Update.

NEWS.txt

@@ -10,15 +10,15 @@ This release extends the 0.9.0rc2 and announces the upcoming final release 0.9.
 
 Highlights (since 0.9.0rc2):
  - Graph clean up and faster compilation
- - Removed warp-synchronous programming
  - New Theano flag conv.assert_shape
- - Fix precision in cuDNN
  - Fix overflow in pooling
- - Test fixes
- - Some useless optimizations are now marked as unsafe
  - Warn if taking softmax over broadcastable dimension
  - Removed old files not used anymore
+ - Test fixes and crash fixes
 
+ - New GPU back-end:
+
+   - Removed warp-synchronous programming, to get good results with newer CUDA drivers
 
 A total of 5 people contributed to this release since 0.9.0rc2 and 122 since 0.8.0, see the lists below.
 

NEWS_DEV.txt

@@ -104,11 +104,10 @@ Others:
  - More stack trace in error message
  - Speed up cholesky grad
  - log(sum(exp(...))) now get stability optimized
- - Some useless optimizations are now marked as unsafe
 
 
 Other more detailed changes:
- - Faster compilation and import with old CUDA backend
+ - Added new Theano flag nvcc.cudafe to enable faster compilation and import with old CUDA back-end
  - Use of 64-bit indexing in sparse ops to allow matrix with more then 2\ :sup:`31`\ -1 elements.
  - Allow more then one output to be an destructive inplace
  - Add flag profiling.ignore_first_call, useful to profile the new gpu back-end