diff --git a/include/mgba/internal/ds/gx.h b/include/mgba/internal/ds/gx.h
index 0502f4425..25e9cb242 100644
--- a/include/mgba/internal/ds/gx.h
+++ b/include/mgba/internal/ds/gx.h
@@ -200,7 +200,7 @@ void DSGXAssociateRenderer(struct DSGX* video, struct DSGXRenderer* renderer);
 uint16_t DSGXWriteRegister(struct DSGX*, uint32_t address, uint16_t value);
 uint32_t DSGXWriteRegister32(struct DSGX*, uint32_t address, uint32_t value);
 
-void DSGXSwapBuffers(struct DSGX*);
+void DSGXFlush(struct DSGX*);
 void DSGXUpdateGXSTAT(struct DSGX*);
 
 CXX_GUARD_END
diff --git a/include/mgba/internal/ds/gx/software.h b/include/mgba/internal/ds/gx/software.h
index 504a97c6f..7c75cd69b 100644
--- a/include/mgba/internal/ds/gx/software.h
+++ b/include/mgba/internal/ds/gx/software.h
@@ -77,6 +77,7 @@ struct DSGXSoftwareRenderer {
 
 	uint16_t depthBuffer[DS_VIDEO_HORIZONTAL_PIXELS];
 	color_t* scanlineCache;
+	bool flushPending;
 
 	struct DSGXVertex* verts;
 };
diff --git a/src/ds/gx.c b/src/ds/gx.c
index 9e1d15bcd..d57831fab 100644
--- a/src/ds/gx.c
+++ b/src/ds/gx.c
@@ -901,6 +901,7 @@ static void DSGXUnpackCommand(struct DSGX* gx, uint32_t command) {
 	gx->outstandingParams[2] = _gxCommandParams[gx->outstandingCommand[2]];
 	gx->outstandingParams[3] = _gxCommandParams[gx->outstandingCommand[3]];
 	_flushOutstanding(gx);
+	DSGXUpdateGXSTAT(gx);
 }
 
 static void DSGXWriteFIFO(struct DSGX* gx, struct DSGXEntry entry) {
@@ -1049,14 +1050,14 @@ uint32_t DSGXWriteRegister32(struct DSGX* gx, uint32_t address, uint32_t value)
 	return value;
 }
 
-void DSGXSwapBuffers(struct DSGX* gx) {
-	gx->swapBuffers = false;
-
-	gx->renderer->setRAM(gx->renderer, gx->vertexBuffer[gx->bufferIndex], gx->polygonBuffer[gx->bufferIndex], gx->polygonIndex);
-
-	gx->bufferIndex ^= 1;
-	gx->vertexIndex = 0;
-	gx->polygonIndex = 0;
+void DSGXFlush(struct DSGX* gx) {
+	if (gx->swapBuffers) {
+		gx->renderer->setRAM(gx->renderer, gx->vertexBuffer[gx->bufferIndex], gx->polygonBuffer[gx->bufferIndex], gx->polygonIndex);
+		gx->swapBuffers = false;
+		gx->bufferIndex ^= 1;
+		gx->vertexIndex = 0;
+		gx->polygonIndex = 0;
+	}
 
 	if (CircleBufferSize(&gx->fifo)) {
 		mTimingSchedule(&gx->p->ds9.timing, &gx->fifoEvent, 0);
diff --git a/src/ds/gx/software.c b/src/ds/gx/software.c
index fe790ad27..d1c4d57e4 100644
--- a/src/ds/gx/software.c
+++ b/src/ds/gx/software.c
@@ -225,12 +225,12 @@ static void DSGXSoftwareRendererInit(struct DSGXRenderer* renderer) {
 	DSGXSoftwareEdgeListInit(&softwareRenderer->activeEdges, DS_GX_POLYGON_BUFFER_SIZE);
 	DSGXSoftwareSpanListInit(&softwareRenderer->activeSpans, DS_GX_POLYGON_BUFFER_SIZE / 2);
 	softwareRenderer->bucket = anonymousMemoryMap(sizeof(*softwareRenderer->bucket) * DS_GX_POLYGON_BUFFER_SIZE);
-	softwareRenderer->scanlineCache = anonymousMemoryMap(sizeof(color_t) * DS_VIDEO_HORIZONTAL_PIXELS * 48);
+	softwareRenderer->scanlineCache = anonymousMemoryMap(sizeof(color_t) * DS_VIDEO_VERTICAL_PIXELS * DS_VIDEO_HORIZONTAL_PIXELS);
 }
 
 static void DSGXSoftwareRendererReset(struct DSGXRenderer* renderer) {
 	struct DSGXSoftwareRenderer* softwareRenderer = (struct DSGXSoftwareRenderer*) renderer;
-	// TODO
+	softwareRenderer->flushPending = false;
 }
 
 static void DSGXSoftwareRendererDeinit(struct DSGXRenderer* renderer) {
@@ -239,7 +239,7 @@ static void DSGXSoftwareRendererDeinit(struct DSGXRenderer* renderer) {
 	DSGXSoftwareEdgeListDeinit(&softwareRenderer->activeEdges);	
 	DSGXSoftwareSpanListDeinit(&softwareRenderer->activeSpans);
 	mappedMemoryFree(softwareRenderer->bucket, sizeof(*softwareRenderer->bucket) * DS_GX_POLYGON_BUFFER_SIZE);
-	mappedMemoryFree(softwareRenderer->scanlineCache, sizeof(color_t) * DS_VIDEO_HORIZONTAL_PIXELS * 48);
+	mappedMemoryFree(softwareRenderer->scanlineCache, sizeof(color_t) * DS_VIDEO_VERTICAL_PIXELS * DS_VIDEO_HORIZONTAL_PIXELS);
 }
 
 static void DSGXSoftwareRendererInvalidateTex(struct DSGXRenderer* renderer, int slot) {
@@ -250,6 +250,7 @@ static void DSGXSoftwareRendererInvalidateTex(struct DSGXRenderer* renderer, int
 static void DSGXSoftwareRendererSetRAM(struct DSGXRenderer* renderer, struct DSGXVertex* verts, struct DSGXPolygon* polys, unsigned polyCount) {
 	struct DSGXSoftwareRenderer* softwareRenderer = (struct DSGXSoftwareRenderer*) renderer;
 
+	softwareRenderer->flushPending = true;
 	softwareRenderer->verts = verts;
 	DSGXSoftwarePolygonListClear(&softwareRenderer->activePolys);
 	DSGXSoftwareEdgeListClear(&softwareRenderer->activeEdges);
@@ -351,6 +352,9 @@ static void DSGXSoftwareRendererSetRAM(struct DSGXRenderer* renderer, struct DSG
 
 static void DSGXSoftwareRendererDrawScanline(struct DSGXRenderer* renderer, int y) {
 	struct DSGXSoftwareRenderer* softwareRenderer = (struct DSGXSoftwareRenderer*) renderer;
+	if (!softwareRenderer->flushPending) {
+		return;
+	}
 	DSGXSoftwareSpanListClear(&softwareRenderer->activeSpans);
 	memset(softwareRenderer->bucket, 0, sizeof(*softwareRenderer->bucket) * DS_GX_POLYGON_BUFFER_SIZE);
 	int i;
@@ -383,7 +387,6 @@ static void DSGXSoftwareRendererDrawScanline(struct DSGXRenderer* renderer, int
 	}
 	qsort(DSGXSoftwareSpanListGetPointer(&softwareRenderer->activeSpans, 0), DSGXSoftwareSpanListSize(&softwareRenderer->activeSpans), sizeof(struct DSGXSoftwareSpan), _spanSort);
 
-	y %= 48;
 	color_t* scanline = &softwareRenderer->scanlineCache[DS_VIDEO_HORIZONTAL_PIXELS * y];
 
 	int nextSpanX = DS_VIDEO_HORIZONTAL_PIXELS;
@@ -438,10 +441,12 @@ static void DSGXSoftwareRendererDrawScanline(struct DSGXRenderer* renderer, int
 			}
 		}
 	}
+	if (y == DS_VIDEO_VERTICAL_PIXELS - 1) {
+		softwareRenderer->flushPending = false;
+	}
 }
 
 static void DSGXSoftwareRendererGetScanline(struct DSGXRenderer* renderer, int y, color_t** output) {
 	struct DSGXSoftwareRenderer* softwareRenderer = (struct DSGXSoftwareRenderer*) renderer;
-	y %= 48;
 	*output = &softwareRenderer->scanlineCache[DS_VIDEO_HORIZONTAL_PIXELS * y];
 }
diff --git a/src/ds/video.c b/src/ds/video.c
index c16f52ad1..ab75406ec 100644
--- a/src/ds/video.c
+++ b/src/ds/video.c
@@ -290,9 +290,7 @@ void _startHdraw9(struct mTiming* timing, void* context, uint32_t cyclesLate) {
 		video->p->ds9.memory.io[DS_REG_DISPSTAT >> 1] = GBARegisterDISPSTATFillInVblank(dispstat);
 		if (video->frameskipCounter <= 0) {
 			video->renderer->finishFrame(video->renderer);
-			if (video->p->gx.swapBuffers) {
-				DSGXSwapBuffers(&video->p->gx);
-			}
+			DSGXFlush(&video->p->gx);
 		}
 		if (GBARegisterDISPSTATIsVblankIRQ(dispstat)) {
 			DSRaiseIRQ(video->p->ds9.cpu, video->p->ds9.memory.io, DS_IRQ_VBLANK);