Timings

We now present some timings. We will vary the parameters raster buffer size and number of threads to see how these affect the results. The total file size of bitmaps (TIFF compressed) written to disk is 287.062 MB. The compression and formatting is done by the sample application but the time required is included in the total time presented below.

Time vs. Concurrent Threads
350 mm x 350 mm at 8000 DPI


raster      number of      number        total elapsed time
buffer        bands      of threads
256 MB         6             1             98 secs
256 MB         6             2             74 secs
256 MB         6             3             61 secs
256 MB         6             4             48 secs
256 MB         6             1             98 secs  (sanity check to check for OS caching)

What can we conclude from the above data?

Increasing the number of concurrent threads improves performance. Since not all the processing time is multi-threaded and there is some disk IO time one does not get completely proportional reduction in time vs threads.


Time vs. RAM
350 mm x 350 mm at 8000 DPI

Since we have 8 GB of RAM (and about 2.5 GB is used by the system) we can increase the raster buffer size. Let's see what effect setting higher raster buffer size gives us.


raster      number of      number        total elapsed time
buffer        bands      of threads
256  MB         6             1             98 secs
512  MB         3             1             96 secs
512  MB         3             1             95 secs  (DBS set to 10,000,000)
768  MB         2             1             94 secs
768  MB         2             2             77 secs

Clearly the extra RAM is not helpful for this particular combination of input data and image size/resolution.


Faster Hardware?

raster      number of      number        total elapsed time
buffer        bands      of threads
256  MB         6             4             48 secs    Intel Core2 8400  2.66 GHZ
256  MB         6             4             34 secs    Intel Core i7 950 3.07 GHz

A faster workstation will result in better throughput.






300 MM Wafer Scale Mask Benchmark

1   |   2   |   3