Computer Chess Club Archives


Search

Terms

Messages

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

Author: Matt Taylor

Date: 21:33:11 03/01/03

Go up one level in this thread


On March 01, 2003 at 20:10:49, Robert Hyatt wrote:

>On March 01, 2003 at 14:26:03, Matt Taylor wrote:
>
>>On March 01, 2003 at 10:55:32, Robert Hyatt wrote:
>>
>>>On March 01, 2003 at 09:56:59, Matt Taylor wrote:
>>>
>>>>On March 01, 2003 at 00:07:01, Robert Hyatt wrote:
>>>>
>>>>>What is the point of the question?
>>>>>
>>>>>Will I buy a 3ghz machine today and buy a 3.2ghz machine in 6 months?  No.
>>>>>But the risk is will I buy that 3.0ghz if your competition is right behind
>>>>>you in speed and significantly below you in cost (AMD vs Intel for example).
>>>>>
>>>>>But if you offer me 3.5ghz today, I'll take it if I am in the market for a
>>>>>new machine, no questions asked.
>>>>
>>>>There is a -lot- of pro-Intel prebias. Even a coworker of mine who knows enough
>>>>assembly to write self-modifying code seems to think AMD is inferior to Intel.
>>>>He once suggested that AMD chips are somehow inferior to Intel chips. While that
>>>>is obviously stupid, I think that is representative of a lot of people. I was in
>>>>CompUSA one day looking for something network-related. I couldn't find it, so I
>>>>was waiting on a sales rep talking to two older women about buying a computer.
>>>>He recommended an AMD-based machine because of cost, and they politely denied it
>>>>and requested Intel specifically because one had a son that told her to get
>>>>Intel.
>>>>
>>>
>>>That has always been a problem.  Even in the PC market, IBM has been a strong
>>>force based on customer demands.  Even the "punch card" was called "IBM card"
>>>which only added to their name recognition.  And they sold crap, and sold lots
>>>of it, because of name recognition rather than technical excellence (IBM is
>>>certainly a reputable company).IE the PS2's were trash.  I remember running
>>>a program "OPtune" that did sector slipping to make disk access faster, and it
>>>reported "This appears to be no faster than a floppy disk, with an average
>>>access time of 100ms +, are you sure you want to run optune on a floppy?"  It
>>>was running on a 20mb hard drive.  :)  But our chair demanded IBM machines when
>>>Tandy had machines with faster processors and much better drives (and everything
>>>else including no micro-channel nonsense.)
>>>
>>>So name recognition plays a role.  Right or wrong.
>>>
>>>>I don't think Intel is completely ignorant of this, either...
>>>
>>>Of course not.  Witness the "Intel Inside" advertising ploy that everyone
>>>has said was the best marketing decision they made.
>>>
>>>
>>>>
>>>>>I still don't see how this is an issue.  How will producing a slower product
>>>>>today help me tomorrow?  Once I lose a customer to a competitor, it is _much_
>>>>>harder to get them _back_.  I'd want to offer the best that I could offer, to
>>>>>drain _their_ customers that need more performance.
>>>>>
>>>>>IE Cray _never_ played these games, _ever_.
>>>>>
>>>>>I don't believe any other vendor does either.
>>>>
>>>>That all depends. You only have to stay just ahead of your competitor in
>>>>performance. Many people buy the high-end part from the high-end company unless
>>>>it is unreasonably expensive. The 3.06 GHz Pentium 4 is generally regarded as
>>>>faster than the AthlonXP 2800 or AthlonXP 3000. (Whether it actually is or not
>>>>is a matter of debate, but popular opinion is what counts here.)
>>>>
>>>>Dell and Gateway purchase Intel processors in mass quantity. Dell phazed out its
>>>>Athlon line several years ago. Gateway did so shortly thereafter. The desktop
>>>>market is -dominated- by prebuilt PCs. Dell and Gateway are major players in
>>>>this market, and both of them use Intel processors exclusively thanks to Intel
>>>>strongarm tactics.
>>>
>>>That's another subject.  IBM did this in the 60's and 70's fighting the "plug
>>>compatible" peripheral folks like Telefile and so forth."  They used the same
>>>sort of tactics to make life miserable for the non-IBM suppliers.
>>>
>>>
>>>>
>>>>Basically what I am saying is that Intel holds all the cards. Even when the
>>>>high-end Pentium 4 was obviously slower than the high-end Athlon, Intel was
>>>>heavily favored. The balance has shifted substantially as Intel is more
>>>>competitive now.
>>>
>>>
>>>I don't disagree there at all, of course.
>>>
>>>
>>>
>>>>
>>>>Assuming Intel has consistent customers (as I have argued), it is not
>>>>unreasonable to think they might take advantage of that. Releasing chips as fast
>>>>as they can means nobody is going to buy the 1.8 GHz because the 2.6 GHz is only
>>>>$50 more. Releasing chips slowly means the consumer buys 1.8 GHz and upgrades to
>>>>3.6 GHz a couple years later. This gives them more revenue in the long run.
>>>>
>>>>-Matt
>>>
>>>
>>>
>>>I'm not sure I agree with that.  For the reason that computer vendors have
>>>a constant stream of upgraders as well as a constant stream of new buyers.  The
>>>new buyers will buy when they need new machines and will probably buy at the top
>>>or near the top of the performance curve knowing the machines will have to be
>>>viable for 3+ years.  The upgraders are going to recycle about every three years
>>>and they will likely buy near the top of the performance range for the same
>>>reason.  So the issue is more about "being at the top by the widest margin" so
>>>that you can charge a premium (compare xeon to non-xeon pricing, for example
>>>and then look at how many xeon-based machines are sold in spite of the very
>>>small overall performance difference for most applications.)
>>
>>There are many, many people who don't buy top-of-the-line simply because it is
>>too expensive. When I spec'd out and purchased my desktop, I bought AthlonMP
>>1600 chips rather than getting 2000's which were available at the time. The
>>difference in price was more than $100 per chip, and the difference in speed is
>>a mere 266 MHz.
>>
>>This does not even apply exclusively to the computer literate/technical crowd,
>>either. A friend of mine asked me for advice a year ago on buying a new
>>computer. She ended up going to Circuit City or Radio Shack and getting a 2 GHz
>>Celeron. Granted I told her that she didn't need a super fast computer to edit
>>documents and send e-mail, but I think a lot of people are at least connected
>>with a computer literate person who would make the same recommendation.
>>
>>>So, in summary, I'm going to replace machines every three years or so, if I can,
>>>and I'm going to buy the fastest processors I can, so that they will last as
>>>long as possible performance-wise before I have to replace them again.  Intel
>>>will get no benefit by "holding back" on clock speeds as I am going to buy when
>>>it is "time".  If they jump the processor clock that will likely make "the time"
>>>come sooner, for me...  If they drag it out, that only extends the life of my
>>>current machines since they won't be that much slower.  Seems obvious to me that
>>>"fast as you can go" is the way to sell the most...
>>
>>Universities, corporations, and other organizations do not necessarily time
>>their purchases according to the speed of available processors on the market.
>>
>>Also, time to market for a processor is something like 5 years. If Intel ramped
>>their processors up to the max now, they wouldn't have a steady income while
>>developing their next product.
>>
>>If Intel went full speed, they really wouldn't sell many 1.8 GHz Northwood chips
>>(because the 2.53 was available). Some are marked down, so perhaps Intel would
>>get the same money per processor. However, Intel's strategy is to get the 1.5
>>GHz Williamette in desktops, then the 2 GHz Williamette in desktops, then the
>>2.53 GHz Northwood in desktops, then the 3.06 GHz Northwood...
>>
>>Consider this, too. Would you have upgraded sooner if 2.8 GHz Xeons had been
>>available sooner? Naively assuming the new 0.13 process could produce 3.06 GHz
>>Xeons as soon as Intel started producing, would you have bought 2.8 GHz Xeons or
>>3.06 GHz Xeons? If you had bought a 3.06 GHz Xeons, you would probably pay about
>>the same price you paid for your 2.8 GHz Xeons. More for your money means less
>>profit for Intel. Now, because you bought 2.8 GHz Xeons instead of 3.06 GHz
>>Xeons, you are also more likely to upgrade sooner. That's another win for Intel.
>
>
>That doesn't describe my purchasing approach at all.  I upgrade when (a) a
>machine is deemed to slow by today's standards _and_ (b) I have the money to
>do so.  At that point, I buy the fastest processor I can afford.  For some
>platforms, this means the fastest that is available _right now_.  I don't
>look at 2.8 vs 3.06 and think "If I wait, I'll get 3.06" because if I wait
>longer I'll get 3.2 or 3.6 or whatever.  So what drives _my_ purchasing
>decisions is "what can I buy _right_ now" and nothing more.

But that's exactly my point -- nobody waits. Intel wins by staggering release
because you upgrade through multiple releases of their CPUs. If they released
everything up-front, we would all buy the high-end chips instead of buying
high-end now, high-end in a couple years, more high-end a few years after that,
etc.

>So if Intel ships 3.0 today, that's what I buy.  If they ship 2.8, that is
>what I buy.

If you buy a slower processor, criteria (a) is met more often, and it doesn't
cost Intel anything extra in the desktop market. Perhaps you might not buy that
frequently, but some people do. I know a fair number of people who will squander
money on the fastest thing Intel has to offer, and they do it every 6 months. If
Intel can sell them the same chip on the same process twice (i.e. 2.53 and 3.06
GHz), Intel makes out like a bandit. A Pentium 4 costs nowhere near $600 to
manufacture.

>For high-performance things, I might have a target mhz.  IE someone running
>on 1.2ghz hardware and he wants his application to run (say) 3x faster to reduce
>delays.  I wouldn't consider buying 2.4 for him.  I'd hope for 3.6 and hope the
>thing scales and test it by benchmarking before I buy.

I wasn't talking so much about requirements. I was thinking more that few people
will upgrade from a 2.8 GHz Xeon to a 3.06 GHz Xeon because it is
cost-prohibitive and returns very little additional speed. The point is that
people will buy the 3.6 if available, otherwise they buy the 2.8 (or whatever
-is- available). If they buy the 2.8, they're more likely to upgrade when the
4.0 is released. If they get the 3.6, there's no chance.

>I've seen _several_ labs here constrain a machine in that way.  It has to
>run this application this fast, or I won't buy, and I've helped them find the
>hardware to deliver that level of performance.  If Intel is lagging, whether
>it be intentionally or because they can't go any faster, then we look at
>alphas and other platforms...

Yes, but that isn't exactly desktop world, is it?

>>It doesn't make a lot of difference when considering 2.8 vs. 3.06, but it does
>>when you're talking about 2.53 vs. 3.06. Some people crave high-end performance
>>and will upgrade through every iteration.
>>
>>-Matt
>
>
>I've never done that and I suspect that is a _tiny_ segment of the market.  I'd
>bet that 90% of upgrades are done cyclical at around three years or (for smaller
>companies) perhaps as long as four years.  For the first time purchaser, I think
>they will go for what they can get based on what they can spend.

And if Intel staggers a CPU core to last 6 years, they've protected their
investment.

>One thing is for sure, I am _much_ more likely to upgrade if the processor goes
>from 2.8 to 4.0, than I am if it goes from 2.8 to 3, to 3.2, to 3.5, to 3.8 and
>finally to 4.0.  For things like chess, as one computational example, I'm not
>going to spend money for almost no speed gain.  So if Intel drags their feet on
>ramping up the clocks, they won't encourage _me_ to upgrade sooner.

Ironic that my example a couple paragraphs above is so similar. Intel isn't
expecting that you will buy the 2.8, 3.06, 3.2, 3.5, etc. Intel is expecting
that you will buy the 2.8 and then upgrade to a 4.0. If they released a 4.0, a
3.6, a 3.2, and a 2.8, they wouldn't get your money for the 2.8's. If they
release incrementally, there is a chance they will get your money for the 2.8's
and the 4.0's. As you said, many new systems are being sold, but Intel isn't
losing their business. Intel has nothing to lose and everything to gain.

-Matt



This page took 0.01 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

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