diff --git a/superbench/benchmarks/model_benchmarks/pytorch_base.py b/superbench/benchmarks/model_benchmarks/pytorch_base.py
index de06b35d0..f2428f2c6 100644
--- a/superbench/benchmarks/model_benchmarks/pytorch_base.py
+++ b/superbench/benchmarks/model_benchmarks/pytorch_base.py
@@ -591,8 +591,8 @@ def _benchmark(self):
         Run the benchmark then handle post-run model log save/compare.
         Set SB_ENABLE_PYTORCH_PROFILER='1' to enable profiling.
         """
-        # Check if this is a Nvidia GPU
-        if not (torch.cuda.is_available() and torch.version.cuda is not None):
+        # Check if this is a Nvidia or AMD GPU
+        if not (torch.cuda.is_available() and (torch.version.cuda is not None or torch.version.hip is not None)):
             ok = super()._benchmark()
             self._post_run_model_log()
             return ok
diff --git a/superbench/runner/runner.py b/superbench/runner/runner.py
index a5ac13cbb..c0b2345e5 100644
--- a/superbench/runner/runner.py
+++ b/superbench/runner/runner.py
@@ -7,6 +7,7 @@
 import sys
 import json
 import random
+import shlex
 import signal
 from pathlib import Path
 from pprint import pformat
@@ -135,12 +136,25 @@ def __get_mode_command(self, benchmark_name, mode, timeout=None):
         enable_nsys = os.environ.get('SB_ENABLE_NSYS', '') == '1'
         trace_dir = os.environ.get('SB_NSYS_TRACE_DIR', self._sb_output_dir)
 
+        # Enable rocprofv2 profiling based on environment variable
+        enable_rocprof = os.environ.get('SB_ENABLE_ROCPROF', '') == '1'
+        rocprof_trace_dir = os.environ.get('SB_ROCPROF_TRACE_DIR', self._sb_output_dir)
+
         mode_command = exec_command
         if mode.name == 'local':
-            trace_command = (
-                f'nsys profile --output {trace_dir}/{benchmark_name}_{mode.proc_rank}_traces '
-                f'--backtrace none --sample none --force-overwrite true --cpuctxsw none --trace cuda,nvtx '
-            ) if enable_nsys and mode.proc_rank == 0 else ''
+            trace_command = ''
+            if enable_nsys and mode.proc_rank == 0:
+                trace_output = shlex.quote(f'{trace_dir}/{benchmark_name}_{mode.proc_rank}_traces')
+                trace_command = (
+                    f'nsys profile --output {trace_output} '
+                    f'--backtrace none --sample none --force-overwrite true --cpuctxsw none --trace cuda,nvtx '
+                )
+            elif enable_rocprof and mode.proc_rank == 0:
+                trace_output = shlex.quote(f'{rocprof_trace_dir}/{benchmark_name}_{mode.proc_rank}_traces')
+                trace_command = (
+                    f'rocprofv2 --hip-trace --kernel-trace --plugin json '
+                    f'-d {trace_output} '
+                )
             # Build the command parts, only including trace if it's not empty
             command_parts = []
             prefix = mode.prefix.format(proc_rank=mode.proc_rank, proc_num=mode.proc_num)
@@ -159,23 +173,41 @@ def __get_mode_command(self, benchmark_name, mode, timeout=None):
                 '--nnodes=$NNODES --node_rank=$NODE_RANK --master_addr=$MASTER_ADDR --master_port=$MASTER_PORT '
             )
 
-            nsys_prefix = (
-                f'nsys profile --output {trace_dir}/{benchmark_name}_traces '
-                f'--backtrace none --sample none --force-overwrite true --cpuctxsw none --trace cuda,nvtx '
-            ) if enable_nsys else ''
+            trace_prefix = ''
+            if enable_nsys:
+                trace_output = shlex.quote(f'{trace_dir}/{benchmark_name}_traces')
+                trace_prefix = (
+                    f'nsys profile --output {trace_output} '
+                    f'--backtrace none --sample none --force-overwrite true --cpuctxsw none --trace cuda,nvtx '
+                )
+            elif enable_rocprof:
+                trace_output = shlex.quote(f'{rocprof_trace_dir}/{benchmark_name}_traces')
+                trace_prefix = (
+                    f'rocprofv2 --hip-trace --kernel-trace --plugin json '
+                    f'-d {trace_output} '
+                )
 
             mode_command = (
-                f'{nsys_prefix}'
+                f'{trace_prefix}'
                 f'torchrun'
                 f' --no_python --nproc_per_node={mode.proc_num} {torch_dist_params}{exec_command}'
                 f' superbench.benchmarks.{benchmark_name}.parameters.distributed_impl=ddp'
                 f' superbench.benchmarks.{benchmark_name}.parameters.distributed_backend=nccl'
             )
         elif mode.name == 'mpi':
-            trace_command = (
-                f'nsys profile --output {trace_dir}/{benchmark_name}_{mode.proc_rank}_traces '
-                f'--backtrace none --sample none --force-overwrite true --cpuctxsw none --trace cuda,nvtx '
-            ) if enable_nsys else ''
+            trace_command = ''
+            if enable_nsys:
+                trace_output = shlex.quote(f'{trace_dir}/{benchmark_name}_{mode.proc_rank}_traces')
+                trace_command = (
+                    f'nsys profile --output {trace_output} '
+                    f'--backtrace none --sample none --force-overwrite true --cpuctxsw none --trace cuda,nvtx '
+                )
+            elif enable_rocprof:
+                trace_output = shlex.quote(f'{rocprof_trace_dir}/{benchmark_name}_{mode.proc_rank}_traces')
+                trace_command = (
+                    f'rocprofv2 --hip-trace --kernel-trace --plugin json '
+                    f'-d {trace_output} '
+                )
             mode_command = (
                 '{trace} '
                 'mpirun '    # use default OpenMPI in image