Computer workstation speeds Page: 7 of 13
This report is part of the collection entitled: Office of Scientific & Technical Information Technical Reports and was provided to UNT Digital Library by the UNT Libraries Government Documents Department.
Extracted Text
The following text was automatically extracted from the image on this page using optical character recognition software:
The SPEC92 ratings generally agree with the speeds measured for the Sandia pro-
grams. Correlation coefficients are commonly used to measure agreement between pairs
of data sequences. Table 2 lists the six correlation coefficients for the two sequences of
SPEC92 ratings paired with the three sequences of speeds for the Sandia programs. A cor-
relation coefficient near 1 indicates strong agreement between the members of the pair.
Table 2. Correlation coefficients between the SPEC92 ratings of the tested com-
puters and the speeds of running the Sandia programs.
SPECint92 SPECfp92
not vectorized, 64 bit, twafer [12] 0.79 0.92
partly vectorized, 32 bit, current [15] 0.74 0.91
highly vectorized, 64 bit, ovend [11] 0.79 0.91
Nevertheless, SPECmarks imperfectly indicate actual performance. Figure 2 com-
pares the SPECfp92 ratings with the normalized speeds of running the highly vectorized
Sandia program. This pair of data sequences has a 0.91 correlation coefficient, yet some
computers with the same SPECmark have speeds that vary by a factor of two.
Figure 2 thus demonstrates the need for independent testing. "Many industry experts
now tend to reject [standardized] tests because it is relatively easy to skew results by com-
piler tuning" [14]. The SPEC92 ratings are thought to be particularly unreliable, and are
being replaced by '95 versions.
Other Computers
Many fast computers were not tested, and many computers that will become available
in 1996 could not be tested. Their performance on the Sandia programs can be extrapolated
from similar machines using processor Hertz and SPECmarks. Unfortunately, comparison
between old and new computers is difficult because older computers have SPEC92 ratings
while newer computers have '95 values.
Table 3 shows SPEC95 ratings that have been measured for some Intel [5] and SGI
[4] computers, or have been estimated for systems built around some anticipated chips [3].
The table's SPEC95 ratings are relative to a Sun 10/40 (SuperSparc @ 40 MHz), which ap-
proximates Figure l's normalization. Reconciling the two sets of data (the SPEC95 ratings
and the speeds of running the Sandia programs) for the Intel and SGI computers suggests,
very roughly, that SPEC95 ratings should be scaled by 0.5 to approximate the speeds mea-
sured for the Sandia programs. This places Table 3's fastest computers between 6 and 10
on Figure l's scale.
General Background
Personal workstations are nearly as fast as mainframe supercomputers because much
of a computer now fits on a single chip. This inexpensively eliminates the delays of trans-
mitting high frequency signals through wires and across circuit boards. Given this degree
of miniaturization, the primary determinants of speed are then (1) clock speed, (2) com-
puter architecture, (3) memory access and (4) word length.6
Upcoming Pages
Here’s what’s next.
Search Inside
This report can be searched. Note: Results may vary based on the legibility of text within the document.
Tools / Downloads
Get a copy of this page or view the extracted text.
Citing and Sharing
Basic information for referencing this web page. We also provide extended guidance on usage rights, references, copying or embedding.
Reference the current page of this Report.
Grcar, Joseph F. Computer workstation speeds, report, June 1, 1996; Livermore, California. (https://digital.library.unt.edu/ark:/67531/metadc674291/m1/7/: accessed April 25, 2024), University of North Texas Libraries, UNT Digital Library, https://digital.library.unt.edu; crediting UNT Libraries Government Documents Department.