Linus Torvalds (torvalds@linux-foundation.org) on 2/25/09 wrote:
---------------------------
>someone (someone@somewhere.com) on 2/25/09 wrote:
>>
>>IDC says x86 server revenue fell faster than the entire
>>server market in Q4. Commodity server sales fell faster
>>than mid range and much faster than high end.
>
>Umm. Hello? There's the obvious explanation for that
>one, but I guess you don't want to hear it.
>
>But in case you do, I'll just mention it: large systems tend
>to have much longer procurement procedures. Just
>about every place I can think of lets managers make
>discretionary buying decisions for things under about $5k,
>so things like a few new workstations can easily get bought
>on the spot. There's no need to do any strategic analysis.
>
>In contrast, buying a new big DB server might take a year or
>two of just planning. That whole "how do I upgrade from
>Alpha/VMS to Itanium/VMS" thread? Remember that one?
>
>Wouldn't you agree? Isn't this obvious?
>
>Now, with those facts in mind, spend the five seconds to
>think about what happens when an economy tanks and people
>start to tighten their belts. Think about how long it takes
>for the effects to ripple down.
>
>Hint: it has absolutely nothing to do with x86 vs IPF, or
>anything technical at all.
>
>Here's another hint: this is also not a "good thing" for
>the big-iron vendors. Yes, the effects of a downturn take
>a bit longer to be felt. But you might also think about
>what happens when the spigots slowly start opening again.
>
>But hey, you can continue to live in your dream world
>where apparently that whole "economic slowdown" didn't
>happen, and where IPF shrinking by just 5% is a great
>thing, because is shrank less than the low-end machines.
>
>Feel free to think that it's about IPF taking over the
>world, and how last quarter people finally started to see
>the light about how great IPF is.
>
>The rest of us? I think we'll stick to the rather more
>obvious explanation.
>
>Linus

Well you have to add up another thing. Governments only *slowly* start to realize now that there is a problem with respect to tax income. So semi-government organisations will get new budgets only a few months from now. And those are a lot less. In fact they are still optimistic for the future compared to the big lack of cash they will have.

IPF is something exclusively for semi-government organisations. Their budgets didn't shrink yet. Endless meetings about budget first have to take place.

In fact a few days ago i spoke with a few guys at a table after midnight, them with a few beers. All of them shared they:
a) didn't see the recession come

Which is typical for semi-government, as other than them about every big financial saw it come (note i had expected it short AFTER US elections, not short before; so i made a stupid mistake by 2 months there).

b) one of the fundaments of economic society: banks
they have a hard time now (discussion whether this form of SM is selfcaused or not is another subject). Yet despite that entire society is shaking in its fundament, for the above guys, as their monthly paycheck continues, they feel confident that in a month from now things will go great again. To quote one of them: "what is the problem of stock exchange going down bigtime?".

One of these guys looks suspected much, but i'm sure this was because of the beer he took, at someone who was an undersecretary of the state of finance.

You know, they play good weather. These type of guys, when in manager position, would order IPF type hardware. As they just have no clue on what 1 euro is worth. 1 euro from someone else or 1 million euro from someone else, same thing for them. Not their money, so they don't care.

IPF of course has 3 fundamental problems:
a) it is a design which performs only well at testsets because of a huge L3 size. So the actual corespeed as you correctly noticed is just a big underperformer.
b) it just never clocks high. 4 cores or 8 cores for tukwila, who cares? First of all if it gets 32 nm it will release earliest 2013 and 2.4Ghz is a lot slower than the 2.9-3.2Ghz from Xeon MP - i7.
c) it is a really expensive hardware solution

Now C a few years ago wasn't a major problem, but it sure will be a few months from now, also for governments. A tiny nation like Netherlands already is publicly broadcasting in media they have a billion or 40 less income (how do they manage to present that as economy going down just 4% or so instead of double digit decline?).

It is survival of the fittest.

Of course intel moves tukwila and demona cpu's to 32 nm or beyond. From summer 2009 until 2012 zero new contract jobs for new IPF hardware can get given by government organisations when using correct analysis of what they need.

They simply will not have the budget. Might tax money drip in again in 2012, it will take until 2013, that governments can increase budget bigtime.

So until 2012 i expect just to hear intel propaganda of a killer IPF cpu that each year doubles in core size on paper.

"Paper supports everything".

In the meantime the tiny processors keep on winning, as you already remarked in this forum a few years ago.

Vincent