ninlar.bsky.social
Software engineer from AZ. My head is usually in the cloud, but I also enjoy game development in GoDot and hardware design.
27 posts
71 followers
204 following
Getting Started
Conversation Starter
comment in response to
post
Going to be a long day
comment in response to
post
My son and I are scheming already. Thanks Kay, we have some great projects from your work.
comment in response to
post
Thank you, that makes sense. I slightly remember Android having a lot of kernel enhancements for power management and other things necessary for phones and tablets.
comment in response to
post
Wait, I thought in WSL 2 it did run the actual Linux kernel now. Is that not true? Have I been lied to.
comment in response to
post
I am very interested in continuing to try Copilot in #VSCode / #GoDot. I am curious as to how helpful it will be. This seems like a more challenging task for Copilot since this type of code has a lot more variation depending on scene building along with complex assets like 3D models and textures.
comment in response to
post
Yep! It took a couple of attempts. The first solution was for Godot3, so it was not compatible, but after a couple of attempts it was able to adjust the vector in the direction of the camera:
comment in response to
post
10)
comment in response to
post
11) So, thank you #Github and #Microsfot for this amazing tool, and thank you to all the open-source developers whose knowledge and experience made this possible.
comment in response to
post
10) I was able to get this miner running and fully utilizing this GPU in just over an hour.
I did have to have enough knowledge to troubleshoot the issues and identify the problem areas, but there is no way I could have implemented this myself so quickly.
comment in response to
post
9) Finally, I had everything up and running. I am absolutely shocked at how good this tool is. I have experience with C/C++, and I have looked at some CUDA samples in the past, but I know very little about it and the SHA256 hashing algorithm.
comment in response to
post
8) Then the application would never exit. It should have tried the ~4 billion possibilities and just exited, so I could tweak the extra nonce and try again, but it wouldn't exit. Of course, there was yet another overflow occurring:
comment in response to
post
7) I noticed that the application completed very quickly and hammered the Bitcoin network with block submissions. It was incorrectly the checking the hash against an incorrect target value:
comment in response to
post
6) I noticed it didn't query the device to see how many cores were available, but an additional prompt corrected this as well. Next, I noticed that the hash rate computation was incorrect. It chose Int32 to store this number which quickly overflowed each second.
comment in response to
post
5) The generated code was excellent; however, it did make several mistakes that needed to be corrected with several prompts. The first implementation did not try a different nonce on each invocation. So, the cores all computed the same result. This was easily corrected with an additional prompt.
comment in response to
post
4) A Python script passes the current block header to the CUDA app, and it was able to exhaustively try every nonce possible for that header in less than a second. So ~4 billion tries a second. Thats impressive, but how helpful was Git Hub Copilot?
comment in response to
post
3) I'm fully aware that mining on a GPU is pointless, due to all the ASICs and the current hash power of the network, but this was just a task to try out #GitHubCopilot.
comment in response to
post
2) I'm not great at CUDA C/C++ where you can write code that runs on GPUs instead of your CPU. So, I decided to give Copilot a try. The task was to create a Bitcoin miner where it would compute the SHA256 double hash of the block header on the CUDA cores in parallel on the GPU.
comment in response to
post
So great to be able to see your content again. Thanks David