pull down to refresh

llama.cpp was in need of some impactful Cuda tuning until yesterday: https://github.com/ggml-org/llama.cpp/pull/16585 and probably still is.
This has been over the decade(s) the persistent issue with Cuda (complexity, cross-version instability/regressions) and why I'd principally love to get some AMD tech. But what I hear from friends working with Vulkan, it has its own problems.
My biggest issue is that Metal works almost out of the box, but the cost is steep...
I didn’t even know what Vulkan was until I installed the new Civilization 7. When you launch the game, it lets you pick between DirectX12 and Vulkan. Fun fact: the game ran like crap on my old graphics card, and it wouldn’t even start with DirectX12. But with Vulkan? It actually worked! Hahaha
reply
TIL DirectX still exists lol
reply