Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Introducing "No-Moore's Law"

Author: Matt Taylor

Date: 19:03:59 03/11/03

Go up one level in this thread


On March 11, 2003 at 12:22:09, Robert Hyatt wrote:

>On March 10, 2003 at 19:56:12, Jeremiah Penery wrote:
>
>>On March 09, 2003 at 22:10:06, Robert Hyatt wrote:
>>
>>>On March 09, 2003 at 01:26:33, Jeremiah Penery wrote:
>>>
>>>>On March 08, 2003 at 23:58:37, Robert Hyatt wrote:
>>>>
>>>>>On March 08, 2003 at 14:52:02, Jeremiah Penery wrote:
>>>>>
>>>>>>By selling them side-by-side, Dell is comparing them.  But Pentium3 is not quite
>>>>>>a Pentium4 in fact...
>>>>>
>>>>>And nobody assumes that.
>>>>
>>>>People assume P3=P4 (in terms of compatibility) just as much as they assume
>>>>K6=P2.
>>>
>>>Maybe some do.  But I'd suspect most don't although the assumption that
>>>executables compiled and running on a PIII will also run on a PIV is a
>>>good assumption.
>>
>>I'm quite sure that almost everyone assumes they are 100% compatible.  Some kind
>>of poll could be useful, otherwise both of us are just guessing.
>>
>>>>>But clearly if someone compares a chip to a PII,
>>>>>the implication is that they are equivalent.  Otherwise the comparison is
>>>>>not very informative.
>>>>
>>>>I've seen P4 compared to 486, but that doesn't mean anyone should assume they're
>>>>totally compatible.  Just like you can compare an Alpha to a SPARC.
>>>>
>>>>But the P3-P4 comparison is much more relevant.  They're placed side-by-side
>>>>_all the time_, and very clearly is it implied that they're equal except that
>>>>the P4 is clocked much higher.  The average buyer is NOT going to know the
>>>>difference, any more than they knew the difference between a P2 and a K6.
>>>
>>>Did you see an advertisement where someone was trying to convince you to
>>>buy over the other?  I doubt it.
>>
>>When P3 and P4 existed concurrently, every single ad where Intel advertised the
>>P4, they were trying to get you to buy P4 over P3.  Of course they didn't
>>explicitly say so, but I don't ever remember seeing an AMD ad where they
>>explicitly mentioned an Intel chip either.  The P3 was, for many things, faster
>>than the higher clocked P4.  Intel prematurely killed the P3 because they wanted
>>to sell P4s, and P3 could have made P4 look bad.
>
>You are arguing this from the _wrong_ side.  The PIV is 100% compatible with the
>PIII.  The inverse is _not_ true, and this is often referred to as "backward
>compatibility".
>
>That is _different_ than their trying to sell a PIII and saying it is equal to
>the PIV but
>much cheaper.  And that case is closest to the K6 vs PII point.

Not really. If that were true, the K6 would have fit in a Pentium 2 socket. The
K6 was for all practical purposes a Pentium on steroids. It fit in a Pentium
socket, had a very similar design, and supported roughly the same parts of the
ISA.

AMD never marketted the K6 as a replacement for a Pentium 2, just as competition
for it. Intel does the same thing with their own lines of products. Look at
laptops where Pentium 3's are STILL sold alongside Pentium 4's. I've seen only 1
AMD ad that compared an AMD product with an Intel product -- a commericial that
was aired somewhere in Europe, I presume in England since the commericial is in
English.

I am suprised that anyone with Linux would bother distributing a binary or would
be confused at all about differences between chips. Most packages that I've
messed with require me to download and compile.

>>>>>It was the _users_ that assumed that PII = K6 since AMD magazine publicity
>>>>>certainly implied it if not outright claimed it.
>>>>
>>>>They imply that P3=P4 too, in terms of compatibility, but they're not completely
>>>>so.
>>>
>>>Again, it depends.  If I do a target=P3, I would expect that to run on a PIV.
>>
>>If I do target=Pentium, I'd expect it to run on a K6.  That's basically the
>>comparison you're making there.
>
>No, I bought a PII and did "target=pentiumII" because _ knew_ that the PII had
>different optimization isues than the original pentium.  The PII had an OOE core
>for example, while the original pentium depended on the compiler to pair
>instructions
>for the super-scalar execution units.  At the time I bought a PII, AMD was
>advertising
>their chip as faster and cheaper.  And it did _not_ say (but not 100%
>compatible) and
>that led to the confusion I mentioned.

That's because it didn't say it was compatible with a Pentium 2. I'd say you
-assumed- it was compatible. I've still never heard of any other case besides
yours and Win95's faulty code where AMD was burned for someone else's mistake,
and I don't think the average consumer even blames AMD.

I would say the average consumer buys Intel because it is what's available. To
make an analogy, if you're starving, do you eat what's put in front of you or do
you insist on food more to your liking? I would also say that the average
consumer also doesn't know the difference. They both taste like chicken unless
you're doing assembly.

>>>And most likely if I do a target=p4 it would run on a P3 although I can
>>>certainly think of reasons why it wouldn't.
>>
>>Most of the time it would, just as most of the time K6 would run P2 executables.
>> The only difference I'm aware of (for K6/P2) is CMOV.  For P3/P4 the difference
>>is SSE2.
>
>But _if_ you compile for the PIII it will absolutely run on the PIV.  If you
>compile for
>the PIV it will run on the PIV but perhaps not on the PIII depending on the
>instructions
>the compiler is capable of producing.  However, _nowhere_ did you see Intel say
>"The
>PIII is 100% compatible with the PIV".  You will find "The PIV is 100% backward
>compatible with the Pentium processor product line."

It's a good thing they didn't, just like AMD has never claimed the K6 supported
everything the Pentium 2 did. I have -never- had that impression, and I was
using a K6 before I knew anything about post-Pentium x86 assembly.

>>>But Intel doesn't market the P3 to be compatible with the P4.
>>
>>No.  They market P4 to be compatible with P3.
>
>Playing word games.  "backward compatible" is not "compatible".
>
>look at their ads.  It's very clear what they are saying and what it means.
>
>
>
>>
>>>>Why don't you use Sun's compiler, which is really fast for SPARC anyway?
>>>
>>>Because we don't buy it.  It's not cheap and we have so many gcc-based machines
>>>we felt the compatibility between compilers was better than the cost of buying
>>>Sun's compiler.  We used to have it years ago, and yes, I found it to be a
>>>bit better than GCC.  But eventually GCC started producing code that was as fast
>>>as Sun's so we dropped sun's compiler.
>>
>>Seems GCC doesn't do so well anymore. :p
>
>Nope.  But when sun became "pseudo-64bit" their compiler started to suck badly.
> I'm
>not sure what Sun thinks they are right now.  Most of the operating system is
>_still_ 32 bit
>code.

Does the OS need to be 64-bit? The biggest reason behind the push for 64-bit
computing is in breaking the 4 GB barrier...

You'll probably see the same thing with Opteron. The default machine word size
is 32-bits even in 64-bit mode. You have to use a prefix to override to 64-bits.
(Internally it zero-extends the 32-bit operands to 64-bits.)

-Matt



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.