diff --git a/.github/workflows/sycl-nightly.yml b/.github/workflows/sycl-nightly.yml
index 06b8b50719dc7..73ace0226b3f0 100644
--- a/.github/workflows/sycl-nightly.yml
+++ b/.github/workflows/sycl-nightly.yml
@@ -323,7 +323,7 @@ jobs:
       # functionality, make sure Linux/Windows names follow the same pattern.
       toolchain_artifact_filename: sycl_windows.tar.gz
       # Disable the spirv-dis requirement as to not require SPIR-V Tools.
-      build_configure_extra_args: -DLLVM_SPIRV_ENABLE_LIBSPIRV_DIS=off -DSYCL_BUILD_INFO="Nightly ${{ needs.get_date.outputs.date }}"
+      build_configure_extra_args: --cuda -DLLVM_SPIRV_ENABLE_LIBSPIRV_DIS=off -DSYCL_BUILD_INFO="Nightly ${{ needs.get_date.outputs.date }}"
       build_target: all
 
   e2e-win:
@@ -337,7 +337,7 @@ jobs:
       fail-fast: false
       matrix:
         enable_new_offload_model: ['False', 'True']
-        gpu_type: ['gen12', 'arc', 'bmg']
+        gpu_type: ['gen12', 'arc', 'bmg', 'cuda']
         include:
           - gpu_type: 'gen12'
             name: Intel Gen12 GPU
@@ -351,11 +351,16 @@ jobs:
             name: Intel Battlemage GPU
             runner: '["Windows", "bmg"]'
 
+          - gpu_type: 'cuda'
+            name: NVIDIA A100 GPU
+            runner: '["Windows", "cuda"]'
+            target_devices: cuda:gpu
+
     uses: ./.github/workflows/sycl-windows-run-tests.yml
     with:
       name: ${{ matrix.name }}
       runner: ${{ matrix.runner }}
-      target_devices: level_zero:gpu
+      target_devices: ${{ matrix.target_devices || 'level_zero:gpu' }}
       toolchain_artifact_filename: ${{ needs.build-win.outputs.toolchain_artifact_filename }}
       extra_lit_opts: --param enable_new_offload_model=${{ matrix.enable_new_offload_model }}
 
diff --git a/.github/workflows/sycl-windows-build.yml b/.github/workflows/sycl-windows-build.yml
index da370d946fe8a..4b21bf63f2f39 100644
--- a/.github/workflows/sycl-windows-build.yml
+++ b/.github/workflows/sycl-windows-build.yml
@@ -105,10 +105,10 @@ jobs:
     outputs:
       build_conclusion: ${{ steps.build.conclusion }}
     env:
-      # Intel GPU architectures of the run-only E2E runners in
+      # AOT GPU architectures of the run-only E2E runners in
       # sycl-windows-precommit.yml. Keep in sync when adding or removing one.
       # Gen12 and older (gen12 runner) are not supported by Windows ocloc.
-      GPU_AOT_TARGET: intel_gpu_acm_g10,intel_gpu_bmg_g21
+      GPU_AOT_TARGET: intel_gpu_acm_g10,intel_gpu_bmg_g21${{ (contains(inputs.build_configure_extra_args, '--cuda') && ',nvidia_gpu_sm_80') || '' }}
     steps:
     - name:  Detect hung tests
       if: always()
@@ -262,7 +262,7 @@ jobs:
         testing_mode: build-only
         target_devices: all
         binaries_artifact: ${{ inputs.e2e_binaries_artifact }}
-        extra_lit_opts: --param sycl_build_targets="spir" --param gpu_aot_target=${{ env.GPU_AOT_TARGET }}
+        extra_lit_opts: --param sycl_build_targets="spir${{ (contains(inputs.build_configure_extra_args, '--cuda') && ';nvidia') || '' }}" --param gpu_aot_target=${{ env.GPU_AOT_TARGET }}
         cxx: ${{ inputs.cxx }}
 
     - name: Build E2E tests with New Offload Model
@@ -277,7 +277,7 @@ jobs:
         testing_mode: build-only
         target_devices: all
         binaries_artifact: ${{ inputs.e2e_binaries_new_offload_model_artifact }}
-        extra_lit_opts: --param sycl_build_targets="spir" --param enable_new_offload_model=True --param gpu_aot_target=${{ env.GPU_AOT_TARGET }}
+        extra_lit_opts: --param sycl_build_targets="spir${{ (contains(inputs.build_configure_extra_args, '--cuda') && ';nvidia') || '' }}" --param enable_new_offload_model=True --param gpu_aot_target=${{ env.GPU_AOT_TARGET }}
         cxx: ${{ inputs.cxx }}
 
     - name:  Detect hung tests
diff --git a/sycl/test-e2e/lit.cfg.py b/sycl/test-e2e/lit.cfg.py
index 55aaa7d3c644c..eb7d9de0a158e 100644
--- a/sycl/test-e2e/lit.cfg.py
+++ b/sycl/test-e2e/lit.cfg.py
@@ -775,7 +775,9 @@ def remove_level_zero_suffix(devices):
     if dev not in available_devices[be] and not "arch-" in dev:
         lit_config.error("Unsupported device {}".format(d))
 
-if "cuda:gpu" in config.sycl_devices:
+if "cuda:gpu" in config.sycl_devices or (
+    config.test_mode == "build-only" and "target-nvidia" in config.sycl_build_targets
+):
     if "CUDA_PATH" not in os.environ:
         if platform.system() == "Windows":
             cuda_root = r"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA"
@@ -836,7 +838,7 @@ def remove_level_zero_suffix(devices):
 
     # Detect CUDA version and add version-specific features for conditional test execution
     cuda_version_json = os.path.join(os.environ["CUDA_PATH"], "version.json")
-    if os.path.exists(cuda_version_json):
+    if config.test_mode != "build-only" and os.path.exists(cuda_version_json):
         try:
             with open(cuda_version_json, "r") as f:
                 version_data = json.load(f)