Test System EPoX 8KHA AMD Athlon Thunderbird 1.2GHz / AMD Athlon4 Palomino 1.2GHz 2 x 128 MB Mushkin High Performance PC2100 DDR MSI Starforce 822 (GeForce3) IBM DPTA 372050 Deskstar ASUS 40x CDROM Alps Electric Floppy disk AMD AT 2700TX 10/100 base-T Ethernet adapter Windows 98 First Edition BIOS settings In the last reviews, we have stressed the importance of BIOS settings for overall system performance, showing up to 10% performance deltas as a function of changing the memory timing settings with otherwise unchanged system configurations. The BIOS settings in this comparison were held constant at the following settings: CAS latency: 2 tRCD: 2 tRP: 2 tRAS: 6 CMD Rate: 1T AGP: 4X Video RAM, BIOS shadow: Disabled Continuous RAM access: Enabled All other settings: Default

Athlon4 "Palomino" in perspective, photographs courtesy of AMD.

Benchmark Software Disclosure

Some of the software used for this comparison is available as free download like Expendable and SiSoft Sandra, even though the version used here is a not-yet released beta (821) revision. Caligari truSpace 4.2 was supplied a few years ago as part of a benchmarking suite by 3Dlabs. Dragon Naturally Speaking, NetShow Encoder, Photoshop 5.0, PhotoDeluxe3.1, Battlezone, Iron Strategy, Midnight GT and Outcast are part of a benchmark suite distributed by Intel upon the release of the PentiumIII. QuakeIII Arena was the commercially available retail version.

The reason for resorting to the Intel-supplied benchmarks is that they were selected to show the performance increase of the PentiumIII over the PentiumII which translates into the effects of SSE instructions being executed as opposed to a CPU without the instructions. Some of the applications are somewhat outdated, however, the value of the applications per se hardly matters in this context since the goal of this review is to show the potential performance delta of the Athlon4 compared to the Thunderbird. Admittedly, there is a 3-7% performance increase in non-SSE optimized applications, some other applications will not show any difference at all but what really counts is how well the Athlon4 performs in an SSE optimized environment.

At one Glance

All benchmarks were run at the default settings with the exception that the Intel-provided script calls for running the benchmarks in sequence to each other. This is a valid approach if the overall system performance, including I/O latencies are the goal of running the benchmarks. On the contrary, I was interested in eliminating I/O latencies as much as possible and ran a sequence of 4 runs of the same benchmark without rebooting. The first result was discarded and the other three scores averaged for a final result.