GPU Accelerated Computing on Cross-Vendor Graphics Cards with Vulkan Kompute - Alejandro Saucedo

[@LiamDennehy]

So glad I found this - learned more about the architecture of Vulkan in a comoute-focused demo than anything I've seen from general-purpose developers. Thanks!

[@wolfgangsanyer3544]

This is a great talk, and a very good introduction to Vulkan too. He does a great job giving, as he describes it, an "intuition" for what's going on in Vulkan

[@LiamDennehy]

Agreed - I barely comprehend C(++), had some intros to Vulkan (mostly from C++ code), but @19:38 and @12:09 crystalised this wonderfully.

[@rudolflovrencic]

great talk!

[@qm3ster]

Where can I find the slides? UwU (Not in the linked repo, that's for sure!)

[@judgeomega]

ONLY 2000 lines of code to print hello world:)

[@foadsf]

we already had a vendor neutral and heterogeneous API: OpenCL. we let it fail, so will Vulkan.

[@SoftBreadSoft]

Comparing apples to oranges here. Vulkan doesnt support generic pointers and cant create a "proper" compute kernel. OpenCL doesnt support a direct gpu shader pipeline and can't create graphics without massive overhead. OpenCLs death is unrelated to Vulkan, but related to HIP/ROC/ROCm being close to catching up with Cuda.