V. Closing

What this lets you do

So let's count our wins. We have an open-source driver which can:

• Pool bandwidth across multiple cables, and across both lanes of a single cable
• Implement libibverbs, we can support vLLM, RCCL, MPI- anything that takes a libibverbs HCA
• Demonstrate actual, real-world value for certain use-cases (finetuning, TP=2 inference)
• Mac support is minimal- we demonstrated our Soft-DMA transport works and integration with JACCL, simple compute- but no real workloads or reason for anyone to do this currently.

The three headline numbers, one chart each:

Raw interconnect — ~95 Gb/s bidi (~48 Gb/s per direction) at 1 MiB, 4-HCA aggregate, IOMMU off:

vLLM MiniMax-M2.7 (AWQ, ~230 B MoE / ~10 B active) TP=2 across both Strix boxes — a model that doesn't fit on a single Strix at all. RDMA wins ~30% over TCP-over-Thunderbolt at batch 1 and the gap narrows to ~5% by batch 8, as compute starts to amortize the interconnect cost:

Gemma 3 27B LoRA FSDP — 1359 s on Ethernet → 126 s on 4-HCA RDMA, a 10.8× speedup:

Why this matters for Hellas

The Hellas mission:

run whatever compute you want, wherever you want

More strictly, we want to bind useful computation to economic value without assuming the compute lives in a hyperscaler datacenter. That only works if ordinary people can bring ordinary hardware to the network and have it do non-trivial work.

This project does not solve that whole problem. It solves a smaller, annoying prerequisite: can cheap consumer boxes be wired together into something that behaves enough like a real AI cluster for existing software to use it?

The answer seems to be yes, at least in the messy research-prototype sense.

Status

This is not a supported Hellas product yet. It is a proof-of-concept for one piece of the stack: making commodity home hardware useful as a multi-node AI machine.

What works today:

• Linux ↔ Linux RDMA over USB4 between Strix Halo boxes
• Standard libibverbs integration, enough for vLLM/RCCL and FSDP experiments
• Real workload wins where the network actually matters
• Partial Apple AD/FA57 interop, enough to prove the path exists

What does not work today:

• Catgrad integration
• Production support
• Safe unattended deployment

The claim here is narrow: if Hellas is going to make owner-operated compute useful, commodity machines need a way to pool memory and bandwidth locally. This experiment makes that look much less impossible than it did a few weeks ago.

Try it

• Code: https://github.com/hellas-ai/thunderbolt-ibverbs
• Built on Nix — nix develop from the project's flake.nix gives you the patched kernel, the out-of-tree module, perftest, and a vLLM/RCCL environment in one shot.
• Who should care: anyone running models on Strix Halo (or any AMD platform with USB4 ports), anyone building multi-box at-home inference, and kernel folks who care about the Thunderbolt subsystem.
• Discord: https://discord.gg/qHqZyuAa3u
• Find me on X: @0xBaltar