"If I were Intel, I would be a little nervous," Diefendorff said at a day-long Thursday seminar focused on comparing high-performance microprocessor designs, at the Microprocessor Forum here.

With its current-generation K-6 processor, AMD has generally lagged behind Intel's fastest parts by two or three speed grades, which kept AMD out of the most profitable part of the microprocessor market. But the K-7 "could allow AMD to play at the very top end of the Intel product line, and be used even in multiprocessor system," Diefendorff said.

However, Intel is spending large amounts of money to make sure that the applications developers adopt the KNI (Katmai New Instructions) extensions next year. "KNI is going to be the programming model. It is going to happen," he said. For certain applications, such as the Guassian blurring used in image manipulation, the KNI extensions to the MMX instruction set will result in greatly improved performance.

However, until KNI is established, the 3DNow instructions are superior to what Intel is currently offering. Moreover, the floating-point performance of the K-7 design from AMD promises to be far superior to that offered in the Pentium II-class processors from Intel, which are based on an older architecture.

"The K-7 is a killer machine when it comes to floating point. It is a fully pipelined machine, and so it can do a multiply-and-add function in the same clock. This design could really kick butt in floating point," he said.

Asked by one engineer to quantify the K-7's advantage over Pentium II processors in terms of floating point, Diefendorff said, "there certainly will be applications where it delivers 2x [i.e., double] on a per-clock basis. If AMD is able to achieve the 500-MHz performance goal — and they seem very very confident of that — then you just about are not going to do any better than this in an X86 architecture. The floating point is not going to match the Alpha, but that is because of the X86 architecture."