Calling the garbage collector explicitly is very expensive and gets progressively slower the longer the program runs. Need to find another way to deal with running out of GPU memory due to reference cycles in PyCUDA.

1 files changed, 1 insertions, 1 deletions

diff --git a/chroma/gpu/photon.py b/chroma/gpu/photon.py
index 2a9a1ff..739c922 100644
--- a/chroma/gpu/photon.py
+++ b/chroma/gpu/photon.py
@@ -145,7 +145,7 @@ class GPUPhotons(object):
         del self.flags
         del self.last_hit_triangles
         # Free up GPU memory quickly if now available
-        gc.collect()
+        #gc.collect()
 
 class GPUPhotonsSlice(GPUPhotons):
     '''A `slice`-like view of a subrange of another GPU photons array.