cancel
Showing results for 
Search instead for 
Did you mean: 

AI Discussions

user_user
Journeyman III

Support XDNA2

Hello forum. if I have the correct information, then you have had attempts to develop ZLUDA. I have an opinion that there is no need to repeat NVIDIA's mistakes, the main problem of which was that they created their nvcc compiler, which does not provide compatible code with the Linux kernel. For your product to be successful, you need to simplify the development of code for processors with XDNA2. to do this, use the GCC compiler. we need to be able to use XDNA2 in the Linux kernel as easily as possible without adapting the Linux kernel itself to develop for XDMA2.
0 Likes
1 Reply
user_user
Journeyman III

You do not need to spend a lot of money on software developers for your products. Just give a good tool so that people from the communities of novice programmers can independently use your NPU, then your product will become interesting and more popular. They know better what they need.

 

 

 I ask myself the question, what should make me buy new products with an NPU if I can't use this NPU? The only operating system we know that legally allows us to change the code of its kernel is LINUX distibutes, but there are many difficulties that are better shown by the example of CUDA and the nvcc compiler, which does not generate code with the possibility of using this code in the Linux kernel. we want to develop this field with artificial intelligence, we want to combine artificial intelligence with the Linux kernel, and we want to do it ourselves. then we will be interested in your products. to do this, we need a tool like GCC and the ability to run NPU code from Linux kernel modules. but the Linux kernel developers are against it. they do not allow you to run custom functions from the kernel module. this greatly slows down development and reduces interest in your products. There's no point in buying something you can't use.