l i n u x - u s e r s - g r o u p - o f - d a v i s
Next Meeting:
April 21: Google Glass
Next Installfest:
Latest News:
Mar. 18: Google Glass at LUGOD's April meeting
Page last updated:
2001 Dec 30 17:13

The following is an archive of a post made to our 'vox-tech mailing list' by one of its subscribers.

Report this post as spam:

(Enter your email address)
Re: [vox-tech] Building a computer
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [vox-tech] Building a computer

forwarded for takashi:

I did not follow the entire of this thread, so my stmt might be
redundant. I appologize that. Anyhow, here's my opinion and what I know...

On Wed, Nov 07, 2001 at 12:20:44PM -0800, Bill Broadley wrote:
> > Umm..  that's where it gets fuzzy.  PCI bus is still at 33MHz.  Memory is
> > at 66, 100, & 133MHz.  Front side buses are at 100 (for the P3, I think),

Chipset is the buffer.
i.e. they don't have to agree w/ the speed.

> > 266 (133 actually but DDR'd, meaning double data rate), and 400MHz (P4,

Let me clear this. DDR uses rising and falling edge to represent the data
transfer. So the data transfer rate is doubled (i.e. x2). However, due to
the setup cost for each data transfer, the performance is not x2 in reality.
i.e. even though the bandwidth has been doubled, the latency was not shortened
at all (actually increased... i guess... I don't know the technical detail
tho. Anyone from architecture lab can answer this? or maybe we can just read
tech doc or research papers if we really want to know).

As Bill said. p4 is actually 100MHz * (32/8) = 400. (i did not know that as
I don't care intel anymore.  anyhow, thaks bill). Wider bus width can tranfer
more data at once naturally. But again where the bottle neck? If the system
still uses SDRAM, then tranfer to the cpu is at least 4 times slower. If 
RDRAM is used, the performance depends on the applications. That is, even
though RDRAM provides much higher bandwidth, the latency of it is very high
(i.e. set-up cost before actually start to tranfer data), and the width is
8-bit (that's why they can achive higher frequency; analogy is system bus
v.s. ethernet). To compensate, RDRAM sends a block of data. This block may
or may not contain useful info, but they feeds to the cpu(actually cache) 
anyways. The worse comes when data-prefech is employed. They guess next data
to be used, then prefetch to the cache. If the prediction was wrong, the 
data tranfer is a completely waste; in addition, since a new data was fed to
the cache, more cache miss can happen hurting the performance further.
RDRAM is still useful in particular applications such as multi-media (e.g
game consoles). Why? Because data used is mostly sequential, so it's easy
to predict the next data to prefetch.

> > but that doesn't mean it actually runs faster than a lower speed Athlon
> > Thunderbird).

Of course not. If you understand what I wrote, you know why.

> Overall though usually the leading edge athlon is faster then the
> leading edge p4 despite the p4 advantage in both clockspeed and FSB.

As mentioned above, p4 mostlikely prefech gabage data. That's one reason
they don't perform well. No good compiler is available because it's new.
If the compiler supports good prediction scheme (for prefetching data), then
things will change. But the biggest problem of p4 is "how to achieve higher
clock speed". They did so by shortening each pipeline stage which made p4
to have 20 or so pipeline stages. Having more pipeline stages is troublesome
because 1) penalty of pipeline register 2) jump instructions.

Again cpu guess which basic block to fetch to utilize the pipeline, but then
if the prediction was wrong, then the cpu must flush the whole contents of
pipeline, then get the correct instruction and data from cache. If the 
pipeline stage is, say 5, then at least 5 clock cycles are wasted. If the
number of stages is 20, at least 20 cycles will be wasted. Assuming the
wrong prediction was propagated to memory system, then cache miss will 
occur naturally.

> AMD has a substantial advantage in price performance:
> 	SIS 735 motherboard = $65
> 	>= 1000 Mhz athlon/duron = $100-$150
> 	512 MB ram = $50-80
> For a dual cpu add $150 ish for the motherboard, and of course another
> cpu.
> My favorite killer 3d card at the moment is the geforce 2 Ti for $115
> or so, small, cool, fast, well supported, linux compatible, even supports
> the new video standard for xfree86-4, so things like dvd playback work
> well.  Of course tribes2, wolf3d, and quake arena benefit as well...

I agree. That'd be a nice system in terms of computational power.
But! what I'm thinking is to build a quiet and cool system.
I have enough computational power (500MHz is enough for me as I don't play
fancy games, and I prefer to wach dvd on a regular tv). What I'm thinking

A. apply for both I and II
A-1. remove case fan, power supply fan.
A-2. use {flex|micro}ATX with built-in video and lan (but I prefer to use
     better sound card).
A-3. use 2.5 hdd
A-4. power supply is 150-200 is more than enough. (i hope).
A-5. micro atx case.

I. ADM duron
I-1. SiS 735 mobo (micro ATX)
I-2. duron under-clocked as much as possible.
   If it runs cool enough, remove the fan.
   Is it better to use XP underclocked around 500MHz?

II. via C3 (preferably .13 core)
II-1. 800MHz underclocked to 500Mhz or so.
      VIA officially says it can operate w/o fan.
II-2. SiS 630 or Via PLE133

I prefer to use AMD, but duron might melt down if heat wasn't dissapated
well. (visit tomshardware.com for detail). C3 is a safe bet, but I need to
buy pretty much everything. Have you guys ever thought about it? Any
suggestion? Thanks.

Takashi Ishihara                 http://wwwcsif.cs.ucdavis.edu/~ishihara
;; The first precept was never to accept a thing as true until I knew it 
;; as such without a single doubt.                (Rene Descartes, 1637)

----- End forwarded message -----

LUGOD Group on LinkedIn
Sign up for LUGOD event announcements
Your email address:
LUGOD Group on Facebook
'Like' LUGOD on Facebook:

Hosting provided by:
Sunset Systems
Sunset Systems offers preconfigured Linux systems, remote system administration and custom software development.

LUGOD: Linux Users' Group of Davis
PO Box 2082, Davis, CA 95617
Contact Us

LUGOD is a 501(c)7 non-profit organization
based in Davis, California
and serving the Sacramento area.
"Linux" is a trademark of Linus Torvalds.

Sponsored in part by:
O'Reilly and Associates
For numerous book donations.