neighorn@qiclab.UUCP (Steven C. Neighorn) (01/15/87)
The following benchmarks were compiled from three machines I have easy
access to. Every attempt at careful measurement and control was made. My
hope is to share the results with others who might be interested in seeing
a comparison of three different AT class machines running small programs at
different clock speeds/wait states. Certainly, this is not intended to be an
ultra-scientific laboratory controlled benchmark listing, as is obvious by
my inclusion of the much maligned Norton SysInfo value. I hope everyone finds
something useful in the table that follows. Comments are more than welcome and
would be appreciated.
COMPUTER SYSTEM/SPECIFICATION/BIOS
.-------------------------------------------------.
----------------------------| IBM PC/AT | QIC 1801 | Intel 80386 |
PROGRAM OR OPERATION | 6Mhz/1 WS | 8Mhz/0 WS | 16Mhz/1 WS |
| IBM BIOS | AWARD BIOS | PHOENIX BIOS |
----------------------------|----------------|----------------|---------------|
Format 360k disk (data) | 52.00 sec | 44.50 sec | 44.30 sec |
----------------------------|----------------|----------------|---------------|
Format 1.2meg disk (data) | 85.33 sec | 72.45 sec | 59.70 sec |
----------------------------|----------------|----------------|---------------|
Copy 285k file RAM to FLOPPY| 36.20 sec | 30.62 sec | 27.80 sec |
----------------------------|----------------|----------------|---------------|
Copy 285k file FLOPPY to RAM| 20.70 sec | 15.32 sec | 19.53 sec |
----------------------------|----------------|----------------|---------------|
Copy 285k file RAM to RAM | 1.21 sec | 0.78 sec | 0.50 sec |
----------------------------|----------------|----------------|---------------|
Copy 285k file HARD to RAM | 2.80 sec | 2.30 sec | 2.00 sec |
----------------------------|----------------|----------------|---------------|
PC Mag BENCH01.EXE 512 bytes| | | |
Random Write RAM/FLOPPY | 3.30/70.36 sec | 2.64/77.23 sec | 1.04/75.00 sec|
File Read Random RAM/FLOPPY | 5.33/47.62 sec | 2.25/45.37 sec | 0.88/57.51 sec|
File Read Seqntl RAM/FLOPPY | 1.54/44.71 sec | 0.99/49.65 sec | 0.66/48.06 sec|
----------------------------|----------------|----------------|---------------|
MIP.COM (many instructions) | 11.42 sec | 5.97 sec | 3.07 sec |
----------------------------|----------------|----------------|---------------|
SIEV.EXE (of Erastothenes) | 28.39 sec | 14.83 sec | 9.08 sec |
----------------------------|----------------|----------------|---------------|
SYSINFO.COM (Norton V3.1) | 5.70 | 8.4 | 13.3 |
----------------------------|----------------|----------------|---------------|
REALSI.COM (factor PC/AT) | 3.0/1.0 | 5.9/1.9 | 10.1/3.1 |
----------------------------|----------------|----------------|---------------|
CLK.COM (timing:4.77 Mhz=PC)| 9.60 Mhz | 15.40 Mhz | 31.10 Mhz |
----------------------------|----------------|----------------|---------------|
CLOCKSP.COM (min RAM access)| 2.38 sec | 1.70 sec | 0.93 sec |
----------------------------|----------------|----------------|---------------|
87BENCH.EXE (cruncher 80287)| 7.25 sec | 5.21 sec | 2.74 sec |
----------------------------|----------------|----------------|---------------|
88BENCH.EXE (same w/o 80287)| 72.75 sec | 41.36 sec | 24.77 sec |
----------------------------|----------------|----------------|---------------|
NOTES :
1) Each test result was the average of five test runs.
2) All tests were timed with a special in-memory timer program and were
double checked with a quartz timer and, when possible, the $T option
available with the PROMPT environment variable.
3) All memory resident programs were removed before running the tests.
4) All tests were run under PCDOS 3.2.
5) All floppy drives used in the tests were manufactured by TEAC.
6) All floppy disk tests used a Maxell MD2-D floppy diskette.
7) All benchmark programs were run from RAMDISK.
8) All AT style RAMDISKS were run in EXTENDED memory using VDISK.SYS /E.
9) The IBM PC/AT had 2176k of 16bit memory. The QIC 1801 had 3072k of 16bit
memory. The Intel 80386 had 2560k of 32bit memory, and 2048k of 16bit
memory.
10) All machines used the standard Western Digital AT disk controller.
11) The IBM PC/AT used a CMI 20meg hard disk (40msec ave access).
12) The Intel 80386 used a Seagate 4051 40meg hard disk (40msec ave access).
13) The QIC 1801 used a Seagate 4038 30meg hard disk (40msec ave access).
14) All machines used BUFFERS=50 in the CONFIG.SYS file.
15) All machines used FCONSOLE 1.15 in place of ANSI.SYS.
16) The IBM PC/AT used a 80827-6, the QIC 1801 used a 80287-8, and the
Intel 80386 machine used a 80287-10.
17) MIP.COM executes 15,728,640 ADD AX,CX instructions, 1,048,576 LOOP
instructions, 64 MOV's, SUB's, DEC's, and JNZ's to control loops,
a SUB AX,AX and an INT 21. The program is designed to check total
estimated clocks versus 'real world' time.
18) REALSI.COM is an improved System Information measure from Relia. It
returns a speedup factor over a stock IBM PC and a 6Mhz IBM PC/AT.
19) CLK.COM runs timing loops, and returns the equivalent clock speed a
standard IBM PC would have to operate at to achieve the same results.
20) CLOCKSP.COM checks for minimal dRAM access. On a standard IBM PC operating
at 4.77Mhz, the program would take 10 seconds to execute.
21) The 87BENCH and 88BENCH programs are based on Savage's Benchmark.
I would be glad to discuss the test programs I used and/or the results
with anyone who is interested. I can be reached at the phone number below via
land line, the address below via U.S. Mail, or the usenet path below via net
email.
--
Steven C. Neighorn tektronix!{psu-cs,reed}!qiclab!neighorn
Portland Public Schools "Where we train young Star Fighters to defend the
(503) 249-2000 ext 337 frontier against Xur and the Ko-dan Armada"aburto@marlin.UUCP (Alfred A. Aburto) (01/23/87)
In article <379@qiclab.UUCP> neighorn@qiclab.UUCP (Steven C. Neighorn) writes: > > > The following benchmarks were compiled from three machines I have easy >access to. Every attempt at careful measurement and control was made. My >hope is to share the results with others who might be interested in seeing >a comparison of three different AT class machines running small programs at >different clock speeds/wait states. Certainly, this is not intended to be an >ultra-scientific laboratory controlled benchmark listing, as is obvious by >my inclusion of the much maligned Norton SysInfo value. I hope everyone finds >something useful in the table that follows. Comments are more than welcome and >would be appreciated. > Steven C. Neighorn: Thanks for the benchmark results. The Sieve and Savage results ( SIEV.EXE 87BENCH.EXE, and 88BENCH.EXE ) are interesting to me because I have run these on a Commodore-Amiga Turbo-Amiga ( 68020/68881 running at 14.32 MHz ). The Savage took 0.39 seconds and the Sieve 0.8 seconds at 14.32 MHz with 32-bit memory. What I am wondering about is: (1) What Language did you use, and (2) could you upload the Sieve and Savage programs because I'm not sure that I was running the same code as yourself (even though the program names are the same). I ran Gibreath's Sieve using Lattice C. The Savage was run with Absoft's 68020/68881 compiler for the Amiga. I found that IEEE single precision was quick but grossly inaccurate in running the Savage so I ran it in double precision. Thanks. Al Aburto