diff --git a/ubb-vga/regs4740.h b/ubb-vga/regs4740.h
index 808e26e..215f9b9 100644
--- a/ubb-vga/regs4740.h
+++ b/ubb-vga/regs4740.h
@@ -50,8 +50,12 @@
 #define	TSCR	_TCU(0x3c)		/* Timer STOP clear */
 #define	TESR	_TCU(0x14)		/* Timer counter enable set */
 #define	TECR	_TCU(0x18)		/* Timer counter enable clear */
+#define	TFR	_TCU(0x20)		/* Timer flag */
+#define	TFSR	_TCU(0x24)		/* Timer flag set */
+#define	TFCR	_TCU(0x28)		/* Timer flag clear */
 #define	TCSR(n)	_TCU(0x4c+0x10*(n))	/* Timer control */
 #define	TDFR(n)	_TCU(0x40+0x10*(n))	/* Timer data full */
+#define	TDHR(n)	_TCU(0x44+0x10*(n))	/* Timer data half */
 #define	TCNT(n)	_TCU(0x48+0x10*(n))	/* Timer counter */
 
 #define	MSC_STRPCL	_MSC(0x00)	/* Start/stop MMC/SD clock */
diff --git a/ubb-vga/ubb-vga.c b/ubb-vga/ubb-vga.c
index 652f415..e9d063d 100644
--- a/ubb-vga/ubb-vga.c
+++ b/ubb-vga/ubb-vga.c
@@ -233,9 +233,14 @@ static void cleanup(void)
 /* ----- Delay logic ------------------------------------------------------- */
 
 
-static void until(uint16_t cycles)
+static void delay(uint16_t cycles)
 {
-	while ((TCNT(TIMER) & 0xffff) < cycles);
+	static uint16_t next = 0;
+
+	next += cycles;
+	TDHR(TIMER) = next;
+	TFCR = 1 << (TIMER+16);
+	while (!(TFR & (1 << (TIMER+16))));
 }
 
 
@@ -292,7 +297,7 @@ static void line(unsigned long line)
 	DTA(DMA) = REG_PADDR(MSC_TXFIFO);	/* MUST set this each time */
 	DTC(DMA) = (mode->xres+63) >> 6;
 
-	until(mode->hback_cycles);
+	delay(mode->hback_cycles);
 
 	/* HSYNC */
 
@@ -313,7 +318,7 @@ static void line(unsigned long line)
 	    (1 << 31) |         /* no descriptor */
 	    1;           
 
-	until(mode->hback_cycles+mode->hsync_cycles);
+	delay(mode->hsync_cycles);
 
 //	MSC_TXFIFO = 0xffffffff;
 
@@ -323,7 +328,7 @@ static void line(unsigned long line)
 	PDDATS = HSYNC;
 	PDFUNC = CMD;
 
-	until(mode->line_cycles);
+	delay(mode->line_cycles-mode->hback_cycles-mode->hsync_cycles);
 	MSC_TXFIFO = 0;
 	if (MSC_STAT & 3)
 		bad++;
@@ -333,11 +338,10 @@ static void line(unsigned long line)
 static void hdelay(int cycles)
 {
 	while (cycles--) {
-		TCNT(TIMER) = 0;
 		PDDATC = HSYNC;
-		until(mode->hsync_cycles);
+		delay(mode->hsync_cycles);
 		PDDATS = HSYNC;
-		until(mode->line_cycles);
+		delay(mode->line_cycles-mode->hsync_cycles);
 	}
 }
 
@@ -359,22 +363,15 @@ static void frame(const unsigned long *f)
 	 * The horizontal back porch of the previous line is handled inside
 	 * "line", so we have to wait for less than a full line here.
 	 */
-	TCNT(TIMER) = 0;
 	PDDATC = HSYNC;
-	until(mode->hsync_cycles);
+	delay(mode->hsync_cycles);
 	PDDATS = HSYNC;
-	until(mode->line_cycles-mode->hback_cycles);
+	delay(mode->line_cycles-mode->hback_cycles-mode->hsync_cycles);
 
-	/*
-	 * Note: resetting the timer just before calling "line" isn't enough.
-	 * We have t reset it before the loop and right after returning from
-	 * "line".
-	 */
-	TCNT(TIMER) = 0;
-	for (p = f; p != f+mode->yres; p++) {
+	for (p = f; p != f+mode->yres; p++)
 		line(*p);
-		TCNT(TIMER) = 0;
-	}
+
+	delay(mode->hback_cycles);
 
 	/* Back porch */
 	hdelay(mode->vback_lines);
@@ -401,6 +398,7 @@ static void session(void (*gen)(void **fb, int xres, int yres), int frames)
 
 	setup_noirq();
 
+	TCNT(TIMER) = 0;
 	for (i = 0; !frames || i != frames; i++) {
 		frame(f_phys);
 		if (DTC(DMA)) {