Mail Archives: pgcc/1998/07/22/22:54:23
At 09:07 PM 7/22/98 +0200, you wrote:
>
>Hi again!
>
>>>>Now K6-350Mhz SDRAM at 112Mhz bus is slower than PII-300 EDO RAM.
>>>
>>>>Few months ago K6-200 used to be as fast as Pentium pro 200...
>>>>...right now nothing can stop PII, and especially not PII because i can
>>>>run parallel soon on it. AMD/IBM K6 and Cyrix M2 regrettably can't run be
>>>>put on a dual or quatro motherboard.
>>>
>>>>Anyway K6 is fast considering its price, but is it smart for a
>>>>compiler to make optimizations for an outdated socket 7 clone?
>>>
>>>>Assuming i have the choice, when selling software, then i'll
>>>>NEVER deliver a K6 optimized version, but always a PII/PRO optimized one,
>>>>which also runs at pentium (so without incompatible instructions like
>>cmove).
>>>
>>>>I'll do that because i think socket 7 is outdated. No future.
>>>>Look to the level 2 cache what they did to it. They put it at the
>>mainboard!!
>>>>Awfull! So socket 7 has no future, the faster your processor, the less
>>>>you profit from it as level 2 cache speed kills you.
>>>
>>>>That's what happens to my program when running on 350+ Mhz K6, it runs
slow
>>>>on it. Slower than on a PII-266 SDRAM even, and one of the reasons is
>>>>level 2 cache.
>>>
>>>There is just one reason why I did not buy a Cray, an Ultra-Sparc or some
>>>other cpu which is faster than PII, it's the money I dont want to spend.
>>>
>>>When you divide the performance thru the price, still AMD is one of the
best.
>>>What does it help when I pay three or four times more for a computer which
>>>will be absolutely outdated in next spring? So I will still buy cheep AMD,
>>>Cyrix or even IDT, because these are the fastet for the money I am willing
>>>to spend.
>
>>I already wrote this down. For it's performance it's cheap. Not needed to
>>tell this another time. It doesn't take away that the socket7 technology
>>is horribly outdated.
>
>gcc is still supporting 80386, ns32000 (I did work on such a beast
>in 1989), m88k, pdp11 and some other old architectures.
>
>I still think that you are comparing different things and argue in a
>wrong way.
>
>First of all, it is *OUR* (mostly Marc's) decision which CPU's he likes
>to support. This is a loosely connection of some people who spare their
>free time to improve performance on whatever CPU's they like. In most
>cases these are the CPU's the own or have access to. So if you *REALLY*
>want them to boost the performance of some special design, then you
>should not flame them, you rather should look for a way to give them
>partial/full access to such a machine. Since this is not a commercial
>company, do not expect that everyone of us has a rich grandmother or some
>other sponsor with never ending money. This is the reason why the cheaper
>(the CHEAPER not the slower) CPU's are better supported.
>
>Then there are patents on Intels PII-Bus, even memory cycles are patented
>(look at cyrix).
>
>>>What about selling your program bundled with a quattro-Xeon-board ;-)
>
>Look at the smile, this was ment as a joke.
>
>>I'm sure next Worldchampionship a lot will run on XEON or faster
>>(previous world championship, octobre 1997 a lot ran on Alpha,
>>two participants even ran at an experimental alpha 767Mhz!).
>>
>>I can't afford such computers, therefore i make my program parallel.
>
>I do not have a problem with someone running his programs on alpha's or
>HAL's or whatever, as long as they do not flame people, which spend their
>free time for fun and not for money in projects *they* decide.
>
>>A dual board is something like 300US dollar, a PII-333 chip will be around
>>250US within a few months, so i have 666Mhz then for a horrible cheap
price,
>... and will then be outdated too ...
>
>>which might equal a 500Mhz XEON (i estimate that i lose around
>>10% because of parallellism so i'm having a PII-600Mhz in fact when
>>compared to single CPU).
>
>The loss of speed could be bigger when you have a lot of concurrent
>memory accesses, but you have to prove that.
>
>>When the all categories world champs chesscomputer are held this quattro
xeon
>>system will be not by far the fastest system. We can expect 1 cray entry,
>>a 32 processor sun/sparc, a 40 processor system (about speed of 200Mhz,
>>a 32 processor PII-300 system, and at least 1 4 processor system.
>
>What about a program which runs on 5 200Mhz - AMD's connected by 100-Mbit
>ethernet? Chess is one of these programs which can perfectly run parallel.
>Did you compare the costs of 5 motherbords with outdated socket-7 cpus with
>the cost of 1 high-performance-up-to-date two-processor board? I think this
>will be equal in price, but the performance could be much better.
First of all they will be slower (slow processor and not having
shared memory), secondly it's cheaper to buy
a PII-400 dual within a few months
and carry that with you, than taking with you
5 computers + keyboards + all power + 100-Mbit ethernet cards + seperate
memory.
100-Mbit ethernet cards protocol is btw rather slow compared to
shared memory.
As you probably don't know: at intel processor there is a special
instruction to lock on atom-level.
Locking and stuff using ethernet cards will require considerable more time.
>>I might be forgetting some others running on multi-processor alpha's,
>>and optionally a french multiprocessor program might enter too at a
>>cray.
>Don't forget Deep Thought (is this the right name?) which was especially
>designed by IBM some years ago.
that were hardware chips connected to a fast mainframe/supercomputer.
Not the general purpose processor did the big job, but the processors.
Not smart to use examples you don't know what it did.
>>Although all of these systems are out of your reach, for others it's not.
>
>Oops! Why do you know that these systems are not in my reach? I do not want
>to buy always the fastest, the newest, the best, the whatever. I do not
>buy computers to show them my neighbours. I just do not want to spend so
>much money in things which get outdated so fast.
>But I agree, brute force is needed for your problem, maybe you can give
>Marc (or someone else) access to such a machine and you will sure see a
>better performance soon, but please argue in a more constructive way.
If you don't care for getting a fast machine then why do you want
to optimize code on it?
I mean: if you still now put all your spare time in making optimizations
for AMD K6 instead of PII/PRO, then your time will be wasted;
at the time a new pgcc version which can achieve reasonably optimization
is ready everyone will be able to buy PII cpu's, like you can buy now AMD K6.
Look how weird gcc is right now.
It can optimize EXCELLENT for 486 processors, but it works horrible on PII
chips.
Even you don't have a 486 processor anymore. For a low budget you get a
K6 now.
>>>>Vincent
>>>--
>>>Wolfgang Formann
>--
>Wolfgang Formann
- Raw text -