-
Verena met with GPU infrastructure guys and we can get Nvidia T4 GPUs. They
will probably be on the technical network to have faster inference for
control (Not for GPTs but we have stuff running on our section GPUs that are constantly running for control which is not great). There are also
A100 lying around that we might be able to get, they have 4 GPUs on
machine but that they are not fastly interconnected and cannot change
that, they have basic PCI connections and not Nvidia ones. We can have
one of these machines (All the A100 are 40Gb). In addition to that they
have ordered H100 GPUs for the end of the year. -
IN are trying to have a whole infrastructure for digital twins (basically
storing models), need to communicate with them about formatting of
storage. -
Anton needs to come up with a new model for his Hysteresis compensation,
something different from LSTMs that they can build on. Verena also needs to make some tests for the TT20 RL task (Can they do online training or do they need to do transfer). -
Meeting with Andrea Latina, he is writing simulation codes for electron cooling and RF track. Might need to add his work to the slides and ask about
him. -
Problem with Florian’s summer student, did not send mail to Chris in time, might not be too late.