Michael S (already5chosen@yahoo.com) on 2/8/09 wrote:
---------------------------
>someone (someone@somewhere.com) on 2/8/09 wrote:
>---------------------------
>>
>> Maybe
>>you just haven't noticed but x86 still takes approximately
>>0% of the mission critical computing market despite
>>being in the server market in a serious way since 1995.
>
>For some very narrow definition of mission-critical (mission-critical=counting big bucks) that could even be true.

Please define what you mean with mission critical.

Some ftp server yours at home that you want to run 24/24?
Some number crunching software where you want to avoid at all costs bitflips in results as the long term calculations might go wrong because of a single memory bitflip?
London Stock Exchange?
Network from a big company?
Novell server from a small financial company (say 3-15 workers)?

>Now 100% of my development-related mission-critical computing is done on x86 and
>it's not even XeonMP. I'd guess you can say exactly the same about yourself.
>
>From my exposure to industrial computing I can assure that x86 has much higher
>presence than any other general-purpose CPU. And ones again those are not XeonMPs.

Price dominates. Most software scales well from big companies, they nowadays use 2 socket hardware and have no real preference for intel or AMD, yet vaste majority there is intel 2 socket Xeon machines. It is actually tiny companies that buy 4 socket hardware sometimes in Europe. For them 1 big fast machine is easier than the hassle of many computers. Many will be tempted by 4 socket AMD or intel if price is not too overdone for it and go for the processor that is just a tad slower.

Price really dominates more than most think. It's not like whether something is 5 dollar more expensive than another solution. It simply has to keep within some sort of budget of say 8k-10k euro.

Suppose a quad socket AMD 2.7Ghz shanghai is 10k euro all together and a 32 core Xeon MP solution is 40k euro.

then obviously the Xeon MP is out of sight.

Previous Xeon MP solutions back in 2003/2004 were really expensive at the time and when they reached the market end 2003, start 2004, the opteron really was that much faster and affordable that it got ordered.

I'd argue a 8 core Xeon MP processor now has no competition at all and will cleansweep if it is not overexpensive. I bet it will be difficult for intel to pick a price for it,
as a much higher price means the profit on 1 cpu is that much higher that it is questionable whether you can make up in sales.

Realize also that past few years the hardware did get a lot faster on testsets, but in reality that soup didn't get eaten as hot as that (sorry for using a dutch saying as a metaphore).

Nehalem is a good cpu. Yet its problem is rather trivial.
4 * 3.2 Ghz = 13.2Ghz

If we compare oldie Q6600 : 4 x 2.4ghz = 9.6Ghz.

9.6Ghz <==> 13.2Ghz

Big deal for the freaks, but to managers who buy hardware it is no big deal.

Now do this math: next Xeon MP machine is (let's assume for now it is clocked similar to i7-965 as i wouldn't know its clockspeed):

8 * 3.2Ghz = 25.6Ghz for 1 cpu.
You can put in 4 of them, so it delivers a total of
100+ Ghz power.

Now that is a BIG difference compared to previous hardware, so it is a reason to order it, IF the company can afford it.

Xeon MP will have really good sales, provided that despite recession companies can still order hardware. It is because it looks just so superior to even the biggest layman and you don't upgrade your hardware to get a factor 30% faster.

Some guys i know around me use "it has to be factor 3 faster" type lemma's and as they would not know how to EVER find something like a benchmark online, let alone know how to test their own software at it, they blindfolded order based upon the above simplistico math.

>Power ISA also presents in significant quantities but I have never seen a single
>"high-RAS" Power chip, just "low-RAS" embedded chips.
>
>There are other segments about which I have a clue - they are all the same. No
>real work is done by mainframes/Power/SPARC/Itanium/XeonMP. The closest to the field
>you find those is billing. A lot of real work is done by DSPs and microcontrollers.
>When occasionally (not that rarely) you see a general-purpose purpose CPU in more
>than 99% of the cases it's either x86 or ARM.
>