[SUGGESTION]: Using 2nd pc for auto ortho while streaming the textures to first PC

[Swappea]

Hi,

I have been sitting here for past 1 hour while auto ortho is fetching the textures and doing it's stuff and I am stuck at loading screen of xplane unable to fly today.

I assume it's my CPU is which is slow being the cause of this.

This got me to thinking can we have a 2nd PC do this work and it would be on same network as main PC? I have a macbook lying around which has very fast CPU.

This is just a thought I had and it could be impossible, but something to think through?

My PC Specs are: Ryzen 5 5600 (6 core 12 threads), RTX 4080 super, 64 GB ram

[karlrado]

Interesting thoughts, maybe @ProgrammingDinosaur can consider.

My take on it:

• It would be a pretty niche feature as not everyone has a second high-powered machine available
• Most of the CPU work AO currently does requires a lot of memory/disk bandwidth, so transferring the data back and forth and network latency is going to be costly, likely to the point of not being worth it
• It would be better to ensure that the CPU resources on the primary machine are being used to the fullest. I know that recent versions have tried to increase parallelism (by moving to Python 3.14 and recoding to use more threads), but I'm not sure yet that the local CPU is being maxed out. I need to measure the latest version, but I'm thinking that if the max parallelism is being enabled, then memory/disk bandwidth may be becoming the limiting factors.
• I've wondered about using the GPU to do some of this work. A Vulkan compute shader could do some of it. JPEG decoding doesn't work well on the GPU, but perhaps mipmap scaling and compression could be done there. The challenges here include copying a lot more memory around in order to give it to and get it from the GPU, and stability due to needing to implement tricky GPU programming. It is horribly inefficient because it would be copying code to the GPU memory, back to the CPU memory, only to give it back to XPlane, which copies it back to the GPU memory. (Yes, I know that CPU/GPU shared(host-visible) buffers can be used, but it is still more copying). Probably not worth doing since we'd strongly prefer to use real data in a mipmap level rather than scaling it from an adjacent level, leaving only compression on the GPU, which probably isn't worth it.