diff --git a/gpu.c b/gpu.c
index 25fb473..3404aa1 100644
--- a/gpu.c
+++ b/gpu.c
@@ -477,23 +477,6 @@ void gpu_init(gpu_ctx_t *ctx, const gpu_config_t *config)
 		assert(result == VK_SUCCESS);
 	}
 
-	// Create vertex buffer
-	ctx->vertex_data_size = config->vertex_data_size;
-	const VkMemoryPropertyFlags vertex_buf_props
-	    = VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
-	    | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
-	create_buffer(
-	    ctx, ctx->vertex_data_size, VK_BUFFER_USAGE_VERTEX_BUFFER_BIT,
-	    vertex_buf_props, &ctx->vertex_buf, &ctx->vertex_buf_mem);
-
-	// Copy vertex data into vertex buffer
-	void *vertex_buf_ptr;
-	vkMapMemory(
-	    ctx->dev, ctx->vertex_buf_mem, 0, ctx->vertex_data_size, 0,
-	    &vertex_buf_ptr);
-	memcpy(vertex_buf_ptr, config->vertex_data, ctx->vertex_data_size);
-	vkUnmapMemory(ctx->dev, ctx->vertex_buf_mem);
-
 	// Create command pool
 	const VkCommandPoolCreateInfo cmd_pool_config = {
 		.sType = VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO,
@@ -532,12 +515,59 @@ void gpu_init(gpu_ctx_t *ctx, const gpu_config_t *config)
 	result
 	    = vkCreateFence(ctx->dev, &fence_config, nullptr, &ctx->in_flight);
 	assert(result == VK_SUCCESS);
+
+	// Set up staging and vertex buffers
+	ctx->vertex_data_size = config->vertex_data_size;
+	VkBuffer staging_buf;
+	VkDeviceMemory staging_buf_mem;
+	create_buffer(
+	    ctx, ctx->vertex_data_size, VK_BUFFER_USAGE_TRANSFER_SRC_BIT,
+	    VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
+	        | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
+	    &staging_buf, &staging_buf_mem);
+	create_buffer(
+	    ctx, ctx->vertex_data_size,
+	    VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_VERTEX_BUFFER_BIT,
+	    VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT, &ctx->vertex_buf,
+	    &ctx->vertex_buf_mem);
+
+	// Copy vertex data into staging buffer
+	void *staging_buf_ptr;
+	vkMapMemory(
+	    ctx->dev, staging_buf_mem, 0, ctx->vertex_data_size, 0,
+	    &staging_buf_ptr);
+	memcpy(staging_buf_ptr, config->vertex_data, ctx->vertex_data_size);
+	vkUnmapMemory(ctx->dev, staging_buf_mem);
+
+	// Copy staging buffer to vertex buffer
+	VkCommandBufferBeginInfo cmd_begin = {
+		.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO,
+		.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT,
+	};
+	vkBeginCommandBuffer(ctx->cmd_buf, &cmd_begin);
+	VkBufferCopy cmd_copy = { .size = ctx->vertex_data_size };
+	vkCmdCopyBuffer(
+	    ctx->cmd_buf, staging_buf, ctx->vertex_buf, 1, &cmd_copy);
+	vkEndCommandBuffer(ctx->cmd_buf);
+	VkSubmitInfo copy_submit = {
+		.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO,
+		.commandBufferCount = 1,
+		.pCommandBuffers = &ctx->cmd_buf,
+	};
+	vkQueueSubmit(ctx->queue, 1, &copy_submit, VK_NULL_HANDLE);
+	vkQueueWaitIdle(ctx->queue);
+
+	// Clean up staging buffer
+	vkDestroyBuffer(ctx->dev, staging_buf, nullptr);
+	vkFreeMemory(ctx->dev, staging_buf_mem, nullptr);
 }
 
 void gpu_cleanup(const gpu_ctx_t *ctx)
 {
 	vkDeviceWaitIdle(ctx->dev);
 
+	vkDestroyBuffer(ctx->dev, ctx->vertex_buf, nullptr);
+	vkFreeMemory(ctx->dev, ctx->vertex_buf_mem, nullptr);
 	vkDestroySemaphore(ctx->dev, ctx->img_avail, nullptr);
 	vkDestroySemaphore(ctx->dev, ctx->render_fin, nullptr);
 	vkDestroyFence(ctx->dev, ctx->in_flight, nullptr);
@@ -552,8 +582,6 @@ void gpu_cleanup(const gpu_ctx_t *ctx)
 		vkDestroyImageView(ctx->dev, ctx->swapchain_img_views[i], nullptr);
 	}
 	vkDestroySwapchainKHR(ctx->dev, ctx->swapchain, nullptr);
-	vkDestroyBuffer(ctx->dev, ctx->vertex_buf, nullptr);
-	vkFreeMemory(ctx->dev, ctx->vertex_buf_mem, nullptr);
 	vkDestroyDevice(ctx->dev, nullptr);
 	vkDestroySurfaceKHR(ctx->inst, ctx->surface, nullptr);
 	vkDestroyInstance(ctx->inst, nullptr);
diff --git a/gpu.h b/gpu.h
index 2a50499..f7d5fe0 100644
--- a/gpu.h
+++ b/gpu.h
@@ -31,9 +31,9 @@ typedef struct {
 	VkFence in_flight;
 	VkSemaphore render_fin;
 	VkSemaphore img_avail;
+	size_t vertex_data_size;
 	VkBuffer vertex_buf;
 	VkDeviceMemory vertex_buf_mem;
-	size_t vertex_data_size;
 } gpu_ctx_t;
 
 typedef struct {