diff --git a/src/gt4py/cartesian/backend/dace_backend.py b/src/gt4py/cartesian/backend/dace_backend.py
index b7a9a0cece..5bcb8e9e08 100644
--- a/src/gt4py/cartesian/backend/dace_backend.py
+++ b/src/gt4py/cartesian/backend/dace_backend.py
@@ -920,7 +920,7 @@ def generate_extension(self) -> None:
 
 @register
 class DaceGPUBackend(BaseDaceBackend):
-    """DaCe python backend using gt4py.cartesian.gtc."""
+    """GPU DaCe python with an optimal KJI loop layout"""
 
     name = "dace:gpu"
     languages: ClassVar[dict] = {"computation": "cuda", "bindings": ["python"]}
@@ -933,3 +933,20 @@ class DaceGPUBackend(BaseDaceBackend):
 
     def generate_extension(self) -> None:
         return self.make_extension(uses_cuda=True)
+
+
+@register
+class DaceGPUBackendIJK(BaseDaceBackend):
+    """GPU DaCe python with an optimal IJK loop layout"""
+
+    name = "dace:gpu_IJK"
+    languages: ClassVar[dict] = {"computation": "cuda", "bindings": ["python"]}
+    storage_info: ClassVar[layout.LayoutInfo] = layout_registry.from_name(name)
+    MODULE_GENERATOR_CLASS = DaCeCUDAPyExtModuleGenerator
+    options: ClassVar[GTBackendOptions] = {
+        **BaseGTBackend.GT_BACKEND_OPTS,
+        "device_sync": {"versioning": True, "type": bool},
+    }
+
+    def generate_extension(self) -> None:
+        return self.make_extension(uses_cuda=True)
diff --git a/src/gt4py/cartesian/utils/compiler.py b/src/gt4py/cartesian/utils/compiler.py
index 409a4c7e09..ec8dc20de0 100644
--- a/src/gt4py/cartesian/utils/compiler.py
+++ b/src/gt4py/cartesian/utils/compiler.py
@@ -137,7 +137,7 @@ class GPUCompilerName(enum.Enum):
 class GPUConfiguration:
     name: GPUCompilerName
     """Name identifier of the compiler"""
-    gpu_compile_flags: list[str]
+    gpu_compile_flags: str
     """Compile flags for device code"""
     binary_path: str
     """Path to binaries for GPU compiler & tools"""
@@ -181,7 +181,7 @@ def gpu_configuration(optimization_level: str) -> GPUConfiguration:
 
     return GPUConfiguration(
         name=name,
-        gpu_compile_flags=gpu_compile_flags,
+        gpu_compile_flags=" ".join(gpu_compile_flags).strip(),
         binary_path=os.path.join(cuda_root, "bin"),
         include_path=os.path.join(cuda_root, "include"),
         library_path=library_path,
diff --git a/src/gt4py/storage/cartesian/layout_registry.py b/src/gt4py/storage/cartesian/layout_registry.py
index 4fadc0f7d0..f80c7e3f8d 100644
--- a/src/gt4py/storage/cartesian/layout_registry.py
+++ b/src/gt4py/storage/cartesian/layout_registry.py
@@ -66,6 +66,15 @@ def register(name: str, info: LayoutInfo) -> None:
         is_optimal_layout=layout_checker_factory(layout_maker_factory((2, 1, 0))),
     ),
 )
+register(
+    "dace:gpu_IJK",
+    LayoutInfo(
+        alignment=32,
+        device="gpu",
+        layout_map=layout_maker_factory((0, 1, 2)),
+        is_optimal_layout=layout_checker_factory(layout_maker_factory((0, 1, 2))),
+    ),
+)
 register(
     "debug",
     LayoutInfo(