[concurrentsquar]

My team can't even evaluate Ball (as a tool for physical simulation of spherical cow-like objects): https://github.com/nate-parrott/ball/issues/9

Currently, we are using Unreal Engine 5 to do our hundreds of architectural physics simulations - the major issue is that UE5 is very slow on *the EC2 instance* (we only have one 2048 core EC2 instance shared between the entire office; we used to use Vercel and Cloudflare but we had to sell our homes to suddenly subscribe to Cloudflare Enterprise (the CF sales guy told us that we would not be allowed to run a CF Worker for more than 30 days without it, even though we had a CF worker run for 37 years, and many of our CF workers have been running before the creation of CF (nobody knows why)) and a giant spike in our Vercel Cuda Function Invocations (for GPGPU compute on the Edge, allowing architects to view the collapse of their buildings with only ~53 ms of latency (compared to ~53 ms without Next.js))). Ball seems much faster (it can run on a Macbook Air), potentially allowing us to save at least several tens of millions of dollars per year on AWS costs.

[Bluestein]

All common issues! Particularly the cost spikes without a reasonable explanation. Also, the lack of NT support the issue you point to shows could be a problem. Ball runs fine on "M" Apple chips, however :)

[nxobject]

It's embarrassing that we don't have NT support already. I have many users of NT on Alpha and MIPS who need Ball for critical services. Here's a convenient patch with a bunch of garbled test cases. Merge that in.

(I'm totally not a state actor looking to socially engineer you to hide an exploit, by the way.)

[Bluestein]

Of course. Were you state, there's a Chinese CDN we can totally recommend to cut latency on this.-

(Totally :)