i got a problem and do not know if my system does not support the new architecture 6.x additional Unified Memory feautures (Especially PageableMemoryAccess and ConcurrentManaged Acess).
My device properties are:
Number of Devices: 1
Device Name: GTX1080
Managed Memory Property CC3.0(prop.managedMemory) : 1
PageableMemoryAccess: 0
ConcurrentManagedAccess: 0
UnifiedAddressing: 1
My system falls back to ZeroCopyMemory with Cuda 8.0 for cudaMallocManaged memory allocation, and i only use a single GPU (Only one Inside).
I ported one of my project to Linux and it works.
My device properties in Linux are:
Number of Devices: 1
Device Name: GTX1080
Managed Memory Property CC3.0(prop.managedMemory) : 1
PageableMemoryAccess: 0
ConcurrentManagedAccess: 1
UnifiedAddressing: 1
The difference is only OS, GPU driver and compiler.
Is the new PageableMemoryAccess and ConcurrentManagedAccess not supported under Window7 64bit with VS2013 or VS2015? (Only supoported under Linux)?
I would prefer to fix the problem, otherwise I have to install on two systems additional Linux and to configure an additional 4 systems and to port additional 3 projects.
I am having the same problem. I have a Titan Xp and I see ConcurrentManagedAccess: 0 in the deviceProp even though it has support for Unified Addressing and ManagedMemory. I am running on Windows 10 and I generated code for compute_60,sm_60.
The symptom I am seeing after allocating memory using cudaMallocManaged is that I am getting access violations when I try to access memory from the device.
it doesn’t look like the same problem. I don’t get access violations.
Acess vialotions only occure in my case when i run several kenrels started by diffrent CPU Threads which use the same GPU Array allocaded by cudaMallocManaged.
In my case a fallback to system memory or zero copy memory occurs.
ConcurrentManagedAccess: 0 is a BUG of NVIDIA on Windows Systems witch occurs with PASCAL architecture.
I know this since a few days, but could not write it here because i was on vacation without internet connection.
For details see the comments of: https://devblogs.nvidia.com/parallelforall/unified-memory-cuda-beginners/
where Mark Harris from NVIDIA confirms the Bug. It should be corrected with CUDA 9. He also tells that it should be communicated to Microsoft to help the caus. But i didn’t found a suitable Microsoft Bug Report Page till now.