Benching with OSmark

Discuss software and how to tweak more performance out of your system.
Post Reply
DonPedro
K6'er Elite
Posts: 578
Joined: Wed Jul 27, 2005 2:11 pm

Benching with OSmark

Post by DonPedro »

I just would like to invite everybody to not only "rely" on synthetic and limited benchmarks like everest, hot cpu tester, sisoft sandra and alike. to measure real life performance it needs a tool that is sound in variety of application benching, I am talking about the open source project OSMARK (OpenSourceMark).

there are 2 versions downloadable:

version 1b7, which I used so far in the past and will continue to so in the future to keep my data straight and consistent. download here http://www.vanshardware.com/files/OSMark.zip (size 24mb)

and

version 1b8, which gives higher numbers compared to 1b7. I guess they optimized the code or used sharper compiler settings. download here https://sourceforge.net/projects/opensourcemark/ (size 57mb)

version 1b8 brings along a documentation very interesting to read and for better understanding that benching is an art (OSMarkDocumentation.html). for everybody's convenience I have copied and pasted some parts from this documentation below for those who want to know firsthand what they get before they start the mb-hefty download. version 1b8 features some more benches than 1b7. I have copied only those bench-description both versions offer.

Understanding Computing Performance

Personal computers are used for a wide variety of tasks. Since no one uses his computer exactly like anyone else, producing a single number to represent overall computing performance is not easy.

A quantity that can be represented by a single number is referred to as “scalar.” Overall computing performance is not scalar, but can be represented by a six-dimensional vector. A “vector” is a quantity that has magnitude and direction, or requires more than one number to describe it.

Here is a breakdown across the six major “performance vectors.”

o 3D Gaming: Modern 3D rendered games create highly interactive, visually rich worlds and impose the most computationally intensive burdens commonly carried out on personal computers. Graphics card choice, screen resolution, color depth, driver implementation, and rendering features have the biggest impacts on gaming performance. However, floating point throughput (FPU, SSE, SSE2, SSE3, etc,), memory amount, memory latency and memory bandwidth are also very important. For games that are not 3D intensive, responsiveness qualities (see below) are typically most vital. OpenSourceMark Official Run tests stressing this vector are: AlphaBlend, AlphaDots, BandwidthBP64, Fern Fractal, Lorenz, Mandelbrot, MemLatency, nBodyOpenGL, nBody, SSE3, PlotTrig, PlotTrig2.

o Artificial Intelligence: Gaming logic, financial modeling, expert systems, robotics path finding and some types of image recognition are all examples of artificial intelligence (AI), a unique performance category. Certain spam filters and anti-virus packages also use AI. OpenSourceMark Official Run tests stressing this vector are: Maze and MemLatency..

o Data Set Transforms: Media encoding and decoding, file compression and decompression, image processing and audio mixing are examples of data set transformations. Many people now watch movies on their computers, and growing numbers of people use their PCs to produce motion pictures. With the digital photography revolution, the personal computer has become an invaluable tool for viewing and manipulating family photographs. File compression/decompression is a common task that often closely tracks encoding performance. Memory bandwidth is very important in this domain, but the CPU also needs to be fast enough to handle the incoming data. Superscalar and/or multicore microprocessors are strong in this domain. OpenSourceMark Official Run tests stressing this vector are: 7zip, BandwidthBP64, Identical Threads, Image Resize, Image Rotate, JPG Decode, Lame Encoder, PNGOut, SSE3 Thread, UPX , ZIP Compress.

o Math/Science/Engineering: For a small subset of computer users, math, scientific and engineering applications constitute an important slice of the computing pie. While some applications in this vector overlap data set transformations, these programs tend to greatly emphasize floating point performance (including transcendentals). In floating point intensive streaming cases, the memory subsystem needs to be fast enough to keep the FPU/SSE2 units fed. Numerical integration carried out in new physics acceleration add-in cards falls within this vector. OpenSourceMark Official Run tests stressing this vector are: BandwidthBP64, Dhrystone, Fern Fractal, Fibonacci, GridBlastFP, Lorenz, Mandelbrot, Montgomery Multiplier, nBodyOpenGL, nBodyTest, Orthogonal Thread, Pi, SSE3 Thread, PlotTrig, Plot Trig2, Whetstone.

o Responsiveness: The most overlooked aspect of computing performance is probably the single most important performance area for general computing. Responsiveness is a measurement of a system's ability to keep up with user input and is characterized by extremely branchy code that typically requires intensive interaction between the CPU, display adapter, memory, and hard drive. A responsive computer is snappy and quick while a computer with poor responsiveness feels sluggish and lethargic. As user interfaces grow more complicated, responsiveness becomes more important. For example, Windows Vista imposes prodigious demands even to render the simplest application windows; many systems will not be able to provide an enjoyable computing experience unless a number of Vista's new interface features are turned off. A system's responsiveness is largely a function of the CPU's integer unit, the CPU's pipeline depth and cache line size, memory latency, memory amount, hard drive speed and interface, the video core's 2D unit, performance state transition latencies and thresholds, screen resolution, refresh rate and color depth and the CPU's interface to the video core. Vista will also make heavy demands on 3D rendering and cryptographic throughput (see Security below). Surprisingly, small and simple power-efficient CPUs can be more responsive than massive, deeply pipelined microprocessors like the Pentium 4. Multicore chips will help responsiveness a lot, especially when antivirus software is running in the background. Thin-clients fall essentially 100% inside this performance vector. Office application suites and web browsing are strongly driven by this vector. AMD Athlon 64 X2 processors are currently the acme of performance for this vector. OpenSourceMark Official Run tests stressing responsiveness are: Filled Circle , DrawEllipses, Fern Fractal, File Copy, GridBlastFP, GridBlast, Draw Lines, MemLatency, Plot Lines, Random Dots, Draw Rectangles, Rich Edit , Web Page.

o Security: Rapidly growing in importance, security involves everything from protecting online transactions to encrypting file systems and data networks on the fly. Not only is security necessary to lock away information from prying eyes, but movie and music providers are insisting on cryptographically secure data streams for next generation media. Facilitating this, Windows Vista steps up cryptography demands by vast leaps and bounds by using the CPU to perform AES encryption on media streams on the fly. While most security functions fall within the data set transform domain, implementation concerns place security within its own unique performance vector. OpenSourceMark Official Run tests stressing this vector are: AES, Montgomery Multiplier, SHA-1, and SHA-256.

In addition to the tools necessary for characterizing a system by the performance vectors listed above, OpenSourceMark also provides benchmarks for measuring notebook performance and battery life.



7zip
Authors: Van Smith, Kolumba.
Description: 7zip is a popular archiving tool that achieves extremely high levels of file compression. The OpenSourceMark 7zip test compresses the entire OSMark directory to a temporary archive located in the root directory of the target drive.
Threaded: Yes
Performance vector: Data Set Transforms
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 59%

BandwidthBP64
Authors: Van Smith
Description: BandwidthBP64 uses the block prefetch 64 assignment from BandwidthBurn. The block prefetch assignment first loads 128 four-byte integer values spaced at every 64 bytes in the integer array before stepping through the normal assignment statements. BandwidthBP64 includes both reads and writes.
Threaded: Yes
Performance vector: 3d Gaming, Data Set Transforms, Math/Science/Engineering
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 62%
Scaling from UP to DP: 47% (will scale even better on NUMA systems)

CalculatePi
Authors: Ray Lischner, Van Smith
Description: Calculates pi using Ray Lischner's Delphi algorithm. Pi is calculated to 1,000, 2,000, 3,000, 4,000 and 5,000 digits. See http://www.tempest-sw.com for a look at some of Ray Lischner's famous books.
Threaded: No
Performance vector: Math/Science/Engineering
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 100%
Scaling from UP to DP: 0%

Circles
Authors: Van Smith
Description: Circles paints randomly sized, randomly position filled circles over the canvas of the output form.
Threaded: No
Performance vector: Responsiveness
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 29%
Scaling from UP to DP: 0%

DhrystoneThreads
Authors: Van Smith, Bayu Prasetio
Description: DhrystoneThreads automatically spawns Dhrystone threads to match the number of logical processors in the system. The Dhrystone test is a port of Reinhold P. Weicker's original code.
Threaded: Yes
Performance vector: Math/Science/Engineering
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 106%
Scaling from UP to DP: 98%

Ellipses
Authors: Van Smith
Description: This is a basic GDI test that draws random ellipses over the output form.
Threaded: No
Performance vector: Responsiveness
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 29%
Scaling from UP to DP: 0%

EncryptDecrypt
Authors: Van Smith, Vijay Hariharan, Tom Crispin
Description: Performs AES encryption / decryption based on Dr. Brian Gladman's routines. VIA’s processors use the Padlock hardware acceleration unit. To avoid inflating overall scores on VIA-based platforms, the EncryptDecrypt test is normalized against a 1.333 GHz VIA C3.
Threaded: Yes
Performance vector: Security
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 99%
Scaling from UP to DP: 97%

FernFractal
Authors: Van Smith
Description: This test draws two fern fractals on the canvas of the output form. The algorithm is floating point intensive and makes heavy use of trigonometric transcendental functions. Fractals are used in many areas of computing from landscape generation in games to image compression to liquid dynamics to even simulations of living objects like trees and ferns. See the Wikipedia article for more on fractals.
Threaded: No
Performance vectors: Math/Science/Engineering, 3d Gaming, Responsiveness
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 86%
Scaling from UP to DP: 0%

FibThreads
Authors: Van Smith
Description: The Fibonacci threads test spawns a thread for each logical processor in the system. Each thread calculates the 40th order Fibonacci number.
Threaded: Yes
Performance vectors: Math/Science/Engineering
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 100%
Scaling from UP to DP: 0%

FileCopy
Authors: Van Smith
Description: Using file streaming calls, this test measures file copy performance with a 100MB source file copied to five targets.
Threaded: No
Performance vectors: Responsiveness
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 0%
Scaling from UP to DP: 0%

GridBlast
Authors: Van Smith
Description: This simple spreadsheet simulation populates a grid with integers and performs iterative calculations through the entire data set.
Threaded: No
Performance vectors: Responsiveness
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 83%
Scaling from UP to DP: 0%

GridBlastFP
Authors: Van Smith
Description: This simple spreadsheet simulation populates a grid with double-precision floating values and performs iterative calculations through the entire data set.
Threaded: No
Performance vectors: Responsiveness, Math/Science/Engineering
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 77%
Scaling from UP to DP: 0%

IThreads
Authors: Van Smith
Description: The Identical Threads test executes two identical quick sort threads. Each thread sorts 5,000,000 elements.
Threaded: Yes
Performance vectors: Data Set Transforms
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 82%
Scaling from UP to DP: 85%

ImageResize
Authors: Van Smith
Description: The Image Resize test gradually resizes a decoded JPG from full window to very small and back again to full window.
Threaded: No
Performance vectors: Data Set Transforms
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 52%
Scaling from UP to DP: 0%

ImageRotate
Authors: Van Smith
Description: The Image Rotate takes a decoded JPG image and rotates it numerous times at 90 degree increments.
Threaded: No
Performance vectors: Data Set Transforms
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 25%
Scaling from UP to DP: 0%

JpgDecode
Authors: Van Smith
Description: “JPG” is the standard file extension for the most popular compressed digital image format used on computers today. Decoding “jpg” files is processing intensive and as image sizes soar with higher and higher digicam megapixel counts, the decoding time can take several seconds on slow systems. The JPG decoding test decodes numerous native JPG images taken from a variety of digital still cameras.
Threaded: No
Performance vectors: Data Set Transforms
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 98%
Scaling from UP to DP: 0%

LameMp3Encode
Authors: Van Smith
Description: MP3 is the most popular audio compression format in use today. Lame is the most commonly used Open Source MP3 encoder. The LameMp3Encoder test compresses a standard sized song in WAV format into an MP3 file.
Threaded: Yes
Performance vectors: Data Set Transforms
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 97%
Scaling from UP to DP: 100%

Lines
Authors: Van Smith
Description: Lines is a simple GDI test that draws random lines across the canvas of the output form. Line drawing is important for GUI rendering.
Threaded: No
Performance vectors: Responsiveness
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 50%
Scaling from UP to DP: 0%

LorenzAttractor
Authors: Van Smith
Description: LorenzAttractor is an OpenGL 3d animated model of the Lorenz Attractor, a famous example of chaos theory. This test utilizes GLScene.
Threaded: No
Performance vectors: 3d Gaming, Math/Science/Engineering
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 54%
Scaling from UP to DP: 0%

Mandelbrot
Authors: Van Smith
Description: Mandelbrot is a threaded rendering of a portion of the well known Mandelbrot set. The test’s complex number arithmetic intensively uses double-precision FPU calculations. Fractals are used in many areas of computing from landscape generation in games to image compression to liquid dynamics to even simulations of living objects like trees and ferns. See the Wikipedia article for more on fractals.
Threaded: Yes
Performance vectors: 3d Gaming, Math/Science/Engineering
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 96%
Scaling from UP to DP: 59%

Mazes
Authors: Van Smith, Chris Yeager
Description: The Mazes test generates complex two-dimensional mazes and solves each one. Failed taken paths are red while the current path is green.
Threaded: No
Performance vectors: Artificial Intelligence
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 85%
Scaling from UP to DP: 0%

MemLatency
Authors: Van Smith
Description: This threaded memory latency test measures main memory latency by making random 32-bit copies over two 16MB blocks of memory. The number of threads spawned automatically corresponds to the number of logical processors in the system.
Threaded: Yes
Performance vectors: Artificial Intelligence, Responsiveness, 3d Gaming
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 40%
Scaling from UP to DP: 2.4% (will be much higher for NUMA systems)

nBody
Authors: Van Smith
Description: The nBody test is a 2d simulation of 4 point source objects interacting via gravity in a plane. This test is optimized for SSE2 and SSE3 and is one of the most floating point intensive tests in OpenSourceMark. The numerical integration method is Euler’s Last Point Approximation.
Threaded: No
Performance vectors: 3d Gaming, Math/Science/Engineering
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 99%
Scaling from UP to DP: 0%

nBodyOpenGL
Authors: Van Smith
Description: The nBodyOpenGL test is the exact 3d model of the 4-body simulation introduced in the nBody test immediately above. This test is optimized for SSE2 and SSE3 and is one of the most floating point intensive tests in OpenSourceMark. The numerical integration method is Euler’s Last Point Approximation.
Threaded: No
Performance vectors: 3d Gaming, Math/Science/Engineering
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 64%
Scaling from UP to DP: 0%

OThreads
Authors: Van Smith
Description: The Orthogonal Thread test was developed as an example of two threads that share neither code nor data, thus minimizing cache line contention. One thread calculates the 40th order Fibonacci number, while the second thread calculates 4,000 digits of pi using Ray Lischner’s Delphi algorithm.
Threaded: Yes
Performance vectors: Math/Science/Engineering
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 100%
Scaling from UP to DP: 81.5%

PiThreads
Authors: Van Smith, Ray Lischner
Description: This test is similar to the CalculatePi test above, but spawns threads for each processor with each thread calculating pi to 10,000 digits using Ray Lischner’s Delphi algorithm. See http://www.tempest-sw.com for a look at some of Ray Lischner's famous books.
Threaded: Yes
Performance vectors: Math/Science/Engineering
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 100%
Scaling from UP to DP: 100%

PlotLines
Authors: Van Smith
Description: Point plotting is very important for GUI rendering. The PlotLines test plots lines one pixel at a time and is remarkably CPU intensive.
Threaded: No
Performance vectors: Responsiveness
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 89%
Scaling from UP to DP: 0%

PlotTrig
Authors: Van Smith
Description: The PlotTrig test plots sine, cosine and tangent curves. Every point calculated is plotted.
Threaded: No
Performance vectors: Math/Science/Engineering, 3dGaming
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 90%
Scaling from UP to DP: 0%

PlotTrig2
Authors: Van Smith
Description: The PlotTrig test plots sine, cosine and tangent curves. Only one out of ever 100 points calculated is plotted.
Threaded: No
Performance vectors: Math/Science/Engineering, 3dGaming
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 99%
Scaling from UP to DP: 0%

PNGOut
Authors: Van Smith, Kalumba
Description: PNG is a new open standard lossless image format that is replacing GIFs. PNGOut is a program that will convert image files to the PNG format. The PngOut test launches an instance of PNGOut for each logical processor in the system. Each thread converts the same JPG file into PNG format.
Threaded: Yes
Performance vectors: Data Set Transforms
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 60%
Scaling from UP to DP: 79%

RandomDots
Authors: Van Smith
Description: Point plotting is very important for GUI rendering. The RandomDots test plots pixels randomly over the output form’s canvas. This simple test is a very good indicator of overall 2d performance.
Threaded: No
Performance vectors: Responsiveness
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 86%
Scaling from UP to DP: 0%

Rectangles
Authors: Van Smith
Description: Rectangles is another GDI primitive test. Rectangle drawing performance is very important for GUI rendering.
Threaded: No
Performance vectors: Responsiveness
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 15%
Scaling from UP to DP: 0%

RichEdit
Authors: Van Smith
Description: Many Windows applications are component based. One of the most commonly used components is Microsoft’s RichEdit which forms the basis of WordPad and even Word. The RichEdit test is a basic test of word processing text rendering performance.
Threaded: No
Performance vectors: Responsiveness
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 14%
Scaling from UP to DP: 0%

UPX
Authors: Van Smith, Kalumba
Description: UPX is a utility to compress executables into files that self extract and run when launched. The UPX test spawn a thread for each logical processor in the system. Each thread compresses "CosbiOpenSourceMark.exe".
Threaded: Yes
Performance vectors: Data Set Transforms
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 81%
Scaling from UP to DP: 89%

WebPageLoad
Authors: Van Smith
Description: WebPageLoad instantiates an instance of Internet Explorer inside the output form and displays a variety of web pages.
Threaded: Yes
Performance vectors: Responsiveness
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 81%
Scaling from UP to DP: 42%

WhetstoneThreads
Authors: Van Smith, Bayu Prasetio
Description: WhetstoneThreads launches a Whetstone thread for each logical processor in the system. Whetstone is a venerable and ubiquitous floating point benchmark.
Threaded: Yes
Performance vectors: Math/Science/Engineering
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 100%
Scaling from UP to DP: 100%

ZipCompress
Authors: Van Smith
Description: “Zip” is the most popular file compression format used on PCs. The ZipCompress test compresses the entire “osmfiles” subdirectory and uses the free ZipForge component that can be obtained at http://www.componentace.com.
Threaded: No
Performance vectors: Data Set Transforms
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 81%
Scaling from UP to DP: 0%

OSMark
Authors: N/A
Description: OSMark is the overall score from an OpenSourceMark Official Run. The OSMark score is the arithmetic average for all of the Official Run tests.
Threaded: Yes
Performance vectors: All
Scaling rate with CPU clock speed (100% = perfect; higher values possible with faster FSB): 67%
DonPedro
K6'er Elite
Posts: 578
Joined: Wed Jul 27, 2005 2:11 pm

Post by DonPedro »

since I recently began benching on boards with sis-chipset or via-chipset after I only had used this bench-program on ali-5 boards I would like to share some experiences.

first I thought that some "problems" I encountered while running it on ali-5 boards were due to this chipset or to the respective hole system-setup (cards plugged in, driver version used, os) but now I know that these problems also occur on other platforms.

sometimes the program would immediately abort after starting it, complaining about some "mci"-device related problem. I was able to solve this by dis-/enable some hw via the device manager, start osmark, went back into device manager to reverse what I had done before and then rän the bench.

also I encountered on all platforms (ali, sis, via) that sometimes the "official run" was not able to complete because one of its sub-benches finished with some error message. in most cases it was the I(dentical)-thread test. after it completed, successfully giving a score, a message comes up saying something about a "parameter" error. and the whole bench-suite comes to a hold .... this is very unconvenient because normally I leave the machine alone while the benches run. when you come back 2 hours later and realize that the bench is broken somewhere in the middle it kind of sucks. I "solved" this by not choosing the official run, but instead set the run counter to 3, unticked the IThread test (and also unticked the 3 tests a little bit "offset" in the bottom left area of the test-box; they are not part of the official test). then I started the bench via the run-button. when the bench-suite finished, I unticked all (there is a button for this), ticked just the IThread test and let it run 3 times. if errors occurred here, it does not mess up the whole thing.
Jim
K6'er Elite
Posts: 1745
Joined: Wed Jan 21, 2004 7:10 pm
Location: Toronto

Post by Jim »

I have come to the conclusion, that Peter's OSMark is probably the best benchmark program there is to determine overall as opposed to specialized performance in some particular area of one's machine. That said, however, I still think OSMark has to be approached cautiously. It stresses the machine being benched, much more highly than other benchmarks do; and for that reason can cause freezes and other problems which may result in OS corruption if the machine is overtweaked.
Superpuppy 3
K6-3+ 450 ACZ (6x100)
DFI K6BV3+/66 Rev B2 (2 Meg) w/ 2x28mm Chipset Fans
2x256 Meg PC 133 Hynix SDRAM
1x 20G Maxtor (7200)
2x 80G Maxtor (7200) Ducted w/ 2x486 Fans Mount
52/24/52/16 LG CDR/RW/DVD
8/4/3/12/24/16/32 LG Super Multi
ATI 9000 aiw Radeon AGP
SB Audigy 1 MP3 Sound
CMD 649 IDE Controller
NEC USB 2 Card
DonPedro
K6'er Elite
Posts: 578
Joined: Wed Jul 27, 2005 2:11 pm

Post by DonPedro »

jim,

the flowers go to the osmark-team.

did you bench around lately, if so, let us know your experience in more detail?

ot and btw, did you read what osmark's author published in november on his site ( http://www.vanshardware.com )? could be of interest to you ....
Jim
K6'er Elite
Posts: 1745
Joined: Wed Jan 21, 2004 7:10 pm
Location: Toronto

Post by Jim »

Facinating! Some people are awake. Have not done any benching lately. Am bogged down getting Glorias ancient DOS games working on her new machine. The computer shop did not load DOS 6.2 before they loaded Windopes, and consequently several of her games are giving "Insufficient memory messages". Got to wipe the machine, repartition it, install DOS 6.2, reinstall Windopes, run Memmaker, to get DOS to load into high memory, then do her whole setup from there. Loads of fun!!
Superpuppy 3
K6-3+ 450 ACZ (6x100)
DFI K6BV3+/66 Rev B2 (2 Meg) w/ 2x28mm Chipset Fans
2x256 Meg PC 133 Hynix SDRAM
1x 20G Maxtor (7200)
2x 80G Maxtor (7200) Ducted w/ 2x486 Fans Mount
52/24/52/16 LG CDR/RW/DVD
8/4/3/12/24/16/32 LG Super Multi
ATI 9000 aiw Radeon AGP
SB Audigy 1 MP3 Sound
CMD 649 IDE Controller
NEC USB 2 Card
Post Reply