From: nan@matt.ksu.ksu.edu (Nan Zou) Subject: Re: Intel, the Pentium and Linux Date: 26 Apr 1993 14:16:27 -0500
jerry@msi.com (Jerry Shekhel) writes:
>Nan Zou (nan@matt.ksu.ksu.edu) wrote:
>: >But how much do they cost? I'll bet the Pentium will be cheaper. What
>:
>: If that's your bet then you'd probably lose. The Pentium costs around
>: $1000 in quantity, the Alpha costs about the same. R4000 chips are much
>: cheaper, ($400?).
>:
>In a year or so, Pentium will cost what the 486 costs now ($400?). The Alpha
I agree the price of Pentium will probably come down due to market
force, but I also think the Alpha has a lot going for it, market wise,
DEC is having a Japanese manufacturer (Mithusta?) second sourcing Alpha
chips, so price probably won't remained too high. They also plan to
release cheap Alpha PCs running NT.
>is less likely to decrease in price as much. In any case, I agree that CPU
>cost is pretty much irrelevant. I do believe, however, that Pentium will
>win big in terms of total system cost, though.
The initial price of above-mentioned Alpha PCs will be less than $5000,
from all accounts the initial price of Pentium based system won't be
less than that, but in the long term, I think the Pentium will win out.
>: The Pentium is pretty much as far as you can push in the 80x86
>: architecture.
>Oh come on. Isn't that what they said about the 286, the 386, *and* the 486?
My statement was a little absolute. I probably shouldn't make predic-
tions that could be proved wrong by the Hexium. I still think it takes
herculean amount of work to extend the x86 architecture as much as Intel
did, anyone know the amount of money spent on the Pentium? I do admire
Intel for their efforts.
>: It's also one prime example why CISC chips are at a dead
>: end against RISC chips, for millions of dollars in R&D, long develop-
>: ment cycle and twice the transistor count (3.1M vs. 1.6M), it still
>: can't touch the Alpha.
>Can't touch the Alpha? I'm not so sure. The slowest Alpha machine in
>existence runs at what, 150MHz? I find the fact that a Pentium running
>at 1/3 the MHz of the Alpha comes within 15% of its performance pretty
>impressive.
Here are some SPEC numbers I have for the various CPUs.
SPECint92 SPECfp92
200MHz Alpha (10000-610) 116.5 193.6
182MHz Alpha (7000-610) 103.1 176.0
160MHz Alpha (4000-610) 94.6 137.6
150MHz Alpha (3000-500) 84.4 127.7
133MHz Alpha (3000-400) 74.8 112.5
66MHz Intel Pentium 64.5 56.9
60MHz Intel Pentium 58.3 52.2
Intel i486/66DX2 32.4 16.8
150MHz MIPS R4400 94 105
40MHz SUN SuperSPARC 53.2 63.4
99MHz HP PA-RISC 80 150.6
62.52MHz IBM RS/6000 59 125
Now I don't think the Pentium numbers look half bad, even when stacked
up against the Alpha. But is it as extensible as the Alpha?
Performance/MHz doesn't really apply in this case, the Alpha was
designed with the objective of being the fastest processor using the
most advanced technology, in the process they have to make certain
trade-offs, the Pentium, HP-PA and R4000 might have it beat when it
comes to performance/mhz, then again, Pentium, HP-PA, and MIPS don't
have any chips running at 200 MHz and easily raised to 400 MHz. How
easily can the Pentium's clock speed be raised?
>impressive. And Pentium machines will simply blow away Suns and other
>workstations in terms of price/performance. Even 486 machines are about
>as fast as SuperSPARCs, and yet I'm sure you're not so critical of that
>architecture.
Actually I'm even less impressed by the SuperSPARC, Sun has lagged
behind HP and MIPS in performance for a long time, but they are selling
more workstations than ever. Simply because they have market momemtum,
good marketing force, established software base, hmm, sounds pretty
similar to the 80x86 market.
>: The 80x86 architecture is severely limited to
>: begin with, any attempts to extent it will likely be difficult.
>Yeah, yeah. Intel bashers have been carrying this tune since the '70's,
>yet Intel has managed to close the PC-workstation CPU performance gap
>with that x86 architecture.
It's not just me who's saying that, even Linus the Master himself often
bemoans the design of the 80386 architecture, I remember hearing him say
a few times that Linux would be easier to write on the 68040.
>: The Alpha is going quad-issue superscalar and have its speed raised to 400
>: Mhz next year. While at 66 MHz Intel is having major heat problems (you
>: probably won't see 66 MHz Pentium chips this year), the pipeling must be
>: hell to do with so few general purpose registers and unorthogonal
>: instruction set, if Intel was smart they'd start with a clean slate on
>: the Hexium (Sexium?), but knowing Intel this probably won't happen.
>[Watch as I steer this discussion back into the realm of this newsgroup :-)]
Now that you mention it, maybe this topic would be more fitting in comp.
sys.amiga.advocacy? ;-)
>Hey, don't get me wrong -- I believe that the Alpha is an innovative
>architecture, and will probably be more easily scalable than the x86.
>Maybe I've just been turned off by the sh*tty system software that currently
>runs on the Alpha and HP (and I doubt that NT will change my mind about that).
>I don't want to sound like an evangelist,
But if you were, Jerry, you'd be preaching to the choir, I love Linux as
much as the next fanatic in this group, that doesn't stop me from
pointing out what I think are the deficiencies of the current architecture.
> but IMHO, Linux simply wipes the
>floor with both OSF/1 and HP-(S)UX, as well as almost everything else I've
>worked with. SunOS/Solaris is the only OS that's in the same league as far
>as I'm concerned. The thing is, I don't think any of this will change until
>the hardware vendors stop thinking they can develop operating systems as
>well. That's why my next machine will probably be a Pentium -- so that I
>can keep running Linux/GNU/XFree86.
If I were considering future upgrades, I'd go with the Pentium route
also, not because of its technological advantages over the other RISC
chips, but because Linux runs on it. As nice as the Alpha and MIPS
chips are, I don't think their impacts will be too significant (compared
to Intel) to the destop PC market. Now the PowerPC on the other hand
might have a chance...
-- Nan