1 Reply Latest reply on Jan 3, 2010 2:53 AM by Benjagan91

    A question regarding 603/604 processors and fsb bottlenecking.



      The question i'm about to ask is one of a technical, and so, i thought it would be best to come here and ask some of the Intel professionals/enthusiasts.


      Ok i'll begin.  Remember Gallatin?  Yes the 80532 designation like Prestonia and Northwood... it's the 130nm range and what some would call the 'Northwood Elite Core' since it was used as the Pentium 4 Extreme Edition line at once stage.  The difference between the Gallatin and Northwood was the Gallatin was a higher binned piece of silicon so it was better suited at overclocking, because it was more likely to get a higher overclock.  It was also likely running one or two hundred MHz lower than what the chip was rated at, to increase stability and extend longevity as desired in server parts.

      Gallatin was also unlocked on the 478 platform since it was marketed as Extreme Edition.  2mb of L3 cache was also included on the chip.  Having the higher Instruction Per Clock cycle (IPC) and added L3 cache, Gallatin was usually able to offset the performance difference between Northwood and Prescott.

      Anyway this Gallatin Extreme Edition (EE) chip had it's FSB clocked at 800mhz like the other flagship 478 processors.  Being clocked at 3.2 and 3.4 GHz i can only ask myself... How big is the bottleneck in a 3ghz Gallatin Xeon MP with it's 400mhz FSB?  and even more so... sharing a single FSB with another Gallatin chip too... sometimes up to 4 being an MP designated chip!


      With the 3ghz Gallatin server chip there was 4mb of L3, so Intel had obviously recognised the bottleneck was increasing (since up to 2.8ghz gallatins had 2mb L3)...


      Is the cache on the 3ghz chip (4mb L3) per chip able to offset the bottleneck of 4 or even 2 chips running on a single 400mhz FSB?  On the Desktop line, Northwood was introduced with 533 at 2.26ghz and then 800 at even 2.4 on Northwood with HyperThreading (HT)... although intel only introduced 800mhz with HT chips (northwood and prescott HT chips) and otherwise used 533 on all non-HT northwood and prescott cores (scaling to 3.06ghz).


      I understand that system bandwidth, memory bandwidth and throughput of a server isn't always required as it usually depends on the tasks at hand... But i ask in a sense of being used as a workstation... someone using CPU intensive applications, and maybe playing the 3D game here and there also (in their spare time, not whilst working of course ).


      I'd also like to extend on my question by asking... since 603 chips can slip into 604 boards but not vice versa...  If you put two gallatin 3ghz (400 fsb) into a 604 board capable of dual 800mhz fsb chips... does this eliminate the traditional Xeon bottleneck due to an 800fsb and there are two 400fsb chips present instead of 800..?


      I'm just curious... and i understand we're not talking dual fsb design like in later intel boards (where intel decided to spend the additional money on theirserver platforms), or like the Athlon MP boards.


      One more thing... are there any socket 603/604 platforms/chipsets supporting dual/multi fsbs?  I mean an individual FSB for each CPU like seen on some 771 boards.


      Thanks for your time guys.  I really appreciate it


      Regards, Ben

        • 1. Re: A question regarding 603/604 processors and fsb bottlenecking.

          I'd also like to add that even Gallatin DP chips ran 533 through the whole line (2.4 - 3.2) which all have HT which makes this match the Northwood HT since Intel added the higher bus spec from 2.4 and also once HT was added.  Note that Intel also included 1-2mb of L3 on these chips... Though i'm not sure if this was because they knew the 533 bus was already inadequate, or if it was to compensate for FSB sharing between dual processors.


          Gallatin EE chips in the P4 line also saw a performance increase when raising the FSB from 800 to 1066 whilst changing the CPU clock between 3.4 and 3.46 (a 60mhz difference which is nearly negligable).  Was the performance increase due to the core being held back by the bus?  Or was it Netburst just being Netburst... loving the additional bandwidth as always?  Was it maybe the core taking better advantage of DDR2 speeds with the faster FSB?  It was still a Northwood so this can't be attributed to being based on a prescott core (though that would usually result in a performance decrease anyways   Another theory could be the transition to a 775 platform/chipset...?




          Regards, Ben