- at least 1 GHz processor is recommended;
- at least 256 MB of RAM is recommended;
- 1024*768 is the minimum screen resolution required;
- JVM - Java Virtual Machine (SPS is written in JAVA, thus is platform independent).
To make a simulation, all you have to do is to: select the desired benchmarks and choose the values of the input parameters. Please note that, by default, the parameters values are set their optimum values.
Also, the Threshold parameter can be linked with the size of the global history register (h). This way, the threshold will be: [1.93*h] + 14 .
After the simulation of the selected traces is done, the results will be displayed.
To save the simulation results, all you have to do is to press the Save results button from the Results tab. You can choose what results to save by simply selecting them...
SPS allows you to simulate over multiple computers that are in a network. This way you can obtain the desired results much faster. Press the Delegate simulation button, select the simulation parameters, input the IP of the computer on which you want to make the simulation and send the request. When the simulation is done, you will see the results in the Results panel.
Note: The server mode must be enabled in order to work over a network. SPS uses a peer-to-peer network architecture.
You can view your simulation results by putting them on a chart. Creating a chart with SPS is easy: select the Charts tab, specify what do you want to see on the chart and press the Generate chart button. After all the required simulations are done (this may take a while), the chart will be displayed. You can save your chart (as a .pgn image) for further investigation, by pressing the Save chart button .
SPS allows you to work with SPEC 2000 benchmarks. In order to use the go to Tools and select Use SPEC 2000 benchmarks.
You can work with SPEC 2000 bechmarks the same way you did with the Stanford benchmarks. But, please note that simulation of SPEC 2000 benchmarks takes much more time (each of the 17 SPEC 2000 benchmarks has 4,000,000 dinamic instructions) .
Also, you can enhance the simple perceptron by using path information (go to: Tools->Enhance simple perceptron), or, by doing a static training of the neural network before each simulation.
Please note that static training is not availale for the enhanced simple perceptron (which is actually the Fast Path Based Neural Predictor written by Daniel Jimenez).
When doing a static training, it is recommended that you use a small global history (up to 8-10), especially for SPEC 2000 (otherwise you may run out of memory!!!).
By default, training statistics are saved in the results\statistics folder. You may choose to stop saving statistics: go to Options->Preferences.
Training statistics can also be filtered by variating the filter percent parameter. This way you can choose to use in the training process only the brances which are biased over filter percent.
Bored of simulations?
Go to Options->Look and feel and check out the look and feels SPS has to offer...