diff --git a/devops/scripts/benchmarks/benches/compute.py b/devops/scripts/benchmarks/benches/compute.py
index d83a0d081af57..b014828b4f6b8 100644
--- a/devops/scripts/benchmarks/benches/compute.py
+++ b/devops/scripts/benchmarks/benches/compute.py
@@ -117,7 +117,7 @@ def benchmarks(self) -> list[Benchmark]:
         if options.sycl is None:
             return []
 
-        if options.ur_adapter == "cuda":
+        if options.ur_adapter == "cuda" or options.ur_adapter == "hip":
             return []
 
         benches = []
diff --git a/devops/scripts/benchmarks/benches/llamacpp.py b/devops/scripts/benchmarks/benches/llamacpp.py
index 86d41ed525292..d5b250b9eb943 100644
--- a/devops/scripts/benchmarks/benches/llamacpp.py
+++ b/devops/scripts/benchmarks/benches/llamacpp.py
@@ -82,7 +82,7 @@ def benchmarks(self) -> list[Benchmark]:
         if options.sycl is None:
             return []
 
-        if options.ur_adapter == "cuda":
+        if options.ur_adapter == "cuda" or options.ur_adapter == "hip":
             return []
 
         return [LlamaBench(self)]
diff --git a/devops/scripts/benchmarks/benches/syclbench.py b/devops/scripts/benchmarks/benches/syclbench.py
index 9854c92d338fc..f6a04918b81d3 100644
--- a/devops/scripts/benchmarks/benches/syclbench.py
+++ b/devops/scripts/benchmarks/benches/syclbench.py
@@ -56,6 +56,11 @@ def setup(self):
                 f"-DCMAKE_CXX_FLAGS=-fsycl -fsycl-targets=nvptx64-nvidia-cuda"
             ]
 
+        if options.ur_adapter == "hip":
+            configure_command += [
+                f"-DCMAKE_CXX_FLAGS=-fsycl -fsycl-targets=amdgcn-amd-amdhsa -Xsycl-target-backend --offload-arch={options.hip_arch}"
+            ]
+
         run(configure_command, add_sycl=True)
         run(f"cmake --build {build_path} -j {options.build_jobs}", add_sycl=True)
 
diff --git a/devops/scripts/benchmarks/benches/velocity.py b/devops/scripts/benchmarks/benches/velocity.py
index 493298dea8b10..df4082f620b8f 100644
--- a/devops/scripts/benchmarks/benches/velocity.py
+++ b/devops/scripts/benchmarks/benches/velocity.py
@@ -47,7 +47,7 @@ def benchmarks(self) -> list[Benchmark]:
         if options.sycl is None:
             return []
 
-        if options.ur_adapter == "cuda":
+        if options.ur_adapter == "cuda" or options.ur_adapter == "hip":
             return [
                 Hashtable(self),
                 Bitcracker(self),
@@ -83,6 +83,11 @@ def download_deps(self):
     def extra_cmake_args(self) -> list[str]:
         if options.ur_adapter == "cuda":
             return [f"-DUSE_NVIDIA_BACKEND=YES", f"-DUSE_SM=80"]
+        if options.ur_adapter == "hip":
+            return [
+                f"-DUSE_AMD_BACKEND=YES",
+                f"-DUSE_AMDHIP_BACKEND={options.hip_arch}",
+            ]
         return []
 
     def ld_libraries(self) -> list[str]:
diff --git a/devops/scripts/benchmarks/main.py b/devops/scripts/benchmarks/main.py
index 859aa96e50903..2d1b99386af87 100755
--- a/devops/scripts/benchmarks/main.py
+++ b/devops/scripts/benchmarks/main.py
@@ -446,13 +446,13 @@ def validate_and_parse_env_args(env_args):
         default=options.current_run_name,
     )
     parser.add_argument(
-        "--cudnn_directory",
+        "--cudnn-directory",
         type=str,
         help="Directory for cudnn library",
         default=None,
     )
     parser.add_argument(
-        "--cublas_directory",
+        "--cublas-directory",
         type=str,
         help="Directory for cublas library",
         default=None,
@@ -469,6 +469,12 @@ def validate_and_parse_env_args(env_args):
         help="Number of build jobs to run simultaneously",
         default=options.build_jobs,
     )
+    parser.add_argument(
+        "--hip-arch",
+        type=str,
+        help="HIP device architecture",
+        default=None,
+    )
 
     args = parser.parse_args()
     additional_env_vars = validate_and_parse_env_args(args.env)
@@ -497,6 +503,7 @@ def validate_and_parse_env_args(env_args):
     options.cublas_directory = args.cublas_directory
     options.custom_results_dir = args.results_dir
     options.build_jobs = args.build_jobs
+    options.hip_arch = args.hip_arch
 
     if args.build_igc and args.compute_runtime is None:
         parser.error("--build-igc requires --compute-runtime to be set")