diff --git a/doom-fire.asm b/doom-fire.asm
index 8e404ee..71f7849 100644
--- a/doom-fire.asm
+++ b/doom-fire.asm
@@ -146,7 +146,7 @@ main_loop:
 	swi OS_Byte
 
 	; Wait for vsync if double buffering
-	.if Screen_Banks <= 2
+	.if Screen_Banks == 2
 	mov r0, #OSByte_Vsync
 	swi OS_Byte
 	.endif
@@ -472,15 +472,15 @@ plot_horizontal_line:
 .endm
 
 .macro DO_RANDOM_BIT
-	; update colour with some randomness
-	and r7, r8, #1				; rnd & 1
-	subs r4, r4, r7				; colour -= rnd & 1
-
 	; randomise destination a bit
-	and r7, r8, #3
+	and r7, r7, #3
 	add r0, r2, r7
 	subs r0, r0, #1				; dest_x += (rnd & 3)-1
 	movlt r0, #0				; or MOD ScreenWidth
+
+	; update colour with some randomness
+	and r7, r7, #1				; rnd & 1
+	subs r4, r4, r7				; colour -= rnd & 1
 .endm
 
 .macro PLOT_PIXEL_BIT
@@ -515,11 +515,17 @@ do_fire:
 	ldr r8, rnd_seed			; seed
 	mov r9, #1					; bit
 
-	mov r3, #Screen_Height - 64
+	mov r3, #Screen_Height - 48
 
 	; R10 = ptr to start of source line
 	add r10, r12, r3, lsl #7	; r10 = screen_addr + y * 128
 	add r10, r10, r3, lsl #5	; r10 += y * 32 = y * 160
+
+	; R3 = ptr to end of screen
+	mov r2, #Screen_Height - 1
+	add r3, r12, r2, lsl #7	; r10 = screen_addr + y * 128
+	add r3, r3, r2, lsl #5	; r10 += y * 32 = y * 160
+
 	; R12 = ptr to start of dest line
 	sub r12, r10, #Screen_Stride
 
@@ -527,43 +533,104 @@ do_fire:
 	mov r2, #0
 
 	.2:
-	; source is contiguous
-	; could read a source word at a time
-	; unroll 8x
-
 	; spread fire
+	RND
 
-	; read source byte
-	ldrb r5, [r10], #1			; load screen byte
+	; source is contiguous
+	; read source word = 8x pixels
+	ldr r5, [r10], #4
 
-	; Left pixel
+	; Byte 0 Left pixel
 	ands r4, r5, #0x0F
 	moveq r0, r2
 	beq .3
-	RND
 	DO_RANDOM_BIT
 	.3:
 	PLOT_PIXEL_BIT
 	add r2, r2, #1
 
-	; Right pixel
-	movs r4, r5, lsr #4			; mask out left hand pixel
+	; Byte 0 Right pixel
+	mov r4, r5, lsr #4
+	ands r4, r4, #0x0F
 	moveq r0, r2
 	beq .4
-	RND
+	mov r7, r8, lsr #4
 	DO_RANDOM_BIT
 	.4:
 	PLOT_PIXEL_BIT
 	add r2, r2, #1
 
+	; Byte 1 Left pixel
+	mov r4, r5, lsr #8
+	ands r4, r4, #0x0F
+	moveq r0, r2
+	beq .5
+	mov r7, r8, lsr #8
+	DO_RANDOM_BIT
+	.5:
+	PLOT_PIXEL_BIT
+	add r2, r2, #1
+
+	; Byte 1 Right pixel
+	mov r4, r5, lsr #12
+	ands r4, r4, #0x0F
+	moveq r0, r2
+	beq .6
+	mov r7, r8, lsr #12
+	DO_RANDOM_BIT
+	.6:
+	PLOT_PIXEL_BIT
+	add r2, r2, #1
+
+	; Byte 2 Left pixel
+	mov r4, r5, lsr #16
+	ands r4, r4, #0x0F
+	moveq r0, r2
+	beq .7
+	mov r7, r8, lsr #16
+	DO_RANDOM_BIT
+	.7:
+	PLOT_PIXEL_BIT
+	add r2, r2, #1
+
+	; Byte 2 Right pixel
+	mov r4, r5, lsr #20
+	ands r4, r4, #0x0F
+	moveq r0, r2
+	beq .8
+	mov r7, r8, lsr #20
+	DO_RANDOM_BIT
+	.8:
+	PLOT_PIXEL_BIT
+	add r2, r2, #1
+
+	; Byte 3 Left pixel
+	mov r4, r5, lsr #24
+	ands r4, r4, #0x0F
+	moveq r0, r2
+	beq .9
+	mov r7, r8, lsr #24
+	DO_RANDOM_BIT
+	.9:
+	PLOT_PIXEL_BIT
+	add r2, r2, #1
+
+	; Byte 2 Right pixel
+	movs r4, r5, lsr #28
+	moveq r0, r2
+	beq .10
+	mov r7, r8, lsr #28
+	DO_RANDOM_BIT
+	.10:
+	PLOT_PIXEL_BIT
+	add r2, r2, #1
+
 	cmp r2, #Screen_Width
 	blt .2
 
 	; Next line
 	add r12, r12, #Screen_Stride
-
-	add r3, r3, #1			; could omit this and cmp end of screen buffer instead
-	cmp r3, #Screen_Height	
+	cmp r12, r3
 	blt .1
 
 	str r8, rnd_seed
@@ -571,7 +638,7 @@ do_fire:
 	ldr pc, [sp], #4
 
 rnd_seed:
-	.long 0x12345678
+	.long 0x87654321
 
 rnd:
 ; enter with seed in R0 (32 bits), R1 (1 bit in least significant bit)