diff --git a/Cargo.lock b/Cargo.lock
index fe2145f..afd0818 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -1471,6 +1471,18 @@ dependencies = [
  "zune-jpeg",
 ]
 
+[[package]]
+name = "image-compare"
+version = "0.4.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "96cd73af13ae2e7220a1c02fe7d6bb53be50612ba7fabbb5c88e7753645f1f3c"
+dependencies = [
+ "image",
+ "itertools",
+ "rayon",
+ "thiserror",
+]
+
 [[package]]
 name = "image-webp"
 version = "0.1.3"
@@ -1952,10 +1964,13 @@ dependencies = [
  "ash",
  "gfx-maths",
  "image",
+ "image-compare",
  "librashader",
  "librashader-runtime",
  "objc2 0.5.2",
  "objc2-metal",
+ "parking_lot",
+ "pollster",
  "wgpu",
  "wgpu-types",
  "windows 0.58.0",
diff --git a/librashader-test/Cargo.toml b/librashader-test/Cargo.toml
index 76e1e8e..e592294 100644
--- a/librashader-test/Cargo.toml
+++ b/librashader-test/Cargo.toml
@@ -13,6 +13,9 @@ image = "0.25.2"
 gfx-maths = "0.2.8"
 
 ash = "0.38.0+1.3.281"
+pollster = "0.3.0"
+parking_lot = "0.12.3"
+image-compare = "0.4.1"
 
 [target.'cfg(windows)'.dependencies.windows]
 workspace = true
diff --git a/librashader-test/src/lib.rs b/librashader-test/src/lib.rs
index 470bcfe..2abe71c 100644
--- a/librashader-test/src/lib.rs
+++ b/librashader-test/src/lib.rs
@@ -1,16 +1,3 @@
-mod triangle;
 
-pub fn add(left: u64, right: u64) -> u64 {
-    left + right
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn it_works() {
-        let result = add(2, 2);
-        assert_eq!(result, 4);
-    }
-}
+/// Render tests
+pub mod render;
diff --git a/librashader-test/src/triangle/d3d11.rs b/librashader-test/src/render/d3d11.rs
similarity index 100%
rename from librashader-test/src/triangle/d3d11.rs
rename to librashader-test/src/render/d3d11.rs
diff --git a/librashader-test/src/render/mod.rs b/librashader-test/src/render/mod.rs
new file mode 100644
index 0000000..362cb46
--- /dev/null
+++ b/librashader-test/src/render/mod.rs
@@ -0,0 +1,66 @@
+pub mod d3d11;
+pub mod wgpu;
+
+use std::path::Path;
+
+/// Test harness to set up a device, render a triangle, and apply a shader
+pub trait RenderTest {
+    /// Render a shader onto an image buffer, applying the provided shader.
+    ///
+    /// The test should render in linear colour space for proper comparison against
+    /// backends.
+    ///
+    /// For testing purposes, it is often that a single image will be reused with multiple
+    /// shader presets, so the actual image that a shader will be applied to
+    /// will often be part of the test harness object.
+    fn render(
+        &self,
+        path: impl AsRef<Path>,
+        frame_count: usize,
+    ) -> anyhow::Result<image::RgbaImage>;
+}
+
+#[cfg(test)]
+mod test {
+    use crate::render::RenderTest;
+    use image::codecs::png::PngEncoder;
+    use std::fs::File;
+
+    const IMAGE_PATH: &str = "../triangle.png";
+    const FILTER_PATH: &str = "../test/shaders_slang/crt/crt-royale.slangp";
+
+    #[test]
+    pub fn test_d3d11() -> anyhow::Result<()> {
+        let d3d11 = super::d3d11::Direct3D11::new(IMAGE_PATH)?;
+        let image = d3d11.render(FILTER_PATH, 100)?;
+
+        let out = File::create("out.png")?;
+        image.write_with_encoder(PngEncoder::new(out))?;
+        Ok(())
+    }
+
+    #[test]
+    pub fn test_wgpu() -> anyhow::Result<()> {
+        let wgpu = super::wgpu::Wgpu::new(IMAGE_PATH)?;
+        let image = wgpu.render(FILTER_PATH, 100)?;
+
+        let out = File::create("out.png")?;
+        image.write_with_encoder(PngEncoder::new(out))?;
+        Ok(())
+    }
+
+    #[test]
+    pub fn compare() -> anyhow::Result<()> {
+        let d3d11 = super::d3d11::Direct3D11::new(IMAGE_PATH)?;
+        let wgpu = super::wgpu::Wgpu::new(IMAGE_PATH)?;
+
+        let wgpu_image = wgpu.render(FILTER_PATH, 100)?;
+        let d3d11_image = d3d11.render(FILTER_PATH, 100)?;
+
+        let similarity = image_compare::rgba_hybrid_compare(&wgpu_image, &d3d11_image)?;
+
+        assert!(similarity.score > 0.95);
+
+        Ok(())
+    }
+}
diff --git a/librashader-test/src/render/wgpu.rs b/librashader-test/src/render/wgpu.rs
new file mode 100644
index 0000000..61ece45
--- /dev/null
+++ b/librashader-test/src/render/wgpu.rs
@@ -0,0 +1,241 @@
+use crate::render::RenderTest;
+use anyhow::anyhow;
+use image::RgbaImage;
+use librashader::runtime::wgpu::*;
+use librashader::runtime::Viewport;
+use librashader_runtime::image::{Image, UVDirection};
+use std::io::{Cursor, Write};
+use std::ops::{Deref, DerefMut};
+use std::path::Path;
+use std::sync::Arc;
+use wgpu::{Adapter, Device, Instance, Queue, Texture};
+use wgpu_types::{
+    BufferAddress, BufferDescriptor, BufferUsages, CommandEncoderDescriptor, ImageCopyBuffer,
+    ImageDataLayout, Maintain, TextureDescriptor, TextureDimension, TextureFormat, TextureUsages,
+};
+
+use parking_lot::Mutex;
+
+pub struct Wgpu {
+    instance: Instance,
+    adapter: Adapter,
+    device: Arc<Device>,
+    queue: Arc<Queue>,
+    image: Image,
+    texture: Arc<Texture>,
+}
+
+struct BufferDimensions {
+    width: usize,
+    height: usize,
+    unpadded_bytes_per_row: usize,
+    padded_bytes_per_row: usize,
+}
+
+impl BufferDimensions {
+    fn new(width: usize, height: usize) -> Self {
+        let bytes_per_pixel = size_of::<u32>();
+        let unpadded_bytes_per_row = width * bytes_per_pixel;
+        let align = wgpu::COPY_BYTES_PER_ROW_ALIGNMENT as usize;
+        let padded_bytes_per_row_padding = (align - unpadded_bytes_per_row % align) % align;
+        let padded_bytes_per_row = unpadded_bytes_per_row + padded_bytes_per_row_padding;
+        Self {
+            width,
+            height,
+            unpadded_bytes_per_row,
+            padded_bytes_per_row,
+        }
+    }
+}
+
+impl RenderTest for Wgpu {
+    fn render(&self, path: impl AsRef<Path>, frame_count: usize) -> anyhow::Result<RgbaImage> {
+        let mut chain = FilterChain::load_from_path(
+            path,
+            Arc::clone(&self.device),
+            Arc::clone(&self.queue),
+            Some(&FilterChainOptions {
+                force_no_mipmaps: false,
+                enable_cache: false,
+                adapter_info: None,
+            }),
+        )?;
+
+        let mut cmd = self
+            .device
+            .create_command_encoder(&CommandEncoderDescriptor { label: None });
+
+        let output_tex = self.device.create_texture(&TextureDescriptor {
+            label: None,
+            size: self.texture.size(),
+            mip_level_count: 1,
+            sample_count: 1,
+            dimension: TextureDimension::D2,
+            format: TextureFormat::Rgba8Unorm,
+            usage: TextureUsages::RENDER_ATTACHMENT | TextureUsages::COPY_SRC,
+            view_formats: &[wgpu::TextureFormat::Rgba8Unorm],
+        });
+
+        let buffer_dimensions =
+            BufferDimensions::new(output_tex.width() as usize, output_tex.height() as usize);
+        let output_buf = Arc::new(self.device.create_buffer(&BufferDescriptor {
+            label: None,
+            size: (buffer_dimensions.padded_bytes_per_row * buffer_dimensions.height)
+                as BufferAddress, // 4bpp
+            usage: BufferUsages::COPY_DST | BufferUsages::MAP_READ,
+            mapped_at_creation: false,
+        }));
+
+        let view = output_tex.create_view(&wgpu::TextureViewDescriptor::default());
+        let output = WgpuOutputView::new_from_raw(
+            &view,
+            output_tex.size().into(),
+            TextureFormat::Rgba8Unorm,
+        );
+
+        chain.frame(
+            Arc::clone(&self.texture),
+            &Viewport::new_render_target_sized_origin(output, None)?,
+            &mut cmd,
+            frame_count,
+            None,
+        )?;
+
+        cmd.copy_texture_to_buffer(
+            output_tex.as_image_copy(),
+            ImageCopyBuffer {
+                buffer: &output_buf,
+                layout: ImageDataLayout {
+                    offset: 0,
+                    bytes_per_row: Some(buffer_dimensions.padded_bytes_per_row as u32),
+                    rows_per_image: None,
+                },
+            },
+            output_tex.size(),
+        );
+
+        let si = self.queue.submit([cmd.finish()]);
+        self.device.poll(Maintain::WaitForSubmissionIndex(si));
+
+        let capturable = Arc::clone(&output_buf);
+
+        let mut pixels = Arc::new(Mutex::new(Vec::new()));
+
+        let pixels_async = Arc::clone(&pixels);
+        output_buf
+            .slice(..)
+            .map_async(wgpu::MapMode::Read, move |r| {
+                if r.is_ok() {
+                    let buffer = capturable.slice(..).get_mapped_range();
+                    let mut pixels = pixels_async.lock();
+                    pixels.resize(buffer.len(), 0);
+
+                    let mut cursor = Cursor::new(pixels.deref_mut());
+                    for chunk in buffer.chunks(buffer_dimensions.padded_bytes_per_row) {
+                        cursor
+                            .write_all(&chunk[..buffer_dimensions.unpadded_bytes_per_row])
+                            .unwrap()
+                    }
+
+                    cursor.into_inner();
+                }
+                capturable.unmap();
+            });
+
+        self.device.poll(Maintain::Wait);
+
+        if pixels.lock().len() == 0 {
+            return Err(anyhow!("failed to copy pixels from buffer"));
+        }
+
+        let image = RgbaImage::from_raw(
+            output_tex.width(),
+            output_tex.height(),
+            pixels.lock().to_vec(),
+        )
+        .ok_or(anyhow!("Unable to create image from data"))?;
+
+        Ok(image)
+    }
+}
+
+impl Wgpu {
+    pub fn new(image: impl AsRef<Path>) -> anyhow::Result<Self> {
+        pollster::block_on(async {
+            let instance = wgpu::Instance::default();
+            let adapter = instance
+                .request_adapter(&wgpu::RequestAdapterOptions {
+                    power_preference: wgpu::PowerPreference::default(),
+                    compatible_surface: None,
+                    force_fallback_adapter: false,
+                })
+                .await
+                .ok_or(anyhow!("Couldn't request WGPU adapter"))?;
+
+            let (device, queue) = adapter
+                .request_device(
+                    &wgpu::DeviceDescriptor {
+                        required_features: wgpu::Features::ADDRESS_MODE_CLAMP_TO_BORDER
+                            | wgpu::Features::PIPELINE_CACHE
+                            | wgpu::Features::TEXTURE_ADAPTER_SPECIFIC_FORMAT_FEATURES
+                            | wgpu::Features::FLOAT32_FILTERABLE,
+                        required_limits: wgpu::Limits::default(),
+                        label: None,
+                        memory_hints: Default::default(),
+                    },
+                    None,
+                )
+                .await?;
+            let (image, texture) = Self::load_image(&device, &queue, image)?;
+
+            Ok(Self {
+                instance,
+                adapter,
+                device: Arc::new(device),
+                queue: Arc::new(queue),
+                image,
+                texture: Arc::new(texture),
+            })
+        })
+    }
+
+    fn load_image(
+        device: &Device,
+        queue: &Queue,
+        path: impl AsRef<Path>,
+    ) -> anyhow::Result<(Image, Texture)> {
+        let image = Image::load(path, UVDirection::TopLeft)?;
+        let texture = device.create_texture(&TextureDescriptor {
+            size: image.size.into(),
+            mip_level_count: 1,
+            sample_count: 1,
+            dimension: TextureDimension::D2,
+            usage: TextureUsages::TEXTURE_BINDING | TextureUsages::COPY_DST,
+            format: wgpu::TextureFormat::Rgba8Unorm,
+            view_formats: &[wgpu::TextureFormat::Rgba8Unorm],
+            label: None,
+        });
+
+        queue.write_texture(
+            wgpu::ImageCopyTexture {
+                texture: &texture,
+                mip_level: 0,
+                origin: wgpu::Origin3d::ZERO,
+                aspect: wgpu::TextureAspect::All,
+            },
+            &image.bytes,
+            wgpu::ImageDataLayout {
+                offset: 0,
+                bytes_per_row: Some(4 * image.size.width),
+                rows_per_image: None,
+            },
+            image.size.into(),
+        );
+
+        let si = queue.submit([]);
+
+        device.poll(Maintain::WaitForSubmissionIndex(si));
+
+        Ok((image, texture))
+    }
+}
diff --git a/librashader-test/src/triangle/mod.rs b/librashader-test/src/triangle/mod.rs
deleted file mode 100644
index cf98831..0000000
--- a/librashader-test/src/triangle/mod.rs
+++ /dev/null
@@ -1,37 +0,0 @@
-mod d3d11;
-
-use std::path::Path;
-
-/// Test harness to set up a device, render a triangle, and apply a shader
-pub trait TriangleTest {
-    /// Render a triangle to an image buffer, applying the provided shader.
-    ///
-    /// The test should render in linear colour space for proper comparison against
-    /// backends.
-    fn triangle(
-        &self,
-        image: impl AsRef<Path>,
-        path: impl AsRef<Path>,
-        frame_count: usize,
-    ) -> anyhow::Result<image::RgbaImage>;
-}
-
-#[cfg(test)]
-mod test {
-    use crate::triangle::TriangleTest;
-    use image::codecs::png::PngEncoder;
-    use std::fs::File;
-
-    const IMAGE_PATH: &str = "../triangle.png";
-    const FILTER_PATH: &str = "../test/shaders_slang/crt/crt-royale.slangp";
-
-    #[test]
-    pub fn test_d3d11() -> anyhow::Result<()> {
-        let d3d11 = super::d3d11::Direct3D11::new()?;
-        let image = d3d11.triangle(IMAGE_PATH, FILTER_PATH, 100)?;
-
-        let out = File::create("out.png")?;
-        image.write_with_encoder(PngEncoder::new(out))?;
-        Ok(())
-    }
-}