home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
RBBS in a Box Volume 1 #3.1
/
RBBSIABOX31.cdr
/
atnu
/
at80.not
< prev
next >
Wrap
Text File
|
1986-09-08
|
8KB
|
197 lines
AT 8 MHz/0 WAIT STATE PERFORMANCE TESTS
Over the past several months a number of AT-compatible
motherboards have appeared on the market. The most common of
these boards are made in Taiwan, feature jumper-selectable speed
and wait state settings, sockets for 1MB of RAM on the
motherboard, and a Phoenix BIOS. They can be bought for $500 or
less, not including RAM. This note describes a number of tests
run on one such board.
When run in the 8 MHz, zero wait state mode (8/0), the board
appears to be over five times as fast as a regular 4.77 MHz PC,
over 25% faster that the new 8/1 IBM PC AT, and over 75% faster
than the older 6/1 IBM PC AT.
Norton System Index (SI) Benchmark
This program (not public domain) is a component of Norton
Utilities 3.0. It is quite a bad indicator of overall system
performance, tending to overstate actual comparative average
performance by 50-100%. Nevertheless, the benchmark numbers are
in widespread circulation. The table below shows the SI figure
for the four modes of the motherboard:
Table 1.
Norton SI Benchmarks
Wait States % Delta
Speed (MHz) 1 0
6 5.7 7.1 24.6%
8 7.7 9.2 19.5
% Delta 35.1 29.6 61.4
The 8 MHz/zero wait state SI figure is 9.2 (times the 1.0 SI
figure for the stock 4.77 MHz vanilla PC). This performance
increase is not true for the general mix of programs.
Note that going from the older 6/1 AT machine to the 8/0
machine implies a performance increase of 61.4 percent. In fact,
the Norton SI measure may understate this performance delta, as
indicated in the next test.
SSE-V2 System Speed Evaluator Version 2.
This set of tests measures system speed across a wider set
of tasks than the Norton test. The program should replace the
Norton SI test, as it appears to generate more realistic
performance numbers. Unfortunately, the program appears to be
hard-wired for a color/graphics display (for no particular
apparent benefit). So those with a monochrome display adapter
will not be able to run the program on their system.
Shown below is the 8/0 machine test in comparison with the
PC, PC with a NEC V20 chip, and the two current flavors of the
IBM PC AT:
System Speed Evaluator Ver #2, by T. Gray
IBM PC IBM PC AT
Routine: 8088 V-20 6 Mhz 8 Mhz Yours
============================================================================
Integer Count 24 (1.0) 23 (1.0) 9 (2.7) 6 (4.0) 6 (4.0)
Real Count 29 (1.0) 26 (1.1) 11 (2.7) 8 (3.6) 6 (4.8)
Table Lookup 28 (1.0) 26 (1.1) 9 (3.1) 7 (4.0) 5 (5.6)
String Manipulation 29 (1.0) 17 (1.7) 10 (2.9) 7 (4.1) 5 (5.8)
Empty Loop 29 (1.0) 25 (1.2) 9 (3.2) 7 (4.1) 5 (5.8)
----------------------------------------------------------------------------
Total Timing 139 (1.0) 117 (1.2) 48 (2.9) 35 (4.0) 27 (5.1)
Note that overall the system shows a 5.1 times speed factor
in comparison with the stock 8088 IBM PC. The table below shows
the speed ratings for the four machine states:
Table 2.
SSE-V2 Benchmarks
Wait States % Delta
Speed (MHz) 1 0
6 2.9 3.8 31.1%
8 4.0 5.1 27.5
% Delta 37.9 34.2 75.9%
The figure of 2.9 times the 8088 PC is more in line with
IBM's own original claims for a speed-up of about 3 times with an
80286 AT in comparison with the PC.
Note here that, if anything, the overall speedup improvement
from going to the 8/0 state from the 6/1 state is even better
than indicated by the SI test: 75.9% versus 61.4% improvement.
PC Tech Journal ATPERF Benchmark Tests
In the August, 1986 issue of PC Tech Journal a number of AT
compatibility programs were introduced. Source and executable
code for these programs are available on PC Tech Journal's
bulletin board, or from local BBS systems under the file name
PCTJ0886.ARC (which contains all the issue's program listings).
The centerpiece program is ATPERF, which measures system
memory refresh overhead, as well as the wait states for
individual components of system memory: RAM, ROM and video board
RAM.
Unfortunately, the program appears to have a bug, in that
"strange" results are registered when tested in the zero wait
state mode: byte accesses that take longer than word accesses,
negative wait states, "strange" word lengths, and so forth.
The below display is for the zero wait state mode:
-------------------------
ATPERF -- PC Tech Journal AT Hardware Performance Test
Version 1.00, Copyright (c) 1986 PC Tech Journal
IBM PC/AT model 339 (8 MHz) = 1.00 for relative measurements.
Byte Word Relative
Average RAM instr. fetch: 0.266 uS 1.52
Average RAM read time: 0.141 uS 0.141 uS 2.84
Average RAM write time: 0.395 uS 0.275 uS 1.46
Average ROM read time: 0.275 uS 0.275 uS 1.46
Average Video write time: 0.803 uS 1.605 uS 1.5
Clock rate: 8 MHz Relative: 1
Refresh overhead: 12.8%
Memory Access width Wait states
RAM read Word -1
RAM write Strange 1
ROM read Word 0
Video write Byte 4
D:\CQ\DLOADS\TMP>
--------------------------
"Strange"?? "-1" wait states? Must be verrry fast BIOS here, or
program glitch.
Above test indicates 8/0 machine is about 50% faster that
the 8/1 AT. This corresponds to two cycles for a memory fetch,
versus two cycles plus one cycle wait state in the regular AT (50
percent longer to execute).
The above test also indicates it takes longer to access a
RAM byte than a RAM word, and has calculated a negative wait
state in the RAM read routine.
The problem appears to be with the program, since the below
test for the 8 Mhz/1 wait state mode displays OK:
-------------------
ATPERF -- PC Tech Journal AT Hardware Performance Test
Version 1.00, Copyright (c) 1986 PC Tech Journal
IBM PC/AT model 339 (8 MHz) = 1.00 for relative measurements.
Byte Word Relative
Average RAM instr. fetch: 0.403 uS 1
Average RAM read time: 0.401 uS 0.401 uS 1
Average RAM write time: 0.402 uS 0.401 uS 1
Average ROM read time: 0.401 uS 0.401 uS 1
Average Video write time: 0.804 uS 1.606 uS 1.5
Clock rate: 8 MHz Relative: 1
Refresh overhead: 7.2%
Memory Access width Wait states
RAM read Word 1
RAM write Word 1
ROM read Word 1
Video write Byte 4
D:\CQ\DLOADS\TMP>
---------------------------
SOURCES FOR THE BOARD
The particular system tested here was purchased from:
Allen Yu (408) 749-9935
Wysecom Inc.
1180 Miraloma Way, Suite F
Sunnyvale, CA 94086
The board will run reliably with 120ns chips.
R. Zider
Zider Brothers, San Francisco
8.11.86