Mojo Roadmap Document

docs.modular.com

9 points·melodyogonna··8 comments

This would’ve generated immense interest one year ago. What happened ?

I guess Intel and NVidia started paying attention to making Python first class on their tooling, so if eventually Python gets the GPU friendly implementations with JIT compilers, what remains as added value from Modular?

https://www.nvidia.com/en-us/on-demand/session/gtcparis25-gp...

https://www.nvidia.com/en-us/on-demand/session/gtc25-s74639/

https://www.intel.com/content/www/us/en/developer/tools/onea...

https://github.com/intel/intel-xpu-backend-for-triton

Will Nvidia implementations work on Amd and Apple GPUs? The main thing Modular wants to solve is fragmentation, vendors create good SDKs for their hardware, this has always been the case.

Modular wants customers, that buy into their platform.

Maybe like middleware games engines and rendering frameworks that is what some folks will gladly pay for, but lets not pretend it is some philanthropic goal for the good of AI developers.

What you pretend or doesn't is not really as relevant as the value a truly cross-platform GPU platform unlocks.

At what price point?

You can already use PyTorch today.

Does it? I'm not sure you know what you're talking about.

But these things are beside the point, the document is about Mojo specifically - which is a programming language not a framework.

That depends on Modular infrastructure to be really useful.

Reply on news.ycombinator.com