CPUMark 99

So what's the deal, then?

The question then, is "why did we release the article, on a part that is unreleased and does not have proper support?" Well, we were trying to dispel the incredible hype behind a closely guarded CPU secret that sounded like it was going to be revolutionary. With so much speculation going on, we hope that the article was able to shed some light on the features and technology powering the K7.

The point of this follow-up article, then is to clarify on some of the questions that have been raised about the K7 in these few days. Some of the questions, as well as answers, are taken from excerpts from readers' emails and message board posts. We have also received an ok to post something that was sent to us from an AMD employee. Please check it out.

Benchmark issues

The majority of the messages, mails, and feedback that we have been receiving centers around the benchmarks that we took and why the K7 performed so well in some of them, while performing rather dismally in others.

Let's look at some of the benchmark results that were obtained. First, please note that the K7 benchmarks were taken on a slightly different system. Typically, we like to run the benchmarks thoroughly in the comfort of our own labs, but unfortunately we could not in this case. As such, we were forced to use the system at hand, and tried to get it to similarly match the some of the systems that we used previously for other benchmarks. Given this knowledge, the benchmarks did differ more than they would had the systems (except for CPU and motherboard) been identical.

CPUMark99 results

First, let's look at the CPUMark99 1.1 result. CPUMark tests the CPU, and its immediate subsystem, which includes RAM and motherboard. We saw the K7 score a 49.1, while a PIII Xeon with 1MB of L2 cache running at 500 MHz scored a 44. However, a K6-3, running at 550 MHz thanks to Kryotech scored a 55.9! The PIII 500 came in at 38.3. This is a good showing for AMD. However, the K7 being topped by the K6-3 is strange.

A possible reason for this is the relatively low 1/3 speed L2 cache present on the K7, which may be holding back its performance. More likely, a processor with a completely new architecture may play havok on a synthetic benchmark test such as CPUMark (remember that the original CPUMark32 included with Winbench99 was replaced because of the notoriously low scores it posted for Celeron processors).