Jupyter-Cloud: problems with TensorFlow in the current version

Message-Id: 202201281400
Affected: Jupyter-Cloud users
Impact: TensorFlow not usable

Since the end of last year the service Jupyter-Cloud has been extended to new servers to
increase capacity. Attempts to use the library TensorFlow on these servers results in an
error currently.
The newer servers lack the CPU instruction set SSE4.1 which in newer versions of
TensorFlow is a hard requirement, but the AMD EPYC CPUs lack. As a result attempting to
import the library in Python results in an error.
Not all servers of the service run on the affected CPUs. As a result not all users are
affected by this and consequently the library hasn’t been removed from the service. But
presently TensorFlow cannot be reliably expected to work in the service.

Solutions for the problems are under investigation.

Alternatively the HPC service of GWDG can be used as well as creating individual VMs in
GWDG Cloud.
But all effort is targeted towards enabling the use of TensorfFow in Jupyter-Cloud again.