Intel had been quiet on the CPU front, and seemed to be letting AMD run rampant with new Athlon 64-based processors. The AMD 64 3200+ and FX-51 opened the floodgates, but the latest Athlon 64 3000+ and 3400+ laid out an impressive line of 64-bit CPUs for many budget and performance needs. The Intel Prescott core was supposed to be the Intel challenger, but delays gave AMD the head start, and forced Intel to play catch up. The official Prescott release is finally here (in 3.4E, 3.2E, 3.0E and 2.80E models), and not only can we get a feel for the overall performance levels of the new Pentium 4-3.2E, but also a taste of the high-end power of the new Pentium 4-3.4 GHz Extreme Edition.
The Pentium 4-3.2E is built on that new 90nm (0.09-micron) Prescott core, but from the outside, it looks quite similar to the Pentium 4 Northwood. The Pentium 4-3.2E uses the current Socket 478 package, runs on the 800 MHz front-side bus, supports Hyper-Threading, includes the same basic board and heat-spreader design, and is compatible with many current Intel chipsets like the i875P and i865PE. The physical differences are relegated to a higher chip-count on the back of the CPU, which leads into one of the major architectural enhancements to the Prescott core.
Basically, Intel has taken the Northwood cache specifications and doubled them. The Prescott core includes a full 16K of L1 cache (8K for the Northwood) and a whopping 1-MB of L2 cache (512K - Northwood), and this double-play goes farther than any previous Pentium 4 core revision. Not only has the L1 cache doubled in size, but it has also moved from 4-way associative (Northwood) to 8-way for the Prescott core. The 1-MB L2 cache is also double that of the Northwood, but its specifications remain consistent at 8-way associative in 128-byte lines.
The other major shift is in manufacturing process, and the Prescott is built upon a 90nm (0.09-micron) process technology, thereby allowing Intel to reach this level ahead of AMD, in the desktop market. These are both important facets of the Prescott core, and while the doubling of cache levels may yield some performance benefits, the 90nm core technology is more integral to the Intel strategy, as it will allow a core speed ramp up far in excess of the initial 3.2 GHz.
These are the two most obvious enhancements, but Intel has made other core-level changes, including enhancements to the NetBurst architecture. Of course, whether these will be considered "enhancements" depend on your point of view. The main shift has been to extend the Pentium 4 pipeline in order to achieve higher frequencies, thereby offering higher potential performance... at least in the long run. This pipeline extension strategy has been a bone of contention between AMD and Intel, which mirrors the MHz Myth doctrine and may bring back the old IPC (instructions per clock) fracas all over again.
The deeper pipeline can also enact higher latencies, though Intel has made further NetBurst enhancements to mitigate its effect. Intel has also added 13 new SSE3 instructions used for multimedia algorithms, along with 2 new instructions for improving thread synchronization. The branch predictor and hardware prefetch have been tweaked, the Execution Core has been tuned and some latencies reduced, and SSE/SSE2/SSE3 performance has been enhanced. Many of these core enhancements are made with Hyper-Threading in mind, and to allow higher multi-threading performance through decreased hardware contention.
What this really boils down to is a "two steps forward, one step back" scenario, where Intel received noticeable frequency benefits from the 90nm process technology, but seems to have frittered away some of the immediate advantages by doubling the data cache, extending the pipeline, increasing latencies, and working to upgrade HT performance. This may yield serious payoffs in the future, but it also means that in most standard, single-threaded games and applications, the benefit of the core enhancements may not be readily apparent, at least at current Northwood frequencies. On the flip side, the ability to rapidly scale clock speeds up is a real benefit, but one that will not be felt right out of the gate.
The Pentium 4-3.4 GHz Extreme Edition is a far more standard processor release, and is simply a 200 MHz core speed upgrade to the previous Pentium 4-3.2 GHz EE model. The basic core specification remain intact, including the basic Northwood 8K L1/512K L2 cache levels, and the full 2-MB L3 cache injection found only on Extreme Edition models. This extra L3 cache is what separates the Pentium 4 Extreme Edition from the pack, and it can be quite useful in ramping up 3D gaming and multimedia performance, but is not as impressive when performing basic business tasks.
The base features of the newest Extreme Edition model are unchanged, and the Pentium 4-3.4 GHz EE is still built on the 0.13-micron process technology, uses an 800 MHz front-side bus, supports Hyper-Threading, and uses the Socket 478 package. What has changed is the overall core speed and performance, which we'll put to the test in our benchmarking section.