diff --git a/buzz/__version__.py b/buzz/__version__.py
index 848258f9e..3a0b6cbfb 100644
--- a/buzz/__version__.py
+++ b/buzz/__version__.py
@@ -1 +1 @@
-VERSION = "1.2.0"
+VERSION = "1.2.1"
diff --git a/buzz/transcriber/recording_transcriber.py b/buzz/transcriber/recording_transcriber.py
index 1654bf188..c9fba33a8 100644
--- a/buzz/transcriber/recording_transcriber.py
+++ b/buzz/transcriber/recording_transcriber.py
@@ -70,11 +70,14 @@ def start(self):
         model_path = self.model_path
         keep_samples = int(self.keep_sample_seconds * self.sample_rate)
 
+        force_cpu = os.getenv("BUZZ_FORCE_CPU", "false")
+        use_cuda = torch.cuda.is_available() and force_cpu == "false"
+
         if torch.cuda.is_available():
             logging.debug(f"CUDA version detected: {torch.version.cuda}")
 
         if self.transcription_options.model.model_type == ModelType.WHISPER:
-            device = "cuda" if torch.cuda.is_available() else "cpu"
+            device = "cuda" if use_cuda else "cpu"
             model = whisper.load_model(model_path, device=device)
         elif self.transcription_options.model.model_type == ModelType.WHISPER_CPP:
             model = WhisperCpp(model_path)
@@ -92,6 +95,9 @@ def start(self):
                 logging.debug("Unsupported CUDA version (<12), using CPU")
                 device = "cpu"
 
+            if force_cpu != "false":
+                device = "cpu"
+
             model = faster_whisper.WhisperModel(
                 model_size_or_path=model_path,
                 download_root=model_root_dir,
diff --git a/buzz/transcriber/whisper_file_transcriber.py b/buzz/transcriber/whisper_file_transcriber.py
index b602b0c91..2e04df41b 100644
--- a/buzz/transcriber/whisper_file_transcriber.py
+++ b/buzz/transcriber/whisper_file_transcriber.py
@@ -148,6 +148,7 @@ def transcribe_faster_whisper(cls, task: FileTranscriptionTask) -> List[Segment]
         model_root_dir = user_cache_dir("Buzz")
         model_root_dir = os.path.join(model_root_dir, "models")
         model_root_dir = os.getenv("BUZZ_MODEL_ROOT", model_root_dir)
+        force_cpu = os.getenv("BUZZ_FORCE_CPU", "false")
 
         device = "auto"
         if platform.system() == "Windows":
@@ -158,6 +159,9 @@ def transcribe_faster_whisper(cls, task: FileTranscriptionTask) -> List[Segment]
             logging.debug("Unsupported CUDA version (<12), using CPU")
             device = "cpu"
 
+        if force_cpu != "false":
+            device = "cpu"
+
         model = faster_whisper.WhisperModel(
             model_size_or_path=model_size_or_path,
             download_root=model_root_dir,
@@ -200,7 +204,10 @@ def transcribe_faster_whisper(cls, task: FileTranscriptionTask) -> List[Segment]
 
     @classmethod
     def transcribe_openai_whisper(cls, task: FileTranscriptionTask) -> List[Segment]:
-        device = "cuda" if torch.cuda.is_available() else "cpu"
+        force_cpu = os.getenv("BUZZ_FORCE_CPU", "false")
+        use_cuda = torch.cuda.is_available() and force_cpu == "false"
+
+        device = "cuda" if use_cuda else "cpu"
         model = whisper.load_model(task.model_path, device=device)
 
         if task.transcription_options.word_level_timings:
diff --git a/buzz/transformers_whisper.py b/buzz/transformers_whisper.py
index e63983909..30a4093fb 100644
--- a/buzz/transformers_whisper.py
+++ b/buzz/transformers_whisper.py
@@ -162,8 +162,10 @@ def transcribe(
         task: str,
         word_timestamps: bool = False,
     ):
-        device = "cuda" if torch.cuda.is_available() else "cpu"
-        torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
+        force_cpu = os.getenv("BUZZ_FORCE_CPU", "false")
+        use_cuda = torch.cuda.is_available() and force_cpu == "false"
+        device = "cuda" if use_cuda else "cpu"
+        torch_dtype = torch.float16 if use_cuda else torch.float32
 
         use_safetensors = True
         if os.path.exists(self.model_id):
diff --git a/docs/docs/preferences.md b/docs/docs/preferences.md
index fd5a86efe..c7328fa16 100644
--- a/docs/docs/preferences.md
+++ b/docs/docs/preferences.md
@@ -94,3 +94,5 @@ Defaults to [user_cache_dir](https://pypi.org/project/platformdirs/).
 **BUZZ_LOCALE** - Buzz UI locale to use. Defaults to one of supported system locales.
 
 **BUZZ_DOWNLOAD_COOKIEFILE** - Location of a [cookiefile](https://github.com/yt-dlp/yt-dlp/wiki/FAQ#how-do-i-pass-cookies-to-yt-dlp) to use for downloading private videos or as workaround for anti-bot protection.
+
+**BUZZ_FORCE_CPU** - Will force Buzz to use CPU and not GPU, useful for setups with older GPU if that is slower than GPU or GPU has issues. Example usage `BUZZ_FORCE_CPU=true`.
diff --git a/pyproject.toml b/pyproject.toml
index 791d2c412..95f15296e 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "buzz-captions"
-version = "1.2.0"
+version = "1.2.1"
 description = ""
 authors = ["Chidi Williams <williamschidi1@gmail.com>"]
 license = "MIT"