AMD's Brotli-G 1.0 Released With CPU & GPU Decompression

[phoronix]

Phoronix: AMD's Brotli-G 1.0 Released With CPU & GPU Decompression

Back in November 2022 AMD announced Brotli-G for GPU-accelerated Brotli compression. Brotli has proven very worthwhile for compressing web assets and other material while AMD's Brotli-G modifies the bitstream format to be more optimal for handling by GPUs rather than just relying on CPU (de)compression. Today Brotli-G 1.0 was finally released...

https://www.phoronix.com/news/AMD-Brotli-G-1.0-SDK

[pWe00Iri3e7Z9lHOX2Qx]

I get the whole Google / pastry name origin, but Broccoli-G would have been way more fun to say.

[sobrus]

Looking at the list of GPU compute milestones AMD has reached in past few years, with all these rocm updates, work to add more GPUs etc, one could think that it is now widely used and poiished platform.

But when it comes to real applications that can use GPU without crashing, there is ....ermm... darktable! yay!.. and it's an OpenCL app.

Thankfully it's opencl, because installing rocm-hip-sdk requires literally 20GB of free disk space (my whole Manjaro system is like 12GB so it's quite ridiculous) and it's so good that it only supports few cards (because other cards are not good enough for such perfection).

Meanwhile Darktable ran flawlessly on RX570, which would cause sort of trouble for rocm I guess.

The way GPGPU evolved over last 10 years is a joke, but it's good to know my gpu could be somehow used for brotli decompression and it's even being supported by GCC.

I wonder what was wrong with OpenCL that nobody wanted to use it. Because it was actually working?

[loligans]

Thankfully it's opencl, because installing rocm-hip-sdk requires literally 20GB of free disk space (my whole Manjaro system is like 12GB so it's quite ridiculous) and it's so good that it only supports few cards (because other cards are not good enough for such perfection).

I have a very similar issue to this. I hate that when I want to use a single program I need to install many gigs of packages just to run that program. In an effort to deal with this, I install these large applications inside docker containers. It's a bit of overhead but at least my system remains lean in that it only has the packages it needs to function and everything else is relegated to running in a docker container.

The thing is, what I do is a lot of effort to set up (even if I have a system for doing it faster), so I only really reserve it for big things like running WINE applications, Steam, or other large things like compilation.