{"id":47113604,"title":"Show HN: OctoFlow v1.0.0 – GPU VM where the GPU runs autonomously, CPU is BIOS","points":1,"user":"mr_octopus","time":1771786500,"time_ago":"8 hours ago","type":"link","content":"

Three days ago I posted OctoFlow 0.83 here (GPU-native programming language, \n2.2 MB binary). The feedback was great. Since then I've pushed v1.0.0 with the \nthing I've actually been building toward: a GPU Virtual Machine.

The idea: the GPU is the computer, the CPU is the BIOS.

You boot a VM, program a dispatch chain of kernel instances, submit once with \nvkQueueSubmit, and everything — layer execution, inter-layer communication, \nself-regulation, compression, database queries — happens on the GPU without CPU \nround-trips. The CPU just provides I/O.

  let vm = vm_boot()\n  let prog = vm_program(vm, kernels, 4)\n  vm_write_register(vm, 0, 0, input)\n  vm_execute(prog)\n  let result = vm_read_register(vm, 3, 30)\n

\n4 VM instances, one submit, no CPU involvement between stages.

The memory model is 5 SSBOs: Registers (per-VM working memory), Metrics \n(regulator signals), Globals (shared mutable — KV cache, DB tables), Control \n(indirect dispatch params), Heap (immutable bulk data — quantized weights).

What makes it interesting:

- Homeostasis regulator: each VM instance has a kernel that monitors activation \n norms, memory pressure, throughput. The GPU self-regulates without asking the \n CPU.

- GPU self-programming: a kernel writes workgroup counts to the Control buffer, \n the next vkCmdDispatchIndirect reads them. The GPU decides its own workload.

- Compression as computation: Q4_K dequantization, delta encoding, dictionary \n lookup — these are just kernels in the dispatch chain, not a special subsystem. \n Adding a new codec = writing an emitter. No Rust changes.

- CPU polling: Metrics and Control are HOST_VISIBLE. CPU can poll GPU state and \n activate dormant VMs without rebuilding the command buffer. The GPU broadcasts \n needs, the CPU fulfills them.

The VM is workload-agnostic. Same architecture handles LLM inference, database \nqueries, physics sims, graph neural networks, DSP pipelines, and game AI. We've \nvalidated all six. The dispatch chain is the universal primitive.

What's new in v1.0.0 beyond GPU VM:\n- 247 stdlib modules (up from 51)\n- Native media codecs (PNG, JPEG, GIF, MP4/H.264 — no ffmpeg)\n- GUI toolkit with 15+ widgets\n- Terminal graphics (Kitty/Sixel)\n- 1,169 tests passing\n- Still 2.3 MB, still zero external dependencies

The zero-dep thing is real — zero Rust crates. The binary links against vulkan-1 \nand system libs, nothing else. cargo audit has nothing to audit.

Landing page: https://octoflow-lang.github.io/octoflow/\nGPU VM details: https://octoflow-lang.github.io/octoflow/gpu-vm.html\nGitHub: https://github.com/octoflow-lang/octoflow\nDownload: https://github.com/octoflow-lang/octoflow/releases/latest

I'm one developer. This is early. The GPU VM works and tests pass bit-exact, but \nthere's a lot of road ahead — real LLM inference at scale, multi-agent \norchestration, the full database engine. I'd love feedback from anyone who works \nwith GPU compute, Vulkan, or language design.","url":"item?id=47113604","comments":[],"comments_count":0}