Intel NetBurst micro-architecture
Hyper-pipelined technology w/ rapid execution engine and execution trace cache
8KB L1 Data Cache, 512KB L2 advanced transfer cache
55 million transistors
0.13-micron manufacturing process w/ copper interconnects
400MHz system bus
Socket-478 form factor
1.5V core voltage
Available in clock speeds of 2.0GHz and 2.2GHz
Northwood's L2 cache
From a features perspective, the only new additions to the Northwood core are the 0.13-micron manufacturing process and the larger, 512KB L2 cache. This marks the first time a desktop processor has contained more than 256KB of cache since the older Katmai Pentium III chips and K75 Athlon cores were around, and remember, neither processor's L2 cache ran at the same clock speed as the processor. Historically doubling the amount of cache has resulted in a performance gain equivalent to one processor speed grade, so it's no surprise to see Intel move in this direction. In fact, it has been speculated for quite awhile and was accidentally supported by a leak on the ASUS website mentioning support for Pentium 4s with 512KB L2 cache.
Like its predecessor, Northwood's L2 cache features a 256-bit interface and is 8-way associative. In addition, the Pentium 4 is unique in that it transfers data on each processor clock. By doubling the amount of L2 cache, Intel has ensured that Northwood's blazing memory bandwidth keeps the processor fed with data. For instance, at 2.2GHz, Northwood's L2 cache bandwidth is 70.4GB/sec (32 bytes x 1 (data transfer per clock) x 2.2GHz = 70.4GB/sec).
Besides a bandwidth advantage, Northwood's 512K L2 cache also gives Intel the edge in onboard memory. While Athlon XP contains 128KB L1 cache and 256KB L2 cache (with an exclusive architecture) for a total of 384KB of onboard cache memory, Northwood contains a 8KB data cache (and 12K execution trace cache) and 512KB L2 cache, giving it a total of 520KB cache on-chip.
As a result of the larger L2 cache, transistor count rises from 42 million in Willamette to 55 million in Northwood, however with Northwood Intel is using a denser SRAM memory cell, effectively allowing them to get double the memory size out of the same footprint. This keeps production costs down as the die size is reduced.
Northwood's system bus remains the same at 400MHz, and provides up to 3.2GB/sec of bandwidth. But as we've mentioned previously, Intel is rumored to be hard at work on increasing the bus to 533MHz. In fact, back in September we took a look at a 533MHz Pentium 4 system running at 1.6GHz. In that article we observed a roughly 9% performance increase -- allowing our 533MHz system bus 1.6GHz system to outpace a 1.8GHz Pentium 4 system with a 400MHz bus.
With Northwood's larger cache and higher clock speeds, the gap will be even more significant if Intel does indeed move to the 533MHz bus in the future.