hey

just to start I don't have any experience in GPU programming.

is there a simple way to program the GPU, I know that there are thousands of cores in a GPU so can I create a program to tell an individual core to do a simple calculation without any complicated programming, so that I can have all the cores doing small and simple calculations to be used elsewhere.

and is there a way to do this without an API, I know that Nvidia GPUs are programmed in CUDA so will my code still work without CUDA?

I know that it is probably much harder to do this, this way, but this seems like how it would work.

anything will be much appreciated, thank you.

I have no real experience in GPU programming, I just read about it and did some experiments.

I think you may want to start with reading about the processor and memory architecture of GPUs in general and your system. I think that should have the answer to most of your questions. Similarly, most likely there is also documentation about why the API is what it is. I have been reading mostly about OpenCL, but that's mostly the same thing as CUDA afaik except you're not locking yourself in to a single manufacturer. I found the API quite reasonable and it looked simple enough to program at the time.

From what I can remember, GPUs are SIMD processors (but check that!!). SIMD means there is 1 program running at all processors (or some subset?? read the documentation of your system!!) but every processor has its own data. Roughly speaking, one algorithm being executed, each processors handles a different pixel. Obviously this is why these things are so freaking fast in 3D transformations and video. They run algorithms in a massively parallel way.

How did the OP earn the “author” tag?

the forum software has some bugs. sometimes flagging a spammer and unhiding thread/posts confuses it.
in this case, i wanted to keep the discussion open because of alberth's insightful comment.

reporting posts is still the best way to get a moderator's attention. if you wish to stop doing that, that's fine too.

that aside, let's keep the thread on-topic.