diff --git a/CHANGES b/CHANGES
index 8ff0d9afb..1bce17f89 100644
--- a/CHANGES
+++ b/CHANGES
@@ -19,6 +19,7 @@ Bugfixes:
  - GBA Video: Don't mask out high bits of BLDY (fixes mgba.io/i/899)
  - GBA Video: Force align 256-color tiles
  - GBA DMA: ROM reads are forced to increment
+ - GB Video: Fix loading states while in mode 3
 Misc:
  - GBA Timer: Use global cycles for timers
  - GBA: Extend oddly-sized ROMs to full address space (fixes mgba.io/i/722)
diff --git a/include/mgba/internal/gb/video.h b/include/mgba/internal/gb/video.h
index a19fa928e..90cfd7bbb 100644
--- a/include/mgba/internal/gb/video.h
+++ b/include/mgba/internal/gb/video.h
@@ -159,7 +159,7 @@ void GBVideoInit(struct GBVideo* video);
 void GBVideoReset(struct GBVideo* video);
 void GBVideoDeinit(struct GBVideo* video);
 void GBVideoAssociateRenderer(struct GBVideo* video, struct GBVideoRenderer* renderer);
-void GBVideoProcessDots(struct GBVideo* video);
+void GBVideoProcessDots(struct GBVideo* video, uint32_t cyclesLate);
 
 void GBVideoWriteLCDC(struct GBVideo* video, GBRegisterLCDC value);
 void GBVideoWriteSTAT(struct GBVideo* video, GBRegisterSTAT value);
diff --git a/src/gb/io.c b/src/gb/io.c
index 2db5d8c6e..8010e6a8a 100644
--- a/src/gb/io.c
+++ b/src/gb/io.c
@@ -392,7 +392,7 @@ void GBIOWrite(struct GB* gb, unsigned address, uint8_t value) {
 		return;
 	case REG_LCDC:
 		// TODO: handle GBC differences
-		GBVideoProcessDots(&gb->video);
+		GBVideoProcessDots(&gb->video, 0);
 		value = gb->video.renderer->writeVideoRegister(gb->video.renderer, address, value);
 		GBVideoWriteLCDC(&gb->video, value);
 		break;
@@ -406,13 +406,13 @@ void GBIOWrite(struct GB* gb, unsigned address, uint8_t value) {
 	case REG_SCX:
 	case REG_WY:
 	case REG_WX:
-		GBVideoProcessDots(&gb->video);
+		GBVideoProcessDots(&gb->video, 0);
 		value = gb->video.renderer->writeVideoRegister(gb->video.renderer, address, value);
 		break;
 	case REG_BGP:
 	case REG_OBP0:
 	case REG_OBP1:
-		GBVideoProcessDots(&gb->video);
+		GBVideoProcessDots(&gb->video, 0);
 		GBVideoWritePalette(&gb->video, address, value);
 		break;
 	case REG_STAT:
@@ -459,7 +459,7 @@ void GBIOWrite(struct GB* gb, unsigned address, uint8_t value) {
 				gb->memory.io[REG_BCPD] = gb->video.palette[gb->video.bcpIndex >> 1] >> (8 * (gb->video.bcpIndex & 1));
 				break;
 			case REG_BCPD:
-				GBVideoProcessDots(&gb->video);
+				GBVideoProcessDots(&gb->video, 0);
 				GBVideoWritePalette(&gb->video, address, value);
 				return;
 			case REG_OCPS:
@@ -468,7 +468,7 @@ void GBIOWrite(struct GB* gb, unsigned address, uint8_t value) {
 				gb->memory.io[REG_OCPD] = gb->video.palette[8 * 4 + (gb->video.ocpIndex >> 1)] >> (8 * (gb->video.ocpIndex & 1));
 				break;
 			case REG_OCPD:
-				GBVideoProcessDots(&gb->video);
+				GBVideoProcessDots(&gb->video, 0);
 				GBVideoWritePalette(&gb->video, address, value);
 				return;
 			case REG_SVBK:
diff --git a/src/gb/video.c b/src/gb/video.c
index 4478b079e..952b51322 100644
--- a/src/gb/video.c
+++ b/src/gb/video.c
@@ -310,7 +310,7 @@ void _endMode2(struct mTiming* timing, void* context, uint32_t cyclesLate) {
 
 void _endMode3(struct mTiming* timing, void* context, uint32_t cyclesLate) {
 	struct GBVideo* video = context;
-	GBVideoProcessDots(video);
+	GBVideoProcessDots(video, cyclesLate);
 	if (video->ly < GB_VIDEO_VERTICAL_PIXELS && video->p->memory.isHdma && video->p->memory.io[REG_HDMA5] != 0xFF) {
 		video->p->memory.hdmaRemaining = 0x10;
 		video->p->cpuBlocked = true;
@@ -400,12 +400,12 @@ static void _cleanOAM(struct GBVideo* video, int y) {
 	video->objMax = o;
 }
 
-void GBVideoProcessDots(struct GBVideo* video) {
+void GBVideoProcessDots(struct GBVideo* video, uint32_t cyclesLate) {
 	if (video->mode != 3) {
 		return;
 	}
 	int oldX = video->x;
-	video->x = (video->p->timing.masterCycles - video->dotClock + video->p->cpu->cycles) >> video->p->doubleSpeed;
+	video->x = (mTimingCurrentTime(&video->p->timing) - video->dotClock - cyclesLate) >> video->p->doubleSpeed;
 	if (video->x > GB_VIDEO_HORIZONTAL_PIXELS) {
 		video->x = GB_VIDEO_HORIZONTAL_PIXELS;
 	} else if (video->x < 0) {
@@ -786,6 +786,7 @@ void GBVideoSerialize(const struct GBVideo* video, struct GBSerializedState* sta
 	STORE_16LE(video->x, 0, &state->video.x);
 	STORE_16LE(video->ly, 0, &state->video.ly);
 	STORE_32LE(video->frameCounter, 0, &state->video.frameCounter);
+	STORE_32LE(video->dotClock, 0, &state->video.dotCounter);
 	state->video.vramCurrentBank = video->vramCurrentBank;
 
 	GBSerializedVideoFlags flags = 0;
@@ -814,6 +815,7 @@ void GBVideoDeserialize(struct GBVideo* video, const struct GBSerializedState* s
 	LOAD_16LE(video->x, 0, &state->video.x);
 	LOAD_16LE(video->ly, 0, &state->video.ly);
 	LOAD_32LE(video->frameCounter, 0, &state->video.frameCounter);
+	LOAD_32LE(video->dotClock, 0, &state->video.dotCounter);
 	video->vramCurrentBank = state->video.vramCurrentBank;
 
 	GBSerializedVideoFlags flags = state->video.flags;