The World of Computer Software

home *** CD-ROM | disk | FTP | other *** search

/ The World of Computer Software / World_Of_Computer_Software-02-385-Vol-1of3.iso / z / zdbench.zip / README < prev next >

Wrap

Text File | 1992-07-15 | 19KB | 526 lines

ZD Labs Benchmark Series Release 1.1 - Help File DESCRIPTION OF MENU SELECTIONS ▒▒▒ File ▒▒▒ ▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄ Load Comparisons The Load Comparisons selection allows you to set which machine's results in the database you wish to compare to the unit under test. To choose a comparison machine, select Load Comparisons from the File menu, select a machine/variant combination, select the date/time for the results for that machine and review the Machine Description before accepting the comparison machine. ▄▄▄▄▄▄▄▄▄ Batch Run The Batch Run selection allows you to run the following tests automatically: . All Processor Tests . DOS Disk Access Test . All DOS File Access Tests . All Video Text Mode Tests . All Video Graphics Mode Tests A scrolling status window presents the test names and results as they are run. You can view the results after the cycle completes by accessing the View Menu. If an error occurs, an error window will appear for 2 seconds and the next test will be run. ▄▄▄▄ Quit When quitting the benchmark program, there is an option to commit the results to the database. As soon as a test is completed, the results are written to a database file. Upon exiting the program, the commit option records these results permanently to the database. If you do not wish to permanently save the results from a session, you may turn off the commit check box before exiting. Then, the next time the program is executed, you will be given the option of using the uncommitted results from the previous test run or deleting those results from the database. ▒▒▒ View ▒▒▒ ▄▄▄▄▄▄▄ Results The results screen allows you to scroll through all the results for the tests that have been run, compare the results to other machine results in the database, graph the results for each test, save the results to a file or print a hardcopy. ▒▒▒ Performance ▒▒▒ ▄▄▄▄▄▄▄▄▄▄▄▄▄ Processor ... [NOTE: Since the execution speed of microcomputers is incredibly fast, the functions described in the Processor Performance Tests are each performed for 10 seconds to produce the published results. While the number of iterations varies from test to test, they are consistent within each test when testing different machines.] ■■■ Instruction Mix ■■■ The Instruction Mix benchmark test times a series of assembly language instructions specific to the 8088, 80286, 80386 or 80486 chip. This test shows how the CPU operates as a result of the interaction of the bus, processor, system memory, and motherboard architecture. A larger number indicates better overall computer performance. ■■■ 128K NOP Loop ■■■ The 128K NOP Loop benchmark test loops through 128K of memory filled with NOP (No Operation) instructions. This test is useful for comparing systems equipped with a processor RAM cache. A higher number indicates better cache system performance. ■■■ Do Nothing Loop ■■■ The Do Nothing Loop benchmark test times the execution of a loop containing one NOP instruction. The higher the number, the more efficient the computer, i.e. less resources are required for system housekeeping (refreshing memory, servicing interrupts, etc.). ■■■ Integer Add ■■■ The Integer Add benchmark test times the execution of a loop containing an ADD instruction. ■■■ Integer Multiply ■■■ The Integer Multiply benchmark test times the execution of a loop containing an IMUL instruction. ■■■ String Sort and Move ■■■ The String Sort and Move benchmark test times the execution of a bubble sort performed on 200 random strings containing 16 characters each. This test is useful in testing systems with a processor RAM cache, as sorting requires reading and writing of the same data repeatedly. Systems with write-back caches will usually perform better on this test than will systems with write-through caches. ■■■ Prime Number Sieve ■■■ The Prime Number Sieve benchmark test times the execution of a routine to find the prime numbers between 0 and 8190. ■■■ Floating Point Mix ■■■ The Floating Point Mix (also known as the Floating Point Calculation Without Coprocessor) benchmark test sets up a floating point emulation program in RAM and then exercises the processor and tests RAM access speeds during floating point calculations. The floating point processes performed are identical to those used in the Coprocessor Test. ■■■ Math Coprocessor ■■■ The Math Coprocessor benchmark test exercises the math coprocessor using the same floating point calculations used in the Floating Point Mix Test. The test can be used to analyze the speed differences of coprocessors in different machines. Additionally, the processing speed a machine gains by using a coprocessor can be seen by comparing its Floating Point Mix scores with its Math Coprocessor scores. ▄▄▄▄▄▄▄▄▄▄ Memory ... ■■■ Conventional Memory Read/Write ■■■ These tests measure random access memory speed of conventional memory performing memory reads and writes. The program allocates 64K of memory and treats it as a series of 64 byte records. 16,384 random records are then either written from or read into local memory. ■■■ Expanded Memory Read/Write ■■■ These tests measure random access memory speed of expanded (Lotus, Intel, Microsoft) memory performing memory reads and writes. The program allocates 128K of memory and treats it as a series of 64 byte records. 16,384 random records are then either written from or read into local memory. ■■■ BIOS Extended Memory Read/Write ■■■ These tests measure random access memory speed of extended memory performing memory reads and writes. The program allocates 128K of memory and treats it as a series of 64 byte records. 16,384 random records are then either written from or read into local memory. Due to advances in the 80386 and 80486 processor chips, the results obtained using the Extended Memory tests with these processors are not representative of what you'll find with all programs using extended memory on 80386 and 80486 machines. The test uses the BIOS function for switching in and out of Protected Mode, and the BIOS call is not commonly used to perform this function on these machines. The test's name indicates its specific use of the BIOS functions. ▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄ DISK ACCESS ... ■■■ BIOS Disk Seek ■■■ The BIOS Disk Seek benchmark test measures mechanical track to track disk drive access times. Both sequential and random access is tested. The test uses the BIOS Interrupt 13h. It cannot be used to test floppy diskettes or mass storage media that do not have an Interrupt 13h interface. The test involves 1,000 sequential seeks (alternating between cylinders 0 and 1) and 1,000 random seeks. Caching disk controllers and certain SCSI adapters which do not force the disk mechanism to move for disk seeks will yield unrealistically fast times. The result is the average access time per track. ■■■ DOS Disk Access ■■■ The DOS Disk Access benchmark test measures the time necessary to perform 1000 read requests at random locations on the disk using the DOS Interrupt 25h. This test should work with any device that DOS recognizes as a disk. The test reports the total time required to complete the requests in seconds. This is an occasion in the Bench 1.1 suite where large numbers are preferable to small ones. [NOTE: If a seek error occurs in either of the above tests a message will appear for a few seconds, then a replacement sector will be chosen and the test will continue. This is NOT necessarily an indication of a bad disk drive. These tests access sectors which may have been legitimately locked out during the formatting of the drive. If you suspect a problem with your drive, use a diagnostic program to test the drive.] ▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄ DOS File Access ... The DOS File Access (Small and Large Records) benchmark test measures disk throughput as a result of mechanical disk drive speed, hard disk controller function, and bus speed. This benchmark tests file creation, writing, and reading (sequentially and randomly) for a 256K file with two buffer sizes. The test may be used to show the effects of disk caching by running twice: once with and once without a cache. Fast times for a particular record length indicate that applications using similarly-sized records will perform well on the system. ▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄ Variable Size (Small and Large) Records ... The DOS Variable File Access (Small and Large Records) benchmark test is similar to the DOS File Access test and has been designed primarily for testing hardware and software disk caches. The small record size is 200 bytes and the large record size is 2048 bytes. File sizes range from 300 to 19,200 records for small records and 30 to 1,920 records for large records. ▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄ EGA/VGA Text ... ■■■ Unscrolled BIOS Write ■■■ The Unscrolled BIOS Write benchmark test times the writing of data to the screen using interrupt 10h BIOS calls. Fast times are advantageous for programs that display large amounts of data in a non scrolling fashion. ■■■ Scrolled BIOS Write ■■■ The Scrolled BIOS Write benchmark test times the writing of data to the screen using interrupt 10h BIOS calls. Once the screen is filled, the test scrolls through an entire screen by adding one new line at a time. This test writes the same amount of data as the non-scrolling test. The difference between the two times indicates the overhead attributable to scrolling the screen. Fast times are advantageous for application programs that frequently cause the screen to scroll. [NOTE: The pair of BIOS write tests give a good indication of video BIOS speed. If the video BIOS is shadowed, (loaded in fast system memory), throughput will be higher for these tests.] ■■■ Direct Screen Write (8-bit, 16-bit, 32-bit) ■■■ The Direct Screen Write tests write data to the video adapter screen memory using the MOVSB (8-bit), MOVSW (16- bit) and MOVSD (32-bit, for machines with at least an 80386 processor) instructions. For a properly installed 16-bit VGA card, the 16-bit test will yield about twice as much throughput as the 8-bit test. The 32 bit test will have only about 25% more throughput than the 16-bit test, however, due to the reduced number of memory cycles required to access the same data. This tests provide the best measure of the text throughput of a display adapter. ▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄ EGA/VGA Graphics ... ■■■ Write Mode 0 Fill (8 bit, 16 bit, 32 bit) ■■■ The Write Mode 0 Fill tests fill the video adapter's memory with varying colors using the STOSB (8 bit), STOSW (16 bit) and STOSD (32 bit, for machines with at least an 80386 processor) instructions with EGA/VGA write mode 0 and all bit planes enabled. Commonly used by graphics software applications, it is one of the best measures of video adapter throughput in graphics mode. ■■■ Memory to Screen Bitblt (16 bit, 32 bit) ■■■ The Memory to Screen Bitblt tests create screen segments, then save these screen images into system memory one video plane at a time. The timed portion of the test copies the screen segments in system memory back to screen memory using the MOVSW (16-bit) and MOVSD (32-bit, for machines with at least an 80386 processor) instructions, one video plane at a time. Very fast video cards will allow you to see the text appear in a "marquee" effect on the screen. Memory to Screen Bitblts are used when updating the images on a display. ■■■ Screen to Memory Bitblt (16 bit, 32 bit) ■■■ The Screen to Memory Bitblt tests are similar to the Memory to Screen Bitblt tests except that the timed portion of the test measures the throughput of copying information in screen memory to system memory using the MOVSW (16 bit) and MOVSD (32 bit, for machines with at least an 80386 processor) instructions, one video plane at a time. Since system memory is being updated, there is no visible activity for the duration of this test. Screen to Memory Bitblts are used when modifying or saving screen images. ■■■ Screen to Screen Bitblt (16 bit, 32 bit) ■■■ The Screen to Screen Bitblt tests divides screen memory into 4 quadrants and fills the first three with uniform fill patterns of different colors. The test repeatedly moves screen data from the first three screen quadrants to the fourth one using the MOVSW (16 bit) and MOVSD (32 bit, for machines with at least an 80386 processor) instructions with write mode 1. Screen to Screen Bitblts are used when moving objects from one area of the screen to another. ▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄ Battery Rundown The Battery Rundown test is designed to exercise a laptop computer's battery system in a worst-case scenario. After charging the computer's battery according to the manufacturer's specifications and disabling all power conservation features, the test is started. The test teletypes 10 screens full of information to the display and then writes a 15K file to disk with a time stamp of elapsed time. This continues until the battery runs out of power. ▄▄▄▄▄▄▄▄▄▄▄▄▄▄ Time the Timer The benchmark programs calculate elapsed time using DOS calls. Some machines may not report a correct time running a version of DOS not configured for the machine. If this is the case, the results you obtain using these tests may be skewed. Time the Timer allows you to measure the accuracy of your system's timer by comparing it to some external source (such as a stopwatch). ▒▒▒ Compatibility ▒▒▒ ▄▄▄ VGA The compatibility test for VGA hardware performs several VGA register-level functions which exercise the different subsystems of a VGA or compatible display board (including the VGA controller chip, video memory, and RAMDAC). Since the test was designed with the original VGA in IBM PS/2 systems in mind, most boards don't pass every test. However, a failure of one or two of the tests doesn't usually indicate serious compatibility problems for most software. ▒▒▒ Quality ▒▒▒ ▄▄▄▄▄▄▄▄▄▄▄▄▄▄▄ EGA/VGA Monitor The quality test permits images to be displayed in various modes from black and white text to VGA color graphics. These tests are used to determine what display modes a monitor/display adapter combination can support. Also, they are used for evaluating laptop screens to see how well they can map colors to gray scale. ▒▒▒ Set ▒▒▒ ▄▄▄▄▄▄▄▄▄▄ Machine ID Calling the Machine ID selection from the set menu permits you to change information about the unit under test. You can give the unit a new variant number and description to reflect a configuration change. Or you can append notes to the Machine Description field to record any interesting findings while running tests. ▒▒▒ Known Problems and Irregularities ▒▒▒ ▄▄▄▄▄▄▄ 10/4/91 We have found that certain high speed video board bus speed combinations have caused the benchmarks to display artifacts ("holes") in the screen display. This has also led to certain graphics tests "locking up" the system, forcing a cold boot. This has been seen in video adapters that run at zero wait states on I/O busses running at 10 MHz. The solution to this problem is to slow down the bus speed, usually through some type of programmable option in the system's CMOS. ▄▄▄▄▄▄▄ 10/4/91 Version 7.00 of the Microsoft mouse driver behaves erratically when running tests, with the cursor disappearing and then reappearing with the mouse buttons inoperative. We recommend using versions prior to or after version 7.00. (Versions 6.14 and 7.04 have been used successfully in the Labs.) ▄▄▄▄▄▄▄ 10/7/91 The IBM XT does not function properly with any tested version of the Microsoft Mouse driver. Apparently, it is not fast enough to service the mouse interrupt while maintaining the display. We recommend not using the mouse if you are running the benchmark on an XT class (8088 based) machine. ▄▄▄▄▄▄▄ 10/7/91 Certain key combinations will cause the system to lock up if executed at the wrong time. Both Alt-F4 and Ctrl- C, which are keyboard accelerators for exiting the program will lock the system up if there are any windows open. You will be forced to reboot the machine if these key combinations are used while a window is open. ▄▄▄▄▄▄▄ 10/8/91 At this time the memory required to run the full suite of benchmark tests is approximately 580,000 bytes. With less memory available you should be able to run all tests except the 128K NOP Loop and the Graphics BitBlts. Future revisions will address this limitation. ▄▄▄▄▄▄▄ 10/8/91...The results for the Screen to Screen BitBlt 16-bit test for the IBM XT included with the benchmark disk is inaccurate and does not reflect the video performance of the IBM EGA adapter installed in the XT. Due to the graphics intensive nature of this test, the XT cannot update the screen memory properly to allow the benchmark to measure its performance. If you have any specific comments or experience irregularities or problems running the benchmarks, you may send a description of them along with either hard copy or electronic copy of test results along with a system description and any other pertinent information to: Attn: Benchmark Department ZD Labs 320 B Lakeside Drive Foster City, CA 94404 ▒▒▒ Additional Information ▒▒▒ Additional information may be found accompanying reviews in Ziff publications. If you have any specific questions or suggestions, please send them along with your registration form or in a separate letter. Thank you for using the ZD Labs Benchmark Series, Release 1.1. Sincerely, Elizabeth Springer Director, ZD Labs NOTE: Please be sure to read the additional files on this disk WHATS.NEW and DATABASE.TXT. These files describe what is new in BENCH 1.1 as well as the functioning of the product's database.