forum.semcad.com

(1) large field sensors save slowly. Make the sensors as small as possible. Consider the possibility of 2-D field sensors.

(2) the absolute saving time is in the same range for both sw and hw (I have in mind that for one of my benchmarks sw-saving-time*1.4 = hw-saving-time). All the data have to be written on the slow hard drive and in hw we have to get the data from the card-memory back to the native PC memory in addition. The factor 1.4 is simulation dependent.

(3) This immediately implies that the ratio between FDTD update time (which has a huge speed up) and saving time gets worse when using the hw acceleration. Just a numeric example (20MCells/s sw vs. 400MCells/s hw).

- sw sim: 20h (update) and 1h (saving) -> 21h simulation time
- hw sim: 1h (update) and 1.4h (saving) -> 2.4h simulation time

Conclusions:
(3a) the hw sim is still much much faster than sw (factor 8)
(3b) saving time is 4.8% of the simulation time in sw, but 58% (!) in hw

(4) the points (3a,3b) explain, that it is sometimes a 'mental' problem of the human beings. While the FDTD update is fast, the saving does not speed up.

(5) disabling the compression, in SEMCAD X preferences, speeds up the saving process but the resulting file size is bigger.

Hi I just installed two solid-state disk in striped RAID configuration in our CIB1500 and assigned it for SEMCAD x calculations (and post-processing) only. The saving is now about 10x faster compared to the normal sata2 drive.

My config: compression off, output files approx 1.6GB

Best,
Jurriaan