The internal latency, throughput, and CPU speed of the Paragon limit the overall rendering speed. (Using naive function calls, we measured the Paragon internal message latency at around 150 microseconds and the best throughput at around 13MB/sec. Using newer versions of the operating system or sunmos [MMRW93] should give better performance.) For instance, Figures 10 and 11 contain performance information on compositing latency and throughput, assuming zero time rendering. The compositing latency is the minimum time it takes for an image to reach the collector from the moment the host asked for the image. The throughput reflects the maximum frame rate achievable. It is interesting to note (as shown before in [SK94]) that the compositing throughput is independent of the number of processors in the pipeline (PVR pipelines inside each cluster), and depends only on the image size, while the latency grows linearly with the number of processors and image sizes.
Figure 12: Image pipeline. Latencies indicated.