diff --git a/tests/test_wgpu_error_fatal.py b/tests/test_wgpu_error_fatal.py
new file mode 100644
index 00000000..03744a31
--- /dev/null
+++ b/tests/test_wgpu_error_fatal.py
@@ -0,0 +1,108 @@
+import wgpu
+from testutils import run_tests
+from renderutils import render_to_texture
+from pytest import raises
+
+
+# from test_wgpu_native_render.py
+default_vertex_shader = """
+@vertex
+fn vs_main(@builtin(vertex_index) vertex_index : u32) -> @builtin(position) vec4<f32> {
+    var positions: array<vec3<f32>, 4> = array<vec3<f32>, 4>(
+        vec3<f32>(-0.5, -0.5, 0.1),
+        vec3<f32>(-0.5,  0.5, 0.1),
+        vec3<f32>( 0.5, -0.5, 0.1),
+        vec3<f32>( 0.5,  0.5, 0.1),
+    );
+    let p: vec3<f32> = positions[vertex_index];
+    return vec4<f32>(p, 1.0);
+}
+"""
+
+# from test_wgpu_native_errors.py
+dedent = lambda s: s.replace("\n        ", "\n").strip()
+
+
+def test_unreachable():
+    # panicked at ~\naga\src\back\spv\block.rs:2401:56:
+    # https://github.com/gfx-rs/wgpu/issues/4517
+    # real world occurance: https://www.shadertoy.com/view/NsffD2
+    # unreachable code due to a pointer and swizzling back.
+    # crashes in wgpuDeviceCreateRenderPipeline, not caught in the shader module.
+    # passes naga validation and wgsl translation, panics to spv translation (which can be caught via subprocess).
+    device = wgpu.utils.get_default_device()
+
+    fragment_shader = """
+    fn test(rng: ptr<function, u32>) {}
+
+    fn woops(uv: vec2<u32>) {
+        var rngs = vec3<u32>(1, 2, 3);
+        test(&rngs.x);
+    }
+
+    @fragment
+    fn fs_main() -> @location(0) vec4<f32> {
+        return vec4<f32>(1.0, 0.499, 0.0, 1.0);
+    }
+    """
+    shader_source = default_vertex_shader + fragment_shader
+
+    bind_group = None
+    pipeline_layout = device.create_pipeline_layout(bind_group_layouts=[])
+    render_args = device, shader_source, pipeline_layout, bind_group
+
+    # would be great to have some kind of error instead of a crash
+    with raises(wgpu.GPUError):
+        render_to_texture(*render_args, size=(64, 64))
+
+
+def test_impossible_loop():
+    # panicked at src\lib.rs:582:5:
+    # should be captured: https://github.com/gfx-rs/wgpu/issues/5926#issuecomment-2216839446
+    # some use cases include the increment var being related to a uniform like iTime.
+    # other cases are generated code that is just wrong.
+    # another case is a translation error from glsl to wgsl where one statement is lost.
+    # https://github.com/gfx-rs/wgpu/issues/6208
+
+    device = wgpu.utils.get_default_device()
+
+    fragment_shader = """
+    @fragment
+    fn fs_main() -> @location(0) vec4<f32> {
+        var a = 0.0;
+        for (var i = 0; i < 10; i -= 1) {
+            a += 0.1;
+        }
+        return vec4<f32>(1.0, a, 0.0, 1.0);
+    }
+    """
+
+    # we do see this error logged, but not captured and raised.
+    # apparently the device gets lost because it takes too long (infinite loop).
+    # Some compilers detect this (DX12) and raise a different error.
+    # the python process crashes.
+    # the additional backtrace might be a bug upstream.
+    expected = """
+        Error in wgpuQueueSubmit: Validation Error
+
+    Caused by:
+    Parent device is lost
+    """
+
+    shader_source = default_vertex_shader + fragment_shader
+    shader_source = dedent(shader_source)
+    expected = dedent(expected)
+
+    bind_group = None
+    pipeline_layout = device.create_pipeline_layout(bind_group_layouts=[])
+    render_args = device, shader_source, pipeline_layout, bind_group
+
+    with raises(wgpu.GPUError) as err:
+        render_to_texture(*render_args, size=(64, 64))
+
+    error = err.value.message
+    assert error == expected, f"Expected:\n\n{expected}"
+
+
+if __name__ == "__main__":
+    run_tests(globals())