Availability: Now

When the very first Pentium II vs. Celeron comparisons were being performed, I remember thinking that while each design had its pros and cons, each seemed to miss the mark by just a bit. The Pentium II featured a large 512K cache, but it ran at only half the speed of the CPU. A Pentium II 450 certainly runs at a 450 MHz core speed, but the CPU L2 cache itself only performs at half the core speed, or 225 MHz. The other issue is the Pentium II cache chips are actually located on the PCB board itself, making them much harder to cool than if they were integrated into the CPU core itself. The initial wave of Slot 1 Pentium III CPUs also had the same design as the Pentium II, with their off-chip cache running at half the core speed of the CPU.

The second generation Intel Celeron design incorporated the L2 cache directly into the CPU core itself, and was the first Intel core to feature an on-die L2 cache design. The Celeron L2 cache runs at the full speed of the CPU, with none of the half-speed limitations of the Pentium II and III CPUs. This integrated design also allowed more effective cooling and helped the Celeron become the most overclockable CPU in history. If there was a fly in the ointment, it was the relatively small 128K cache of the Celeron. The Celeron 128K of L2 cache is only 1/4 the size of the Pentium II/III CPUs (512K), which led to lower performance with many business applications. Games traditionally don't make heavy use of the CPU cache, but demanding business apps like PhotoShop or web server duties really show off the Pentium II/III outperforming a Celeron of the same speed.

This discussion makes me think back to the grand old days of Socket 7 where the CPU didn't even have an on-chip L2 cache. The cache was instead incorporated on the motherboard, and depending on configuration, would be 64K, 128K or 256K of SRAM memory. At the time, this usually got quite expensive as you increased the cache size. Performance testing was consistent in finding that moving from 64 to 128K provided the best bang for your buck, but that 256K still offered the best overall performance. In later Socket 7 revisions, the motherboard cache levels were raised to 512K, but overall performance did not rise correspondingly to the increase in cache size. In the usual bang for the buck equation, 256K was cited as the prime spot on the cache scale.