Pentium To The Extreme: Part 2

THE UNTHINKABLE did happen - but no, it was not Britain finally getting a responsible government! It was just an unexpected glitch in my new time machine design, causing Louis Burns' IDF keynote to somehow be covered well in The Inq quite a few hours before the actual keynote happened, and with me half-a-world away from IDF at that. Talk about space-and-time warp (a new combo package, including a Windoze-based control engine for guaranteed bulk supply of space-time warp viruses, bugs, worms, wormholes, black holes, space-time bloats into other dimensions etc).

Pentium 4 Extreme Edition was the star of that keynote, as our readers knew. Extra 2 MB L3 cache on top of the existing 512 KB L2, plus keeping all the nice 3.2+ GHz clock and 800 MHz FSB, plus "possibly" a reliable, well-tested repackaged XeonMP die used (sorry, this one just a baseless speculation).

I did some thinking about it - this is the king of all overclockers' CPUs! For instance, if you Peltier or vapor-chill or kryocool your P4 Extreme Edition to the extreme, but the mainboard FSB cannot keep up (dang, it stops at 1.1 GHz! - angry geek face appears), this humongous cache helps buffer the effects of having to rely on maximum multiplier value where the FSB and memory is way, way slower than the CPU. So, the real application performance scaling curve can continue a bit further upwards.

Since the die now gets dominated by a large cache area, it will be interesting to monitor where the thermal "hot-spots" on the die will be now, and whether that will affect overclockability. Also, will P4 EE need bigger/faster/noisier fans?

Finally, it seems to be that this will ultimately be a very successful product, which may spawn a whole "new" long-term product line, possibly repackaging the XeonMP versions of the each new-generation CPU into the high-end desktop like at higher clocks and faster FSB to create ultimate enthusiast processors.

So, a year or so from now, you may have a, say, "Pentium 5 Extreme Edition" (again a pure speculation or time machine garbage) in a Prescott Socket T or whatever, basically a repackaged Potomac XeonMP CPU with 1 MB L2 and 4 MB L3 cache, possibly a 1066 MHz FSB throughput!

Make no jokes about it - this looks like an excellent strategy. As mentioned before, the original XeonMP development costs will be recovered much, much faster if the effort is shared with "Pentium Extreme". It may also result in cheaper XeonMP to fight the increasingly successful Opteron. And, it will help Intel in the perennial battle for the ultimate desktop performance crown - and that still does matter. µ