diff --git a/tritonbench/utils/parser.py b/tritonbench/utils/parser.py
index 6efefb33..9e0666cb 100644
--- a/tritonbench/utils/parser.py
+++ b/tritonbench/utils/parser.py
@@ -223,15 +223,15 @@ def get_parser(args=None):
         type=str,
         help="Load input file from Tritonbench data JSON.",
     )
+    parser.add_argument(
+        "--logging-group",
+        type=str,
+        default=None,
+        help="Name of group for benchmarking.",
+    )
 
     if is_fbcode():
         parser.add_argument("--log-scuba", action="store_true", help="Log to scuba.")
-        parser.add_argument(
-            "--logging-group",
-            type=str,
-            default=None,
-            help="Override default name for logging in scuba.",
-        )
         parser.add_argument(
             "--production-shapes",
             action="store_true",
diff --git a/tritonbench/utils/triton_op.py b/tritonbench/utils/triton_op.py
index d60348e3..5ae2fe48 100644
--- a/tritonbench/utils/triton_op.py
+++ b/tritonbench/utils/triton_op.py
@@ -616,18 +616,6 @@ def _inner(self, *args, **kwargs):
     return decorator
 
 
-def register_benchmark_manually(
-    operator_name: str,
-    func_name: str,
-    baseline: bool = False,
-    enabled: bool = True,
-    label: Optional[str] = None,
-):
-    return register_benchmark(
-        operator_name, func_name, baseline, enabled, fwd_only=False, label=label
-    )
-
-
 def register_metric(
     # Metrics that only apply to non-baseline impls
     # E.g., accuracy, speedup
@@ -1097,8 +1085,35 @@ def get_example_inputs(self):
         except StopIteration:
             return None
 
-    def get_temp_path(self, path: Union[str, Path]) -> Path:
-        return Path(tempfile.gettempdir()) / "tritonbench" / self.name / Path(path)
+    def get_temp_path(
+        self,
+        fn_name: Optional[str] = None,
+    ) -> Path:
+        unix_user: Optional[str] = os.environ.get("USER", None)
+        logging_group: Optional[str] = self.logging_group
+        parts = [x for x in ["tritonbench", unix_user, logging_group] if x]
+        tritonbench_dir_name = "_".join(parts)
+        benchmark_name = self.benchmark_name
+        fn_part = f"{fn_name}_{self._input_id}" if fn_name else ""
+        out_part = Path(tempfile.gettempdir()) / tritonbench_dir_name / benchmark_name
+        return out_part / fn_part if fn_part else out_part
+
+    @property
+    def precision(self) -> str:
+        if self.tb_args.precision == "bypass" or self.tb_args.precision == "fp8":
+            return ""
+        return self.tb_args.precision
+
+    @property
+    def benchmark_name(self, default: bool = False) -> str:
+        if not default and self.tb_args.benchmark_name:
+            return self.tb_args.benchmark_name
+        parts = [x for x in [self.precision, self.name, self.mode.value] if x]
+        return "_".join(parts)
+
+    @property
+    def logging_group(self) -> Optional[str]:
+        return self.tb_args.logging_group
 
     def accuracy(self, fn: Callable, baseline_fn: Callable) -> bool:
         output = fn()
@@ -1331,7 +1346,7 @@ def _init_extra_metrics() -> Dict[str, Any]:
                     do_compile_kineto_trace_in_task,
                 )
 
-                kineto_trace_output_dir = self.get_temp_path("kineto_trace")
+                kineto_trace_output_dir = self.get_temp_path(fn_name)
                 kineto_trace_output_dir.mkdir(parents=True, exist_ok=True)
                 metrics.extra_metrics["_compile_time_kineto_trace_in_task"] = (
                     do_compile_kineto_trace_in_task(
@@ -1562,10 +1577,10 @@ def _get_op_task_args(
 
     def nsys_rep(self, input_id: int, fn_name: str) -> str:
         op_task_args = self._get_op_task_args(input_id, fn_name, "_nsys_rep_in_task")
-        nsys_output_dir = self.get_temp_path(f"nsys_traces/{fn_name}_{input_id}")
+        nsys_output_dir = self.get_temp_path(fn_name)
         nsys_output_dir.mkdir(parents=True, exist_ok=True)
         ext = ".nsys-rep"
-        nsys_output_file = nsys_output_dir.joinpath(f"nsys_output{ext}").resolve()
+        nsys_output_file = nsys_output_dir.joinpath(f"nsys_rep{ext}").resolve()
         nsys_trace_cmd = [
             "nsys",
             "profile",
@@ -1639,11 +1654,11 @@ def service_exists(service_name):
                 logger.warn(
                     "DCGM may not have been successfully disabled. Proceeding to collect NCU trace anyway..."
                 )
-        ncu_output_dir = self.get_temp_path(f"ncu_traces/{fn_name}_{input_id}")
+        ncu_output_dir = self.get_temp_path(fn_name)
         ncu_output_dir.mkdir(parents=True, exist_ok=True)
         ext = ".csv" if not replay else ".ncu-rep"
         ncu_output_file = ncu_output_dir.joinpath(
-            f"ncu_output{'_ir' if profile_ir else ''}{ext}"
+            f"ncu_rep{'_ir' if profile_ir else ''}{ext}"
         ).resolve()
         ncu_args = [
             "ncu",
@@ -1686,14 +1701,14 @@ def service_exists(service_name):
 
     def att_trace(self, input_id: int, fn_name: str) -> str:
         op_task_args = self._get_op_task_args(input_id, fn_name, "_ncu_trace_in_task")
-        att_output_dir = self.get_temp_path(f"att_traces/{fn_name}_{input_id}")
+        att_output_dir = self.get_temp_path(fn_name)
         att_trace_dir = launch_att(att_output_dir, op_task_args)
         return att_trace_dir
 
     def kineto_trace(self, input_id: int, fn: Callable) -> str:
         from tritonbench.components.kineto import do_bench_kineto
 
-        kineto_output_dir = self.get_temp_path(f"kineto_traces/{fn._name}_{input_id}")
+        kineto_output_dir = self.get_temp_path(fn._name)
         kineto_output_dir.mkdir(parents=True, exist_ok=True)
         return do_bench_kineto(
             fn=fn,
@@ -1834,7 +1849,7 @@ def run_and_capture(self, *args, **kwargs):
             fn()
 
         if len(compiled_kernels) > 0:
-            ir_dir = self.get_temp_path("ir")
+            ir_dir = self.get_temp_path(fn._name)
             ir_dir.mkdir(parents=True, exist_ok=True)
             logger.info(
                 "Writing %s Triton IRs to %s",