diff --git a/GPy/util/gpu_init.py b/GPy/util/gpu_init.py
index 845d38a1..03d07d77 100644
--- a/GPy/util/gpu_init.py
+++ b/GPy/util/gpu_init.py
@@ -17,7 +17,7 @@ except:
     pass
 
 try:
-    if MPI_enabled and MPI.COMM_WORLD.size>1:
+    if MPI_enabled: #and MPI.COMM_WORLD.size>1:
         from .parallel import get_id_within_node
         gpuid = get_id_within_node()
         import pycuda.driver
diff --git a/GPy/util/linalg_gpu.py b/GPy/util/linalg_gpu.py
index 1b9b0594..d969d14f 100644
--- a/GPy/util/linalg_gpu.py
+++ b/GPy/util/linalg_gpu.py
@@ -12,9 +12,6 @@ from ..util import gpu_init
 try:
     from pycuda.reduction import ReductionKernel
     from pycuda.elementwise import ElementwiseKernel
-    import scikits.cuda.linalg as culinalg
-    from scikits.cuda import cublas
-    from scikits.cuda.cula import culaExceptions
     
     # log|A| for A is a low triangle matrix
     # logDiagSum(A, A.shape[0]+1)
@@ -64,6 +61,13 @@ try:
 except:
     pass
 
+try:
+    import scikits.cuda.linalg as culinalg
+    from scikits.cuda import cublas
+    from scikits.cuda.cula import culaExceptions
+except:
+    pass
+
 def jitchol(A, L, cublas_handle, maxtries=5):
     try:
         cublas.cublasDcopy(cublas_handle, A.size, A.gpudata, 1, L.gpudata, 1)