Interesting news from the Microprocessor Forum that LGPT is attending this week. AMD claims "killer floating-point performance" on their new K-7 processor and quantify this by saying possibly 2x the performance compared to the Pentium II on some applications.
Based on EE Times articles, it appears that there has been numerous discussions at the Microprocessor Forum regarding floating-point capabilities...very good for LGPT!
Microprocessor Forum: Microprocessor Report: AMD's K-7 set to out-muscle Pentium II
By David Lammers
EE Times
(10/16/98, 2:24 p.m. EDT)
SAN JOSE, Calif. — Intel Corp. will face strong competition from
Advanced Micro Devices Inc.'s K-7 processor in the second half of
next year, said Keith Diefendorff, editor-in-chief of the
Microprocessor Report.
“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.” |
