Six GB10s, three models, zero switches
A short tour of the JMNI Labs fleet — and the publishing discipline behind it.
The lab runs six GB10-class systems: NVIDIA DGX Spark and ASUS GX10 machines. Each one is a full Blackwell-generation GPU with 128 GB of unified memory — a datacenter chip that fits on a desk. That's the whole point of this hardware class: small enough to own, big enough to be interesting.
Interesting, though, is not the same as easy. A pair of these systems can serve genuinely frontier open-weight models — if you can get the pieces to cooperate. That "if" is where most people stall, and it's where we spend our time.
The ring
The trick we care most about: instead of putting a switch in the model-traffic path, the nodes are cabled directly to each other over their ConnectX-7 links in a ring, and collective traffic is forwarded through the NICs' own hardware. In plain terms — every node can talk to every other at high speed, with no network box in the middle. Fewer parts, less to configure, less to fail.
The GB10 community has done remarkable work making these stacks real. Our job is to operationalize them: take what works, package it, document it, and put numbers next to it.
What runs here
On top of that fabric, among others: GLM-5.3-Flash in NVFP4 across a Spark pair; Qwen3.8-Flash-Next on single Sparks and pairs, including a hybrid checkpoint we published ourselves; and DeepSeek-V4.1-Flash across a four-node switchless ring. Each one leaves the lab as a deployment kit with its measured results attached.
Publish or it didn't happen
An operations recipe without evidence is marketing. So every configuration that leaves this lab comes with numbers and a build log, in public — if it doesn't hold up in the open, it doesn't ship. That rule is basically the whole company.
More notes soon.