GPU out of memory?

Just Starting
1

Dear PyFR Developer,

I’ve downloaded the 3d case which is uploaded at example datasets.

When I ran using cuda backend, it gives an error message like this :

[andiraarif@localhost Benchmark Problem]$ pyfr run -b cuda -p cube_hex24.pyfrm config.ini

Traceback (most recent call last):
File "/usr/bin/pyfr", line 9, in <module>
load_entry_point('pyfr==1.4.0', 'console_scripts', 'pyfr')()
File "/usr/lib/python3.4/site-packages/pyfr-1.4.0-py3.4.egg/pyfr/scripts/main.py", line 109, in main
File "/usr/lib/python3.4/site-packages/pyfr-1.4.0-py3.4.egg/pyfr/scripts/main.py", line 230, in process_run
File "/usr/lib/python3.4/site-packages/pyfr-1.4.0-py3.4.egg/pyfr/scripts/main.py", line 214, in _process_common
File "/usr/lib/python3.4/site-packages/pyfr-1.4.0-py3.4.egg/pyfr/solvers/__init__.py", line 14, in get_solver
File "/usr/lib/python3.4/site-packages/pyfr-1.4.0-py3.4.egg/pyfr/integrators/__init__.py", line 26, in get_integrator
File "/usr/lib/python3.4/site-packages/pyfr-1.4.0-py3.4.egg/pyfr/integrators/controllers.py", line 15, in __init__
File "/usr/lib/python3.4/site-packages/pyfr-1.4.0-py3.4.egg/pyfr/integrators/steppers.py", line 9, in __init__
File "/usr/lib/python3.4/site-packages/pyfr-1.4.0-py3.4.egg/pyfr/integrators/base.py", line 41, in __init__
File "/usr/lib/python3.4/site-packages/pyfr-1.4.0-py3.4.egg/pyfr/solvers/base/system.py", line 52, in __init__
File "/usr/lib/python3.4/site-packages/pyfr-1.4.0-py3.4.egg/pyfr/solvers/base/system.py", line 110, in _load_int_inters
File "/usr/lib/python3.4/site-packages/pyfr-1.4.0-py3.4.egg/pyfr/solvers/navstokes/inters.py", line 13, in __init__
File "/usr/lib/python3.4/site-packages/pyfr-1.4.0-py3.4.egg/pyfr/solvers/baseadvecdiff/inters.py", line 12, in __init__
File "/usr/lib/python3.4/site-packages/pyfr-1.4.0-py3.4.egg/pyfr/solvers/baseadvec/inters.py", line 24, in __init__
File "/usr/lib/python3.4/site-packages/pyfr-1.4.0-py3.4.egg/pyfr/solvers/base/inters.py", line 56, in _const_mat
File "/usr/lib/python3.4/site-packages/pyfr-1.4.0-py3.4.egg/pyfr/backends/base/backend.py", line 17, in newfn
File "/usr/lib/python3.4/site-packages/pyfr-1.4.0-py3.4.egg/pyfr/backends/base/backend.py", line 117, in const_matrix
File "/usr/lib/python3.4/site-packages/pyfr-1.4.0-py3.4.egg/pyfr/backends/base/types.py", line 138, in __init__

File "/usr/lib/python3.4/site-packages/pyfr-1.4.0-py3.4.egg/pyfr/backends/base/types.py", line 62, in __init__
'metaclass' entry removed. If no kwds argument is passed in, this will
File "/usr/lib/python3.4/site-packages/pyfr-1.4.0-py3.4.egg/pyfr/backends/base/backend.py", line 57, in malloc
File "/usr/lib/python3.4/site-packages/pyfr-1.4.0-py3.4.egg/pyfr/backends/cuda/base.py", line 82, in _malloc_impl
pycuda._driver.MemoryError: cuMemAlloc failed: out of memory

What’s wrong with this calculation? What memory in computer is used to perform GPU computing using CUDA Backend ? Fyi I used NVIDIA GeForce GTX460 GDDR5 1GB and 16GB RAM

2

Hi Arif,

What’s wrong with this calculation?

It seems you are running out of (GPU) memory.

What memory in computer is used to perform GPU computing using CUDA Backend ?

GPU memory. In your case 1GB.

You can try lowering the order in the .ini fee and see if it launches.

Cheers

Peter

3

Thanks a lot Dr Peter E Vincent,
glad to get the answer from you. It works with second and third order.
Does it mean that cuda backend cannot solve the relatively heavy case with 1 GB GPU memory ?

May i ask you something else?
Can i run PyFR with multi-core in single processor ( in this case i use 1 CPU Intel i7 with 8 cores(4cores hyperthreading)) + GPU ( Nvidia GTX460 ) simultaneously?What configuration and command do i have to type?

I have read the user guide, and i’m still confused about running in this mode. Thank you :slight_smile:

4

Hi Arif,

Freddie has answered how to go about setting up a heterogeneous run using both CPU and GPU configurations in the PyFR mailing list. The thread with detailed instructions can be found at:

https://groups.google.com/forum/#!topic/pyfrmailinglist/3-WhWBeH2ok

Best Regards,

Zach