> Even if there was some way to tell the GC about that 'other' > memory, the (incremental) GC will see the memory pressure after > the first 10 GB allocation, but it could never catch up at that > pace or that granularity relative to the total amount of memory. Pardon my ignorance, but I thought (i.e. just assumed) that the GC looks at bytes allocated since the last GC step, when deciding how much to collect in the current step, so if 10G were allocated since the last step, the current step should not stop until either 10G is freed or all dead objects are freed. Your answer indicates that it's more complicated than that. I'm asking because I always wondered why there's no size arg to ffi.gc() -- would that not help much then?