Intel Pentium III 1.2GHz 0.13-micron Tualatin: The Celeron of the Future

The future of the Tualatin

We’ve mentioned countless times that the Pentium III is a dying breed, soon to be replaced entirely by the Pentium 4 in the desktop market (the Pentium 4 is too big and too hot for a mobile CPU).  In fact, Intel doesn’t really want people buying the desktop Tualatin.  At the end of August 2001 the 1.2GHz Pentium III (Tualatin) will be priced around $260 in 1,000 unit quantities; at the same time, the 1.5GHz Pentium 4 will be priced at $130 in similar quantities.  The extremely costly nature of the Tualatin-based Pentium III will keep it out of the hands of many, since for the same price you can easily get a Socket-A motherboard and a CPU.  The desktop Pentium III will most likely max out at 1.2GHz, with the only Tualatin based parts being the 1.13GHz and 1.2GHz CPUs. 

However, the real goal of bringing the Tualatin core to the desktop market is to pave the way for the true home of the Tualatin, the Celeron.  By the end of this year Intel is expected to debut the new Celeron at 1.2GHz.  Over the past two years the Celeron has lost a great deal of its competitive edge.  Only recently did it gain the use of the 100MHz FSB, and its 4-way set associative L2 cache held its performance back even more.  The AMD Duron, currently priced at extremely affordable levels, is clearly a superior choice in terms of price and performance.  A Tualatin-based Celeron could turn things around for the price/performance value of Intel’s entry-level processors. 

In continuing the tradition of the Celeron, it seems as if Intel is going to cripple the Tualatin core much more than necessary in order to clearly distinguish it from the flagship Pentium 4.  The current roadmaps state that the 0.13-micron Tualatin-based Celeron will only feature a 128KB L2 cache (potentially only 4-way set associative, but, hopefully, it will feature an 8-way set associative L2 cache, which would result in a much higher hit rate), and it will still only officially use the 100MHz FSB.  These two limitations will significantly hamper the performance of the Celeron. 

First of all, with only 128KB L2 cache, the Celeron will definitely have a hard time performing well in a lot of the new applications being released.  As we’ve already proven, Windows XP eats up a considerable amount of performance over Windows 2000, as do the XP versions of popular applications (for example, Office XP vs. Office 2000).  These applications do have larger footprints that may easily exceed the Celeron’s 128KB L2 cache, giving users even more incentive to upgrade to the Pentium 4, which will eventually be given a full 512KB L2 cache by the end of this year.  

Secondly, with only 800MB/s of FSB bandwidth (100MHz FSB), the Tualatin’s Data Prefetch Logic will be even less useful than it is on the Pentium III.  Remember, the less bandwidth that the DPL can use to prefetch data, the less it can prefetch into L2 cache.  This brings up another limitation of the 128KB L2 cache; there is less room for prefetched data courtesy of the DPL. 

If Intel truly wants the Celeron to be competitive with the Duron (and the potential of a Duron MP processor), they need to release a Celeron with the Pentium III’s Tualatin core.  A Tualatin-based Celeron with an 8-way set associative 256KB L2 cache, and a 133MHz FSB could give the Duron a run for its money.  The only benefit of a 128KB L2 cache Tualatin-based Celeron would be the low manufacturing costs, courtesy of an extremely small 0.13-micron die size.