AMD Acquires AI Software Company Nod.ai

[Marees]

AMD announced today that it has agreed to acquire open-source AI software company Nod.ai to improve its open source AI capabilities. Nod.ai, founded in 2013, has developed a software ecosystem of developer tools, libraries, and models that speed the deployment of AI solutions for AMD's silicon, including its Ryzen AI chips for consumer PCs and EPYC CPUs, Radeon GPUs, and Versal processors for the data center. Nod.ai was recently valued at $36.5 million, but AMD hasn't disclosed the financial terms of its acquisition.

Nod.ai's open source software meshes nicely with AMD's software software strategy, with its Shark compiler-based automation software speeding the deployment of AI models and runs on AMD's CPUs (Zen), GPUs (RDNA), and FPGAs (XDNA). Shark is a cross-platform Machine Learning Distribution that works in Windows, MacOS, and Linux. In fact, we recently used Nod.ai's Shark-based Stable Diffusion for testing in Windows with a broad range of AMD GPUs, but the software can scale all the way up to data center and hyperscaler applications as well.


https://www.tomshardware.com/news/a...ompany-nodai-to-bolster-open-source-portfolio

 

[Lakados]

Well AMD needed to do something and this is certainly something they could do.

 

[staknhalo]

Well AMD needed to do something and this is certainly something they could do.

Will they do anything with it or just let it sit and rot on github so to speak is the question from here on.

 

[Twisted Kidney]

M&A is one way to accelerate your software development.

 

[Lakados]

Nod doesn’t really use the tensor cores and has a pretty stable Vulkan path, so it’s agnostic but LLM’sthat make use of the tensor cores crush it last I checked.

This could be a big thing for game AI, dialog, decision trees, things that are less intensive than the systems where the Nvidia stuff shines.

 

[longblock454]

Wish they'd acquire at least one full time ROCm Dev.

 

[uOpt]

I would give it a nod if you could use random desktop/gaming AMD cards for this, like you can with NVidia.

 

[Lakados]

Wish they'd acquire at least one full time ROCm Dev.

Well I’m sure they will have the Nod guys roll ROCm into Nod.
ROCNod…

 

[ZeroBarrier]

Wish they'd acquire at least one full time ROCm Dev.

 

[staknhalo]

Well I’m sure they will have the Nod guys roll ROCm into Nod.
ROCNod…

History would suggest scrap ROCm and launch a new product

What was the one before ROCm again? OpenCL and one before I think?

 

[Lakados]

History would suggest scrap ROCm and launch a new product

What was the one before ROCm again? OpenCL and one before I think?

CL and ML are languages, Nod and ROC are environments. But you are right they need a new name, I suggest SOCm.

 

[staknhalo]

CL and ML are languages, Nod and ROC are environments. But you are right they need a new name, I suggest SOCm.

No but I mean for a time AMD themselves were just like - "just use OpenCL or something go ask someone else stop bothering me" - until they were like "we have ROCm now hey come use this!" and I think that's because they used to have something even older they eventually dropped but I can't recall ATM

 

[Lakados]

No but I mean for a time AMD themselves were just like - "just use OpenCL or something go ask someone else stop bothering me" - until they were like "we have ROCm now hey come use this!" and I think that's because they used to have something even older they eventually dropped but I can't recall ATM

But SOCm should follow ROCm especially for machine learning.

But now I can’t remember the name of that fuzzy logic library…
This is going to bother me, I know you could do it in Matlab, R, and J, but I’m pretty sure it wasn’t one of those you could do it in GL, but my brain says it was something else.