``` 230 FOR J= 0 TO ITER 240 LOAD "ANIM"+ STR$ (J)+".SL2" LAYER 250 NEXT J 260 NEXT %I 360 LAYER 2,0: LAYER 0 ``` The program generates ellipses that rotate counter to one another and after drawing each frame, saves the entire screen on the SD card. Once it's done generating (when **CURITER** reaches **ITER**), it uses **LOAD … LAYER** (which we will look at in depth in *Chapter 19*) to load and display the *Layer 2* screens the previous part generated. Unlike the previous example using *Layer 0* which only moved **2K** at a time, this loads and displays **48K** at a time. Compared to the previous example using the *RAMdisk*, this appears much smoother and the reason is simple; there are many more frames generated by the program than what the previous one did. The question is can it be made smoother and if at all possible, faster? ### Memory Based Frame Playback It's time to delegate frame playback to RAM. Replace line 90 with this, longer, version: ``` 90 IF CURITER <=ITER THEN SAVE "ANIM"+STR$ (CURITER)+",SL2" LAYER: BANK 9 COPY TO 111-(CURITER*3): BANK 10 COPY TO 110-(CURITER*3): BANK 11 COPY TO 109-(CURITER*3): CURITER +=1 ``` and then add the following lines at the end: ``` 270 PRINT AT 22,0;"Done Loading from SD. Press any key to load from memory" 280 PAUSE 0 290 FOR %I=0 TO 5 300 FOR %J=0 TO % INT {ITER} 310 BANK %111-(3*J) COPY TO %9 320 BANK %110-(3*J) COPY TO %10 330 BANK %109-(3*J) COPY TO %11 340 NEXT %J 350 NEXT %I 360 LAYER 2,0: LAYER 0 ``` Run the program again and now compare the playback using the SD card, with the playback of all the screens using the memory. You can see that the playback is even smoother AND faster than the SD card and the reason is simple and that is because memory is a much faster medium than your SD card. Now there are several things of note here. First of all, this is not very efficient code, memory wise; *Layer 2* uses 3 banks of 16K each making an entire screen **48K** long. For the 20 iterations we made, that's **20 * 3 * 16K = 960K** making this program unlikely to work on a non-expanded KS1 ZX Spectrum Next[^p127-1]. Secondly, not the entire screen is moving. Only a small window does and that makes saving the remainder of each screen wasteful in mem- [^p127-1]: If you modify variable ITER however to a value around 10 it will work since we already know that banks 0 to 12 are being used by the system and 10*3*16 gives us a figure of 480K which is a memory size available on an unexpanded Next.