commentary What CPU technologies will affect the way your next PC performs?

Coincidences can be a pleasant surprise, particularly when a new AMD Athlon64 3200+ and Intel P4 3.2GHz “Prescott” both land on the Lab’s doorstep at the same time. It then got better — we also received a loaner P4 Extreme processor, also clocked at 3.2GHz.

We have been dying to compare the performance of these babies for a while now and the fact that they arrived at the Lab for a head to head was a lucky break.

Of course it was not all plain sailing, the Intel D875PBZ board initially delivered was a rev. 301 with a beta BIOS. Even when we upgraded the BIOS the supplied software installation disk claimed the board was not an Intel product. We were then supplied with a rev. 302 board and everything was fine.

Not to be outdone, the MSI K8T Neo Version 1.0 motherboard supplied with the Athlon64 was flaky as well and had to be replaced — a most unusual state of affairs.

Let’s take a look at the three processors and compare any similarities and differences, starting with the two Intel CPUs.

The easy one first — the P4 Extreme — is based on the current Northwood core using 130nm fabrication technology so it has an 8kB L1 cache and a 512kB L2 cache, nothing different here, however Intel managed to fit a new 2MB L3 cache onto the chip. That’s pretty much all there is to it, however the extra L3 cache should give the CPU a noticeable performance advantage over a straight Northwood P4 and is targeted squarely at gamers.

The Prescott is a more substantial upgrade using a 90nm fabrication and double-sized L1 and L2 data caches, the Instruction Trace cache increases from 12k to 16k micro OPs, which should reduce latencies and, in the former case, heat. This has been offset however by a longer integer pipeline of 31 stages compared to the Northwood’s 20 stages and the Athlon64’s 12-stage ALU. In addition, the Prescott has another 13 SSE instructions (this iteration called SSE3) and, according to Intel, improved Hyper-Threading, improved NetBurst architecture and better power management based on the new M mobile processor technology.

Now adding the extra stages to the Prescott’s pipeline should result in a noticeable drop in performance, quite simply if this pipeline stalls it’s going to take longer to start it up again compared with the Northwood. However, our initial testing indicates that the Prescott’s performance is not all that different from a similarly clocked Northwood. How can that be? Simply put, Intel has compensated by increasing the accuracy of the Branch Predictor.

The Northwood core was expected to top out at around the 3.8GHz mark and with the latest clocked at 3.4GHz there is not a lot of headroom, the Prescott however is supposed to be good for up to 5GHz.

The AMD Athlon64 3200+ is actually clocked at just 2GHz, compared with the P4’s 3.2GHz, yet in our preliminary testing it certainly gives the faster Intel CPU’s a run for their money. As the name suggests, the Athlon64 is a 64-bit processor with sixteen 64-bit integer registers, whereas the P4 is a 32-bit processor; but unlike Itanium the Athlon64 is backwardly compatible with 32-bit x86 instructions. In addition the Intel P4 only supports 32-bit addressing, so 4GB is the memory ceiling. The Athlon64 has 40-bit physical addresses which is 256 times larger. The Athlon64 also has an integrated memory controller, it no longer resides in separate Northbridge chip, however the controller is only capable of handling up to four registered DIMM’s, each up to 4GB, so the real memory limit is 16GB, but hey that’s a big improvement.

Steve Turvey is Lab Manager of the RMIT IT Test Labs.

This article was first published in Technology & Business magazine.
Click here for subscription information.