Doug Siebert (foo@bar.bar) on 2/6/09 wrote:
---------------------------
>savantu (savantu@email.ro) on 2/5/09 wrote:
>---------------------------
>>Let's take another POV : does Tukwilla really need to beat Nehalem and X86 in general ?
>>
>>I'd say the answer is clearly no.In fact Tukwilla needs to equal Power and handily beat Sparc ( which will happen ).
>>Intel can happily push Tukwilla as a direct competitor to other high end server/mainframe
>>architectures while using Nehalem as a hammer to squeeze RISC/mainframes from above and below.
>
>
>For the highest end of the server market you are right. But Itanium left the workstation
>market, then left the low end server market, then it left the HPC market (or rather
>was left by the market - who would buy it now versus the x86 alternatives unless
>Intel essentially gives it away for free for marketing value?)
>

I remember reading pdfs from SGI where 1.6GHz Montecito trashed x86 ( 3GHz Opteron ) in some HPC apps.The whole problem with IPF is to tune the app for it, it can achieve unmatched levels of IPC.

Secondly, if you want large SSI, IPF is the only option.

>Now improved virtualization and clustering of virtual servers in the x86 world
>shrinks the market for high-RAS servers, and the PA-RISC upgrade cycle is just about
>complete so IPF's main growth conduit is petering out.
>
>Sure, it will beat Sparc, though Rock is different enough it will probably have
>its niche where it beats Tukwila handily, but those in the market for high-RAS servers
>won't be buying version 1.0 of Rock anyway. I'm skeptical Tukwila can equal POWER,
>but let's say it does. Its still a matter of IBM shops will buy POWER, HP shops
>will buy IPF, and both will be slowly but continually bleeding customers to x86 with each upgrade cycle.
>
>Delay after delay only hurts Tukwila, and I think its a fantasy to believe that
>Poulsen will hit 2010 given Intel's record with IPF schedules, and the fact that
>if it did Tukwila would have a pretty short life....even if they really will get
>Poulsen out on time they might not want to make it known, lest people hold off on
>upgrades during the down economy to wait for Poulsen!
>

The whole point is that it doesn't matter if Tukwilla loses to Nehalem , either way it is Intel who profits.It does matter however when x86 is a non-player in some niches ( can you call a $20B market segment a niche ? ), Intel needs something to compete there.That's where IPF is relegated and where it needs to succed.It does not need to beat x86 in a single benchmark ( SPARC did neither for the last 5 years , exagerationg of course but you get the point) unless x86 is AMDs turf.

So as long as X86 is owned by Intel and there is a multibillion market in the high end, IPF has a reason d'etre.
Tukwilla will IMO beat Power in pretty much everything. It's all about system uarch , a Tukwilla core is IMO more capable than a Power one.