no problem there. no worse than alpha or
ARM anyhow.

if you want an open source CPU you are going to have to start learning
VHDL or something and buying FPGAs.

> > *cough* Stop thinking X86 *please*.
> > compared to an ARM the VIA C3 is a greedy power sucking monster.
> 
> I was talking in terms of processors that you could actually buy today
> and put in your desktop computer for a reasonable price/performance.

You can.

> > My dad has a desktop machine with an ARM CPU in it (so do I, but
> > mines 15 years old). it runs at 600MHz, consuming LESS than 1Watt of
> > power.
> 
> Who sells _THAT_???

Castle technology.

> And for how much? 

1299ukp. which is pretty damn good given the size of the market for a 32
bit RISC desktop PC.

It has no FPU, and is not superscalar. it IS fast though, and have 64
bit 66MHz PCI.

I recon a fixed-point-math version of Q3 would run rather nicely on it.

> > X86 has a LOT to answer for. its shit, and will NEVER EVER be low 
> > power.
> 
> I am well aware of that. 
> That's why I still uphold the dream of Alpha on the desktop!
> :_( 

Forget it. Alpha is dying. Sad, but true.

> > I could fit nearly 100 strongARM CPU cores onto the die of a P-IV. 
> > thats 100*300MHz of CPU grunt, or, in marketing terms, 30GHz of CPU 
> > grunt. It'd be a fantastic project.
> 
> problems: 
> 
> 1. That many chips would choke on the bus. =( One solution might be to
> provide four seperate 32-bit quad band memory channels..

Yes. you would need local caches, and fast busses. however, 50 cores on
a chip doesnt seem unrealistic, even with all this baggage. the SA core
is *miniscule*.

> 2. A typical linux system would only be able to use 10 or so of them
> effectively, the rest being idle most of the time. (not enough threads
> in the ready state)... 

True. but thats 3GHz worth, and another 27GHz in reserve ;-)

> Paralell programming is difficult and takes considerable expertise
> above what it takes to write normal programs.

Not always.

Consider a 100 CPU system software-rendering a scene from quake3 at
1280x1024

you would dedicate about 10 CPUs to vector manipulation and
rasterisation (trivial).
once that is done, each CPU graps ~10 lines, and renders them. the data
bandwidth is not too high (400MB/sec for 1600x1200 at 75Hz, 32bit colour).

See - not everything is hard to do.

a typical linux system would flie along with *1* 300MHz StrongARM, if
the code wasnt heavily X86 optimised (lots of floats, little ARM
assembly).

Just think - a general purpose CPU as powerful as a 3D video card, yet
with no worse thermal dissipation than a typical Via C3 @ 800MHz.

> What you really want to
> do is design languages and environments which are _IMPLICITLY_
> paralell. 

Nah. just learn how to use the stuff effectively. Some tasks just dont
parallelise. tough luck. but many MANY will with a slightly different
approach.

> Ofcourse to do language research I need an operating system to support
> this.

:-)
 
> Even though I hate leenooks I realize that it is probably my best
> option for the forseeable future. 

I really think you should drop the 'I hate linux'. IT wont win friends
or support here, and probably get you glared at.

It has its flaws, but its actually PERFECT for your purposes. its a ROCK
SOLID base to write languages on.

> For research purposes I need a small, tightly controlled, environment
> that I can easily manage. This is why I chose Linux from scratch.
> Through this I should be able to build a zero-bull system that will
> serve my needs. 
> 
> So, Ian, I'm already working on the very problem of supporting
> massively paralell operating environments. ;)

Well, I'll let you know when I start my 1000core MegaARM when Im a post
grad researcher...
-- 
Unsubscribe: send email to listar at linuxfromscratch.org
and put 'unsubscribe lfs-chat' in the subject header of the message



More information about the lfs-chat mailing list