I have a small modern Fortran code with OpenACC that I want to test on my Jetson Nano Developer Kit (regular 4GB version). I’ve successfully compiled it with NVIDIA HPC SDK 20.11, SBSA version:
$ nvfortran -acc=gpu -Minfo=acc ddot.f90 -o ddot.x
dot_product:
42, Generating create(result,vecb(:),veca(:)) [if not already present]
45, Generating present(vecb(:),veca(:))
Generating copyin(n) [if not already present]
Generating Tesla code
46, !$acc loop gang, vector(128) ! blockidx%x threadidx%x
53, Generating present(result)
Scalar last value needed after loop for result at line 63,66,78,81,79
Accelerator serial kernel generated
Generating Tesla code
63, Generating present(result,vecb(:),veca(:))
Generating copyin(n) [if not already present]
Generating Tesla code
65, !$acc loop gang, vector(128) ! blockidx%x threadidx%x
Generating reduction(+:result)
73, Generating update self(result)
However, when I try to run it, I got an error message I’ve never seen before:
$ echo 100000000 | ./ddot.x
Input vector length n:
Using n = 100000000
Failing in Thread:1
call to cuModuleLoadDataEx returned error 209: No binary for GPU
Any help would be appreciated.
PS: Yeah, I know NVHPC 21.2 is available now, but haven’t got time to deploy it on the Jetson Nano
Yes, it is Maxwell. It looks like it’s still supported, even though Maxwell is deprecated. I also tried compiling using -gpu=cc50, which succeeded, but when I ran the executable, it gives the same error message.
While we didn’t explicitly disable the Maxwell code gen on ARM, we don’t formally support it and only test using Volta and newer. Though, what CUDA driver version are you using? We do have a minimum driver of 450.36 (CUDA 11).
That might be the problem. The JetPack version that is currently installed carries CUDA 10.2, though I tried using the CUDA 11.1 that is bundled with NVHPC.
This is the result for the CUDA example deviceQuery, under 1_Utilities.
$ ./deviceQuery
./deviceQuery Starting...
CUDA Device Query (Runtime API) version (CUDART static linking)
Detected 1 CUDA Capable device(s)
Device 0: "NVIDIA Tegra X1"
CUDA Driver Version / Runtime Version 10.2 / 10.2
CUDA Capability Major/Minor version number: 5.3
Total amount of global memory: 3964 MBytes (4156682240 bytes)
( 1) Multiprocessors, (128) CUDA Cores/MP: 128 CUDA Cores
GPU Max Clock rate: 922 MHz (0.92 GHz)
Memory Clock rate: 13 Mhz
Memory Bus Width: 64-bit
L2 Cache Size: 262144 bytes
Maximum Texture Dimension Size (x,y,z) 1D=(65536), 2D=(65536, 65536), 3D=(4096, 4096, 4096)
Maximum Layered 1D Texture Size, (num) layers 1D=(16384), 2048 layers
Maximum Layered 2D Texture Size, (num) layers 2D=(16384, 16384), 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: 32768
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: Yes
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: No
Supports Cooperative Kernel Launch: No
Supports MultiDevice Co-op Kernel Launch: No
Device PCI Domain ID / Bus ID / location ID: 0 / 0 / 0
Compute Mode:
< Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >
deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 10.2, CUDA Runtime Version = 10.2, NumDevs = 1
Result = PASS
Now the issue becomes how to upgrade CUDA on the Jetson Nano. If I use the CUDA Toolkit download links, it breaks the system. At least the “deb (network)” option does that.
Not sure this will work and I can’t test it, but try setting CUDA_HOME to a local CUDA 10.2 SDK install so the compiler will pick-up this CUDA toolset.