today's leftovers
-
Fedora Community Blog: Creative Freedom Art Challenge Info & Guidelines
We welcome you (yes, you!) to join in an Art Challenge hosted by the Creative Freedom Summit and Fedora Design Team!
-
Fedora Infrastructure Status: koji database upgrade
We will be moving the koji buildsystem database (and the virthost it runs on) to RHEL9 and postgresql 15 (from RHEL8 and postgresql 12). This outage will happen while the outage of s390x builders is occuring to consolidate outages. During the outage window koji will be unavailable and builds will …
-
Fedora Infrastructure Status: wiki upgrade outage
We will be upgrading our wiki and it's database server along with the virthosts they are hosted on.
-
Tomeu Vizoso: Etnaviv NPU update 1: Planning for performance
As I wrote in the last update, my OpenCL branch is able to correctly run MobileNet v1 with the GPU delegate in TensorFlow-Lite, albeit much slower than with VeriSilicon's proprietary stack.
In the weeks that passed I have been investigating the performance difference, understanding better how the HW works and what could the explanation be. Inference with Etnaviv took 1200 ms, while the proprietary stack did the same in less than 10 ms (120x faster!).
When trying to understand the big performance difference I discovered that the existing reverse engineering tools that I had been using to understand how to run OpenCL workloads weren't working. They detected a single OpenCL kernel at the end of the execution, and there was no way that single kernel could be executing the whole network.