diff --git a/server/voice_changer/common/deviceManager/DeviceManager.py b/server/voice_changer/common/deviceManager/DeviceManager.py index 02754d7d2..e0cf86098 100644 --- a/server/voice_changer/common/deviceManager/DeviceManager.py +++ b/server/voice_changer/common/deviceManager/DeviceManager.py @@ -73,7 +73,7 @@ def reset_nvidia_clocks(self, device_id: int | None = None): if self.device.type != 'cuda': return device_id = self.device.index - logger.info(f"Resetting core/memory frequencies for cuda:{device_id}...") + logger.info(f"Resetting core/memory clocks for cuda:{device_id}...") subprocess.call(shlex.split(NVSMI_RESET_GPU_CLOCKS % self.device.index), stderr=subprocess.DEVNULL, stdout=subprocess.DEVNULL) subprocess.call(shlex.split(NVSMI_RESET_MEMORY_CLOCKS % self.device.index), stderr=subprocess.DEVNULL, stdout=subprocess.DEVNULL) @@ -83,7 +83,7 @@ def force_nvidia_clocks(self, device: torch.device): output = subprocess.check_output(shlex.split(NVSMI_QUERY_DEVICE_CLOCKS % device.index)) lines = output.decode("utf-8").split(os.linesep) gpu_clock, memory_clock = lines[0].split(', ') - logger.info(f"Using core/memory clocks: {gpu_clock} MHz / {memory_clock} MHz") + logger.info(f"Setting core/memory clocks for cuda:{device.index}: {gpu_clock} MHz / {memory_clock} MHz") subprocess.call(shlex.split(NVSMI_LOCK_GPU_CLOCKS % (device.index, int(gpu_clock))), stderr=subprocess.DEVNULL, stdout=subprocess.DEVNULL) subprocess.call(shlex.split(NVSMI_LOCK_MEMORY_CLOCKS % (device.index, int(memory_clock))), stderr=subprocess.DEVNULL, stdout=subprocess.DEVNULL)