GPU Device 0: "Graphics Device" on Pegasus does not support Unified Memory

It seems that the Device 0: “Graphics Device” (dGPU?) does not support Unified Memory, and I wonder to know if this is intentionally design or just a bug and will be fixed in the future. (e.g. cudaMallocManaged() call cannot work on “Device 0”)

I also test the sample codes provided by NVIDIA (e.g. CUDA, CUPTI), and there are some discrepancies between actual and the website info. (e.g. “Turing” Discrete GPUs (TU104 SXM2) - 130 TOPS connected to Xavier SoC over NVIDIA®NVLink™ (20 GB/s)) The sample code return “No Nvlink supported device found”, however, I supposed that CPU to dGPU is connected via NvLink (Please correct me if I am wrong), and the memory bandwidth results of Host<->Device, Device<->Device either on iGPU(>250 GB/s) or dGPU(>750 GB/s) cannot reach the theoretical values.

In addition, is there any APIs can directly transfer data from iGPU to dGPU and vice versa?
What is the best approach for memory management if there is no Unified Memory provided on dGPU?

The deviceQuery gives the following output:

Device 0: “Graphics Device”
CUDA Driver Version / Runtime Version 10.0 / 10.0
CUDA Capability Major/Minor version number: 7.5
Total amount of global memory: 7680 MBytes (8052998144 bytes)
(44) Multiprocessors, ( 64) CUDA Cores/MP: 2816 CUDA Cores
GPU Max Clock rate: 1500 MHz (1.50 GHz)
Memory Clock rate: 13 Mhz
Memory Bus Width: 256-bit
L2 Cache Size: 4194304 bytes
Maximum Texture Dimension Size (x,y,z) 1D=(131072), 2D=(131072, 65536), 3D=(16384, 16384, 16384)
Maximum Layered 1D Texture Size, (num) layers 1D=(32768), 2048 layers
Maximum Layered 2D Texture Size, (num) layers 2D=(32768, 32768), 2048 layers
Total amount of constant memory: 65536 bytes
Total amount of shared memory per block: 49152 bytes
Total number of registers available per block: 65536
Warp size: 32
Maximum number of threads per multiprocessor: 1024
Maximum number of threads per block: 1024
Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
Max dimension size of a grid size (x,y,z): (2147483647, 65535, 65535)
Maximum memory pitch: 2147483647 bytes
Texture alignment: 512 bytes
Concurrent copy and kernel execution: Yes with 3 copy engine(s)
Run time limit on kernels: No
Integrated GPU sharing Host Memory: No
Support host page-locked memory mapping: Yes
Alignment requirement for Surfaces: Yes
Device has ECC support: Disabled
Device supports Unified Addressing (UVA): Yes
Device supports Compute Preemption: Yes
Supports Cooperative Kernel Launch: Yes
Supports MultiDevice Co-op Kernel Launch: Yes
Device PCI Domain ID / Bus ID / location ID: 1 / 1 / 0
Compute Mode:
< Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >

Device 1: “Xavier”
CUDA Driver Version / Runtime Version 10.0 / 10.0
CUDA Capability Major/Minor version number: 7.2
Total amount of global memory: 24729 MBytes (25929990144 bytes)
( 8) Multiprocessors, ( 64) CUDA Cores/MP: 512 CUDA Cores
GPU Max Clock rate: 1109 MHz (1.11 GHz)
Memory Clock rate: 1109 Mhz
Memory Bus Width: 256-bit
L2 Cache Size: 524288 bytes
Maximum Texture Dimension Size (x,y,z) 1D=(131072), 2D=(131072, 65536), 3D=(16384, 16384, 16384)
Maximum Layered 1D Texture Size, (num) layers 1D=(32768), 2048 layers
Maximum Layered 2D Texture Size, (num) layers 2D=(32768, 32768), 2048 layers
Total amount of constant memory: 65536 bytes
Total amount of shared memory per block: 49152 bytes
Total number of registers available per block: 65536
Warp size: 32
Maximum number of threads per multiprocessor: 2048
Maximum number of threads per block: 1024
Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
Max dimension size of a grid size (x,y,z): (2147483647, 65535, 65535)
Maximum memory pitch: 2147483647 bytes
Texture alignment: 512 bytes
Concurrent copy and kernel execution: Yes with 1 copy engine(s)
Run time limit on kernels: No
Integrated GPU sharing Host Memory: Yes
Support host page-locked memory mapping: Yes
Alignment requirement for Surfaces: Yes
Device has ECC support: Disabled
Device supports Unified Addressing (UVA): Yes
Device supports Compute Preemption: Yes
Supports Cooperative Kernel Launch: Yes
Supports MultiDevice Co-op Kernel Launch: Yes
Device PCI Domain ID / Bus ID / location ID: 0 / 0 / 0
Compute Mode:
< Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >

Peer access from Graphics Device (GPU0) -> Xavier (GPU1) : No
Peer access from Xavier (GPU1) -> Graphics Device (GPU0) : No

deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 10.0, CUDA Runtime Version = 10.0, NumDevs = 2
Result = PASS

Dear 454615,
Yes. Unified memory is not supported on dGPU.
For nvlink isssue, may I know which sample have tested?
If you are asking to transfer data from iGPU to dGPU, you may have to bring the data onto CPU using cudamemcpy calls in between. However, you can choose EGLStream to transfer frames from iGPU to dGPU to avoid unneccessary memcpy operations. You can choose pinned memory as unified memory is not available. Please check https://docs.nvidia.com/cuda/cuda-for-tegra-appnote/index.html#memory-selection