GpuBatchedDot: tweak default stream_threshold

theano/sandbox/cuda/blas.py

@@ -18,7 +18,7 @@ from theano.tensor import as_tensor_variable
 class GpuBatchedDot(GpuOp):
     __props__ = ("stream_threshold",)
 
-    def __init__(self, stream_threshold=128):
+    def __init__(self, stream_threshold=650):
         self.stream_threshold = stream_threshold
 
     def make_node(self, inp1, inp2):