c++ - About pinned memory in CUDA, is there an upper limit on it? -


this question has answer here:

i wrote application benefit pinned memory, , best performance can achieved when there large pinned memory.

we ordered server, consisting of 768gb memory , plan allocate ~500gb memory pinned, there problem that, since have never allocated pinned memory before, far on our servers smaller memory installed, dont see problem of allocating large chunk of memory page-locked/pinned.

the targetting os of application both windows server 2008 r2 , rhel6+, cuda version being 6.0 rc or 5.5, gpus tesla k20x, btw.

there no maximum pinned memory limit in cuda. determined amount of main memory machine have, memory required other process(oss , applications) etc. sometimes, os puts limit on maximum amount of page-locked memory process can request has soft limit , hard limit. can change both.


Comments

Popular posts from this blog

Android layout hidden on keyboard show -

google app engine - 403 Forbidden POST - Flask WTForms -

c - Why would PK11_GenerateRandom() return an error -8023? -