Re: Intel Moves Towards Itanium and Xeon Convergence

JF Mezei wrote:
n.rieck@xxxxxxxxxxxx wrote:
Intel Moves Towards Itanium and Xeon Convergence

http://www.dailytech.com/article.aspx?newsid=6236


The CSI was announced back in 2004 (or early 2005). (Common system chipset that will allow the 8086 to scale to same size as IA64).

Not necessarily: Intel could choose to hobble x86 if it wanted to - the surrounding chipset may allow the same number of sockets, but the on-CPU use of it could have different socket (and local RAM support) limits.


Note that HP apparently intends to continue to use its own proprietary system chipsets, so it is not clear how this will play out within HP.

Have a reference for that? At this point, I'd be somewhat surprised to see HP rolling its own CSI chipsets: while years ago the theory was that Intel would only provide, say, direct support for 4 sockets in CSI and leave larger system support up to its major OEMs, the competition on the x86 front (with the likelihood that AMD will be supporting larger configurations natively before then: 8 sockets by the end of this year in a more effective configuration than its current 8-socket offerings, and 32 sockets some time next year) may make this strategy less feasible.

Furthermore, making the other Itanic OEMs invest in developing their own CSI chipsets would be a good way to drive them away entirely: they have far less dependency on the platform than HP does, and might well choose to drop it rather than either play second-fiddle to HP or shovel additional bucket-loads of cash down that rat-hole. SGI could be the sole exception here, since they need to support systems (whether x86 or Itanic) larger than anyone else, hence have no choice but to develop such support on their own if they wish to continue in that market space.


Also, if the dates quoted in this article are correct, it means that Intel is ahead of its delayed schedule. Last I have heard, the 8086 was to get that CSI in 2008 and that IA64 thing would get CSI with Tukwilla late 2009.

I haven't heard anything that disputes shipping CSI support with x86 next year, and Gelsinger's statement about the 'realization' of Tukwila in 'late 2008' could well refer to something earlier than an actual ship date (though probably something later than first silicon - otherwise, they'd be shipping closer to 2010). Then again, Intel has been scrambling lately, and it's not completely inconceivable that they could pull in Tukwila ship to, say, December of next year: Itanic will sure as hell be hurting badly until Tukwila hits the streets, given that Montvale seems to be only a somewhat warmed-over version of Montecito (which POWER5+ already blows out of the water commercially, with POWER6 promising to about double POWER5+'s per-core performance this year and POWER6+ due some time in 2009 - not to mention the continuing x86 improvements by both Intel and AMD this year and next and improved competition from Fujitsu/Sun's SPARC64 platforms).


(Back in 2004 when the announcement was made, both were to get the CSI at the same time in 2007).

Which makes Gelsinger's claim that "the first Montecito slip aside, we're back on track" just a tad disingenuous. Perhaps he just meant that, after the 1.5 year Montecito slip, there were no *additional* amounts of slippage yet perceived as likely in the program (i.e., the whole program had slipped by that amount).


What is significant here is that the 8086 will get the same system level features as IA64. (Remember that RAS argument made by the HP rep who tried to find one reason for IA64, well by that time, RAS won't be a differentiating factor even to marketers)

That, again, depends on the CPU chips, not the surrounding chipset: Intel *could* still choose to differentiate x86 from Itanic on that basis if it doesn't feel compelled to compete in that area with AMD x86 products.

- bill
.

Relevant Pages

  • Re: PCMCIA wireless cards....just works?
    ... > If you have a card that 'just works', please post the name, model number, and revision in this thread. ... Prism chipsets were manufactured by Intersil - apparently they ... So - if you buy a WiFi product that has open-source support, ...
    (Fedora)
  • [PATCH] AGP 648[FX] cleaned+fixed but still have problem. Plz help.
    ... This patch is provides AGP support for sis 648 and 648FX chipsets ... This is fatal during initialisation because the bridge must be ...
    (Linux-Kernel)
  • Re: Recommendations for Wireless Network Cards
    ... Classic Gold PC Card"; it is mf'd by Agere Systems, ... Prism chipsets were manufactured by Intersil - apparently they ... So - if you buy a WiFi product that has open-source support, ...
    (Fedora)
  • Re: [RFC v13][PATCH 00/14] Kernel based checkpoint/restart
    ... Open TCP/IP sockets can legitimately get reset as well. ... what happens if I checkpoint a process in 2.6.30 and restore it in ... http://selenic.com: development and support for Mercurial and Linux ...
    (Linux-Kernel)
  • Re: Best supported motherboard maker and linux?
    ... >> solid unit with regard to linux. ... FWIW, I've had no problems with VIA chipsets, except for the stupid DMA ... > README.ati which states that currently agp support is only implemented ... a Mach64 soldered to it was a piece of defective crap. ...
    (comp.os.linux.hardware)