BLOG

Engineering Blog

Deep dives into GPU observability, inference infrastructure, and accelerator tooling.

2026-08-10

plow is open source: a compiler that emits packet streams

Ahead-of-time compilation from checkpoint to packets — and the runtime that executes them

plow is now Apache-2.0. What the compiler emits, what a packet stream is, how one execution model covers NVIDIA and AMD across eight GPUs and several resident models, and the gated agent harness that brings up a new model.

Read article →
2026-08-02

The cold-start tax

Why vLLM recompiles itself at every boot — and how plow removes it

vLLM and SGLang cold starts are CPU-bound JIT work, not GPU compute. A phase-by-phase breakdown, measured boots and loaded footprints on GH200 and RTX PRO 6000, and how plow’s static compile path takes that tax off the critical path.

Read article →
2026-07-31

Kimi-K3 day-zero bring-up on 8× MI355X

A 93-layer MLA + KDA hybrid MoE, from “cannot load” to a correct answer

What day-zero support for a 93-layer MLA+KDA hybrid with 896 experts looks like from inside a compiled runtime — the interpreter, counter dataflow, the Lean-gated compiler, the wave/warp switch, tensor parallelism, per-XCD queues and the counter hierarchy they unlock, the correctness bugs a near-right model hides, an end-to-end accuracy check through the served endpoint, and a measured 28.9 ms/token decode taken apart against its bandwidth and protocol-chain floors.

Read article →
2026-07-30

Long-context decode on 8× AMD MI350X

Per-token latency vs context: plow grows 1.52× where vLLM grows up to 2.9×

From 1k to 128k context, vLLM’s per-token latency grows up to 2.9× — plow’s grows 1.52×. Where the fixed execution plan pays off, crossovers included.

Read article →
2026-07-27

Gemma-4 decode on RTX PRO 6000 Blackwell

A methodology-first, correctness-gated comparison with vLLM

Batch-1 decode on an RTX PRO 6000 Blackwell: plow fp8 beats vLLM fp8 on Gemma-4-26B MoE from 16k to 128k context — methodology, roofline analysis, and the not-yet-tuned paths included.

Read article →
2026-07-25

Plow: the compiled inference runtime

Why we moved optimization and orchestration out of the request path

The case for moving orchestration and optimization out of the request path — and a clear separation between measured gains, architectural properties, and open evaluation work.

Read article →
2026-06-04

Plow SMI: cross-vendor GPU observability and control

Production-grade GPU telemetry, monitoring, and power/clock control for heterogeneous AI infrastructure

A unified, vendor-agnostic platform — built in Rust — that brings first-class GPU telemetry, interactive monitoring, and control into a single, extensible toolkit.

Read article →