diff --git a/cmd/nvidia-ctk/cdi/generate/generate.go b/cmd/nvidia-ctk/cdi/generate/generate.go
index 9edd8598f..d112e51e9 100644
--- a/cmd/nvidia-ctk/cdi/generate/generate.go
+++ b/cmd/nvidia-ctk/cdi/generate/generate.go
@@ -73,6 +73,8 @@ type options struct {
 		ignorePatterns []string
 	}
 
+	noAllDevice bool
+
 	// the following are used for dependency injection during spec generation.
 	nvmllib nvml.Interface
 }
@@ -232,6 +234,12 @@ func (m command) build() *cli.Command {
 				Destination: &opts.featureFlags,
 				Sources:     cli.EnvVars("NVIDIA_CTK_CDI_GENERATE_FEATURE_FLAGS"),
 			},
+			&cli.BoolFlag{
+				Name:        "no-all-device",
+				Usage:       "Don't generate an `all` device for the resultant spec",
+				Destination: &opts.noAllDevice,
+				Sources:     cli.EnvVars("NVIDIA_CTK_CDI_GENERATE_NO_ALL_DEVICE"),
+			},
 		},
 	}
 
@@ -387,13 +395,18 @@ func (m command) generateSpecs(opts *options) ([]generatedSpecs, error) {
 		spec.WithVendor(opts.vendor),
 		spec.WithEdits(*commonEdits.ContainerEdits),
 		spec.WithFormat(opts.format),
-		spec.WithMergedDeviceOptions(
-			transform.WithName(allDeviceName),
-			transform.WithSkipIfExists(true),
-		),
 		spec.WithPermissions(0644),
 	}
 
+	if !opts.noAllDevice {
+		commonSpecOptions = append(commonSpecOptions,
+			spec.WithMergedDeviceOptions(
+				transform.WithName(allDeviceName),
+				transform.WithSkipIfExists(true),
+			),
+		)
+	}
+
 	fullSpec, err := spec.New(
 		append(commonSpecOptions,
 			spec.WithClass(opts.class),