diff --git a/src/core/PICA/gpu.cpp b/src/core/PICA/gpu.cpp
index adec82da..10d6b80b 100644
--- a/src/core/PICA/gpu.cpp
+++ b/src/core/PICA/gpu.cpp
@@ -224,21 +224,34 @@ Vertex GPU::getImmediateModeVertex() {
 	return v;
 }
 void GPU::fireDMA(u32 dest, u32 source, u32 size) {
+	log("[GPU] DMA of %08X bytes from %08X to %08X\n", size, source, dest);
 	constexpr u32 vramStart = VirtualAddrs::VramStart;
 	constexpr u32 vramSize = VirtualAddrs::VramSize;
 
 	const u32 fcramStart = mem.getLinearHeapVaddr();
 	constexpr u32 fcramSize = VirtualAddrs::FcramTotalSize;
 
+	// Shows whether this transfer is an FCRAM->VRAM transfer that's trivially optimizable
+	bool cpuToVRAM = true;
+
 	if (dest - vramStart >= vramSize || size > (vramSize - (dest - vramStart))) [[unlikely]] {
 		Helpers::panic("GPU DMA does not target VRAM");
 	}
 
-	if (source - fcramStart >= fcramSize || size > (fcramSize - (dest - fcramStart))) {
-		Helpers::panic("GPU DMA does not have FCRAM as its source");
+	if (source - fcramStart >= fcramSize || size > (fcramSize - (dest - fcramStart))) [[unlikely]] {
+		cpuToVRAM = false;
+		// Helpers::panic("GPU DMA does not have FCRAM as its source");
 	}
 
-	// Valid, optimized FCRAM->VRAM DMA. TODO: Is VRAM->VRAM DMA allowed?
-	u8* fcram = mem.getFCRAM();
-	std::memcpy(&vram[dest - vramStart], &fcram[source - fcramStart], size);
+	if (cpuToVRAM) [[likely]] {
+		// Valid, optimized FCRAM->VRAM DMA. TODO: Is VRAM->VRAM DMA allowed?
+		u8* fcram = mem.getFCRAM();
+		std::memcpy(&vram[dest - vramStart], &fcram[source - fcramStart], size);
+	} else {
+		printf("Non-trivially optimizable GPU DMA. Falling back to byte-by-byte transfer");
+
+		for (u32 i = 0; i < size; i++) {
+			mem.write8(dest + i, mem.read8(source + i));
+		}
+	}
 }