home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
Software Vault: The Diamond Collection
/
The Diamond Collection (Software Vault)(Digital Impact).ISO
/
cdr29
/
dfb16b.zip
/
READ.ME
< prev
next >
Wrap
Text File
|
1994-12-07
|
12KB
|
266 lines
DDDDDD FFFFFFF BBBBB
DDDDDDD FFFFFFF BBBBBBB
DD DD FF BB BB
DD DD FFFFF BBBBB
DD DD FFFFF BB BB
DDDDDDD FF BBBBBBB
DDDDDD FF BBBBB
Data Flow Benchmark V 1.6
(c) 1994 by D.Engert
1. Legal stuff
There is no warranty. Use this software on your own risk. Due to the
complexity and variety of today's hardware and software which may be
used to run this program, I am not responsible for any damage or loss of
data caused by use of this software. It was tested very well and is
expected to work correctly, but nobody can actually guarantee this for
any circumstances. And because this software is free, you get what you
pay for ...
This program can be used freely for private or educational purposes. If
you want to use it for commercial purposes or find any bugs or have
suggestions about further enhancement, please contact the author.
Author: Detlef D. Engert
Gruentenweg 14
D-90471 Nuernberg
Germany
Fax: +49-911-861319
Mailer: +49-911-861319 UTC 18:00 - 23:00
EMail: 2:2490/1145.9@fidonet
2:2490/2576@fidonet
engert@ibm.net
2. Purpose and intent of this program
Todays hardware gets more and more powerful but more complicated too. Modern
motherboards using up to date chip sets may turn out to be very difficult to
configure. And to make things worse, there are different manufacturers of CPU
chips besides Intel now with new features and options. The memory subsystems
implemented on these motherboards are even harder to configure, taking into
consideration different cache strategies, RAM speeds and access modes.
Beyond the core of any computer system lay the peripherals (video, magnetic
storage..) connected by a variety of bus implementations like ISA, EISA, VLB or
PCI. Chip sets used on these peripherals are often of even higher complexity
than the computer core.
Even skilled users are often overwhelmed by the sheer complexity and variety
of options offered. Nobody will them the real power available to them by a
given computer system using a particular configuration set. How should a user
optimize his or her computer or how should a buyer choose between similar
looking components based on hard facts ? May be this program will help you !
3. What will this program offer ?
Let's have a look on the output (framed) of the current version run on my own
machine.
Machine configuration:
GigaByte Motherboard,
Intel 486DX-50 CPU,
SiS 411 EISA/VLB chip set,
EISA/VLB bus system running at 50MHz,
256 KByte of 12ns cache RAM, 2 Banks a 16 MByte of 70ns DRAM,
Diamond Viper Pro Video P9100 video board, 4 MByte VRAM
All configuration settings are optimized for maximum throughput. That gives
the following results:
+-----------------------------------------------------------------------------+
| Data flow benchmark v1.6 |
| |
| copyright (c) 1994 D.Engert; partially based on a DOS implementation by |
| A.Stiller, c't |
| |
| Processor : Intel/AMD 486DX |
| Clock : 50.0 MHz |
| Coprocessor : present |
| Internal bus width : 32 bit between processor and primary cache |
| External bus width : 32 bit between primary and secondary cache |
| DRAM page size : 16 KByte, interleaved |
| MMU cache : 32 entries 4-way set associative, 4KByte per entry |
| Primary cache : 8 KByte 4-way set associative |
| Secondary cache : 256 KByte direct mapped |
| Cache line size : 16 Bytes |
| Cache strategy : write through |
+-----------------------------------------------------------------------------+
These figures are quite selfexplanatory. Type and speed of the CPU are detected,
the width of the data paths between CPU core and primary cache - typically
located on the same chip as the CPU core - and between the core or primary
cache and the secondary cache (if present) or main memory.
The program next tries to determine the effective page size, if page mode is
implemented by the chip set. The following 3 lines show information about the
address translation lookaside buffer (MMU cache), the primary and secondary
cache. Size and associativity are checked, the length of a cache line is
determined and the strategy used by the cache subsystem (write-through or write-
back) is sensed.
+-----------------------------------------------------------------------------+
| Data flow and bus performance memory |
| |
| -- memory -> CPU --------- |
| Maximum 4K FETCH (Hits) : 19.5µs ( 973c) => 200.7MB/s (0.25c/Byte) |
| 4K FETCH (Miss+Hit) : 29.2µs ( 1461c) => 133.6MB/s (0.37c/Byte) |
| Minimum 4K FETCH (Misses) : 59.7µs ( 2985c) => 65.4MB/s (0.76c/Byte) |
| Maximum 4K LODSD (hits) : 83.2µs ( 4160c) => 49.3MB/s (1.02c/Byte) |
| 4K LODSD (miss+hit) : 97.6µs ( 4882c) => 42.0MB/s (1.19c/Byte) |
| Minimum 4K LODSD (misses) : 134.5µs ( 6730c) => 30.4MB/s (1.64c/Byte) |
| -- CPU -> memory --------- |
| Maximum 4K STOSD (hits) : 82.1µs ( 4106c) => 49.9MB/s (1.00c/Byte) |
| Minimum 4K STOSD (misses) : 82.1µs ( 4108c) => 49.9MB/s (1.00c/Byte) |
| -- memory -> memory ------ |
| Maximum 4K MOVSD (hits) : 65.5µs ( 3275c) => 62.6MB/s (0.80c/Byte) |
| 4K MOVSD (miss+hit) : 129.9µs ( 6498c) => 31.5MB/s (1.59c/Byte) |
| 4K MOVSD (clean) : 202.7µs ( 10142c) => 20.2MB/s (2.48c/Byte) |
| 4K MOVSD (dirty) : 141.3µs ( 7071c) => 29.0MB/s (1.73c/Byte) |
| Minimum 4K MOVSD (misses) : 203.2µs ( 10166c) => 20.2MB/s (2.48c/Byte) |
+-----------------------------------------------------------------------------+
These are the performance figures of the CPU <--> memory data path.
There are four disciplines:
- opcode fetch
- data load
- data store
- data move.
Depending on the discipline several scenarios are tested (denoted in paren-
theses):
- hits in all memory caches (hits)
- hit in secondary cache, but not in primary (miss+hit)
- hit with replace in clean secondary cache, no write back necessary (clean)
- hit with replace in dirty secondary cache, write back carried out (dirty)
- misses in all caches (misses)
The first (hits) should give maximum performance down to the last (misses) with
minimum speed.
The test transfer size depends on the cache and page sizes. In the case above
it is 4KByte.
There are four result columns:
- absolute time needed for one test of the mentioned size
- the same in clock cycles
- the resulting transfer speed in MBytes per second
- the cost of the operation in cycles per Byte
+-----------------------------------------------------------------------------+
| VIO info : XGA, 0 KByte video memory |
| Device info : manufacturer Weitek, chip set W5186, 1 MByte video memory |
| Screen : 1280x1024x256 |
| Aperture : 4 MByte @ 0xC0800000 |
| Bus width : 32 bit between CPU and video memory |
+-----------------------------------------------------------------------------+
The information about the video system is queried from different parts of OS/2,
so there may be different figures for the same item. That depends more or less
on how careful the developer of the video drivers did the job...
+-----------------------------------------------------------------------------+
| Data flow and bus performance video |
| |
| -- Video -> CPU ---------- |
| Maximum 4K LODSD (Hits) : 390.5µs ( 19536c) => 10.5MB/s (4.77c/Byte) |
| Minimum 4K LODSD (Misses) : 390.4µs ( 19528c) => 10.5MB/s (4.77c/Byte) |
| -- CPU -> Video ---------- |
| 4K STOSD : 82.3µs ( 4116c) => 49.8MB/s (1.00c/Byte) |
| -- Memory -> Video ------- |
| Maximum 4K MOVSD (Hits) : 86.0µs ( 4301c) => 47.6MB/s (1.05c/Byte) |
| 4K MOVSD (Miss+Hit) : 133.4µs ( 6674c) => 30.7MB/s (1.63c/Byte) |
| 4K MOVSD (Clean) : 536.0µs ( 26812c) => 7.6MB/s (6.55c/Byte) |
| Minimum 4K MOVSD (Misses) : 539.5µs ( 26986c) => 7.6MB/s (6.59c/Byte) |
| -- Video -> Memory ------- |
| Maximum 4K MOVSD (Hits) : 536.0µs ( 26812c) => 7.6MB/s (6.55c/Byte) |
| Minimum 4K MOVSD (Misses) : 539.5µs ( 26986c) => 7.6MB/s (6.59c/Byte) |
| -- Video -> Video -------- |
| Maximum 4K MOVSD (Hits) : 562.1µs ( 28118c) => 7.3MB/s (6.86c/Byte) |
| Minimum 4K MOVSD (Misses) : 678.6µs ( 33946c) => 6.0MB/s (8.29c/Byte) |
+-----------------------------------------------------------------------------+
This is the same as above, obviously the discipline opcode fetch is left out,
but there are more transfer data paths.
The figures for data store and move from primary cache into video memory are
more or less senseless on local bus systems and coprocessed video cards, but
give at least an idea how careful these buses are implemented.
I didn't comment the actual figures, because each - and most probably your -
system is different. Compare yourself, I only say that this system is a fast
one in its category...
4. How do I start this program ?
That's easy: go to a command line and type
DFB [options]
The following options are currently implemented:
/NOV[ideo] : suppress video testing
/CC:number : set country code to number, default is from CONFIG.SYS
49 : Germany (deutsch)
else : international (english)
/MORE : stops output after each section to ease reading
/DMP : dump test values to stderr
may be redirected to file via 'DFB /DMP [...] 2>filename'
Options are not case sensitive !
The commands 'help DFB0000' or 'DFB /?' will give you the same information about
the usage of the latest version of DFB.
If you start DFB in a full screen session, video testing will be left out,
since there is no video aperture available. Better use a windowed OS/2 box.
To access memory and video for testing, the device driver SSMDD.SYS must be
loaded. It is part of MMPM/2 and included in the distribution of DFB also. Be
sure, the statement
DEVICE=[path]\SSMDD.SYS
is in your config.sys file. DFB will tell you if it is not. In this case, only
a small part of the DFB functionality is available (basic CPU type checking).
If you create a minimum boot disk with floppy support only, there is enough
room left to put SSMDD.SYS and DFB.EXE onto it too. So you may enter your
favourite computer shop and check out the different machines offered by a mere
boot from this floppy disk.
5. Is there danger to use this program ?
Yes, there is !
First, I am human so I am error prone :-)
Second, DFB goes to the bones of your computer. Therefore I don't guarantee that
it will interact with any running program or any active device in a totally
harmless manner. If you plan to start DFB, I recommend to stop any other running
user process and wait until all sensitive devices are idle. That is not a must
but reduces any risks (I run DFB parallel to my communications software, active
CD-ROM and active audio system).
6. A whish of the author
Since I have access only to Intel 486 class machines, I would appreciate if you
run the tests using the dump option and drop me an email with a description
of your system and the resulting dump output. If you can't reach me through
Fidonet you may send your message to Compuserve 100275,3253. I am interested in
any non i486 CPU.
Everybody who provides me new information may consider him/herself as a
registered user of a forthcoming shareware version (if there will be one...)
---