Keras 3/TF vs. PyTorch – small model performance tests on a Nvidia 4060 TI

There are many PROs and CONs regarding the choice of a Machine Learning [ML] framework for private studies on a Linux Workstations. Two mainly used frameworks are PyTorch and a Keras/Tensorflow combination. One aspect for productive work with ML models certainly is performance. And as I personally do not have TPUs or other advanced chips available, but just a consumer Nvidia 4060 TI graphics card, performance and optimal GPU usage are of major interest – even for the training of relatively small models.

With this post I just want to point out that the question of performance advantages of some framework on a CUDA controlled graphics card can not be answered in a unique way. Even for small neural network [NN] models the performance may depend on a variety of relevant settings, on jit-/xla-compilation and the chosen precision level of your training or inference runs.

Continue reading

Leap 15.6, Nvidia driver 570 – resume from suspend to RAM not working / workaround

Hint: After some experiments and further Internet digging, this post was rewritten and supplemented on the 5th of March, 2025. Sorry for any inconvenience.
—–

Recently, I have upgraded Opensuse Leap to version 15.6 on 5 PC-systems – all with (different) Nvidia graphic cards. I use KDE/Plasma on all these systems.

My daily working system is equipped with a 4060 TI Nvidia card. Nvidia drivers of version 570.124.06-1 on this particular system came from the Nvidia CUDA repository for Opensuse system at

https://developer.download.nvidia.com/ compute/ cuda/ repos/ opensuse15/ x86_64.

I sadly must say that the named particular driver, but also the present Nvidia drivers of version 570.86.16 on other systems, are at least in their corporation with the Linux kernel (6.4.0) and other components of the present Leap 15.6, unreliable or even buggy (for KDE/Plasma):

The resume process from “Suspend to RAM” does not work reliably on any of the systems.

Continue reading

Leap 15.6 – upgrade from Leap 15.5 on laptop with Optimus architecture

The last 4 months I was primarily occupied with physics. I got a bit sloppy regarding upgrades of my Linux systems. An upgrade of an rather old laptop to Leap 15.6 was overdue. This laptop had an Optimus configuration: To display graphics one can use either the dedicated Nvidia card or a CPU-integrated Intel graphics or both via an “offload” option for certain applications.

General steps to perform the upgrade

I just list up some elementary steps for the upgrade of an Opensuse Leap system – without going into details or potential error handling:

Step 1: Make a backup of the present installation
You can, for example, create images of the partitions or LVM volumes that contain your Leap-installation and transfer them to an external disk. Details depend of course on whether and how you have distributed system files over partitions or (LVM) volumes. In the simple case of just one partition, you may simply boot a rescue system, mount an external disk to /mnt and then use the “dd”-command;

# dd status=progress if=/dev/YOUR_PARTITION of=/mnt/bup_leap155.img  bs=4M 

Step 2: Update the installed packages of the present Leap installation
Perform an update of (all) installed packages – if newer versions are available. Check that your system runs flawlessly afterwards.

Step 3: Change the addresses of repositories to use the ${releasever} variable
You can e.g. use YaST to change the release number in the definition of your repositories’ addresses to the variable ${releasever}. The name of the SLES repository may then look like “https://download.opensuse.org/update/leap/${releasever}/sle/”.

Continue reading