# DGXArley: cluster deployment and transport reports

An operator-maintained K3s/Ansible repository for a four-Spark cluster. It contains deployment notes and reported transport/inference comparisons. Select the exact model profile and experiment before reusing a number.

- Type: Performance report
- Evidence: source_reported; site reproduced: false
- Model family: Multiple
- Model: Multiple model profiles; see the selected report
- Original author: vroomfondel
- Original project: https://github.com/vroomfondel/dgxarley
- Pinned document: https://github.com/vroomfondel/dgxarley/tree/5f1f3762797eac0ea1ad43e403fa979e3e2c3934
- Source revision: 5f1f3762797eac0ea1ad43e403fa979e3e2c3934
- Source commit time (not experiment date): 2026-10-16T07:08:17Z
- Checked: 2026-10-17
- Experiment date: None
- Runtime topics: SGLang
- Hardware topics: spark
- Reported setup: {"system": "NVIDIA DGX Spark / K3s cluster", "node_count": 4, "parallelism": null, "kv_dtype": null, "context_limit_tokens": null, "experiment_date": null}
- Performance figures: attributed selections with per-point conditions; no cross-source ranking

## Conditions

Unknown is null. Recipe options are not measured benchmark settings.

```json
{
  "conditions": {
    "model_revision": null,
    "weight_hash": null,
    "runtime_execution_revision": null,
    "input_tokens": null,
    "output_tokens": null,
    "concurrency": null,
    "cache_state": null
  },
  "test_profiles": [],
  "recipe_parameters": null,
  "deployment_options": []
}
```

## Reading notes

- The repository contains multiple model profiles and parallelism choices.
- Transport, topology and concurrency belong to each reported experiment.
- Per-file licensing may differ from the repository’s main license declaration.

## Limits

- A default deployment profile is not necessarily the profile used in a benchmark paragraph.
- Prompt lengths, output policy, cache and model revision need verification in the underlying report.
- Do not compare these results directly with the three-node MiniMax configuration.

## License declaration

{"declared_spdx": "LGPL-3.0", "scope": "Repository declaration only; models, containers, datasets and per-file terms remain separate."}


## Editorial answer

This document describes NCCL all-reduce profiling for a Spark serving cluster. It recommends inspecting small messages as well as the large-message bandwidth plateau. The author discusses roughly 185–190 versus 92–95 Gbit/s under different port-spanning conditions, but those are transport observations without an identified workload row here. They are not language-model decode speeds.

By Run AI Here editorial · 2026-10-17

AI-assisted source reading and original editorial synthesis; no independent model execution.

## What is the report measuring?

This document describes NCCL all-reduce profiling for a Spark serving cluster. It recommends inspecting small messages as well as the large-message bandwidth plateau. The author discusses roughly 185–190 versus 92–95 Gbit/s under different port-spanning conditions, but those are transport observations without an identified workload row here. They are not language-model decode speeds.

Sources: https://github.com/vroomfondel/dgxarley/blob/5f1f3762797eac0ea1ad43e403fa979e3e2c3934/NCCL_BANDWIDTH_PROFILING.md#L12

## Why is a healthy headline link rate insufficient?

A standalone benchmark can inherit different interface, channel or RoCE settings from the serving container. Large messages can also hide the latency-bound region that tensor-parallel decode uses. Record the actual serving path and inspect its small-message behavior before attributing an inference slowdown to model kernels.

Sources: https://github.com/vroomfondel/dgxarley/blob/5f1f3762797eac0ea1ad43e403fa979e3e2c3934/NCCL_BANDWIDTH_PROFILING.md#L25

## How is the sweep defined?

The tool’s default range starts at 1 MB and ends at 1 GB, with a factor-two size step, five warmup and twenty timed iterations per size. The document asks operators to lower the floor to 8 KB for decode-oriented profiling. The ratio below 0.4 is an author-defined diagnostic flag, not an industry pass/fail threshold or a prediction of tokens per second.

Sources: https://github.com/vroomfondel/dgxarley/blob/5f1f3762797eac0ea1ad43e403fa979e3e2c3934/NCCL_BANDWIDTH_PROFILING.md#L49

## What belongs in a usable transport receipt?

Keep node count, per-node network device, MTU path, transport, process/container network settings, message size and both algorithm and bus bandwidth. Match those settings to the inference process before linking network results to generation performance. Maintenance and live-workload conflicts require the operator’s own planning; this site does not inspect or alter any cluster.

Sources: https://github.com/vroomfondel/dgxarley/blob/5f1f3762797eac0ea1ad43e403fa979e3e2c3934/NCCL_BANDWIDTH_PROFILING.md#L130

## Attributed configuration facts

```json
[
  {
    "name": "Decode-oriented message region",
    "value": "8 KB–1 MB",
    "classification": "publisher_declaration",
    "evidence_url": "https://github.com/vroomfondel/dgxarley/blob/5f1f3762797eac0ea1ad43e403fa979e3e2c3934/NCCL_BANDWIDTH_PROFILING.md#L33",
    "source_document_sha256": "293d119b6374d8073f040fe6dd277fe42f50f5f9baaadfdfa0d2120e65f12568"
  },
  {
    "name": "Default warmup / timed iterations",
    "value": "5 / 20",
    "classification": "publisher_declaration",
    "evidence_url": "https://github.com/vroomfondel/dgxarley/blob/5f1f3762797eac0ea1ad43e403fa979e3e2c3934/NCCL_BANDWIDTH_PROFILING.md#L52",
    "source_document_sha256": "293d119b6374d8073f040fe6dd277fe42f50f5f9baaadfdfa0d2120e65f12568"
  },
  {
    "name": "Author’s small-region warning threshold",
    "value": "ratio below 0.4× overall peak",
    "classification": "publisher_declaration",
    "evidence_url": "https://github.com/vroomfondel/dgxarley/blob/5f1f3762797eac0ea1ad43e403fa979e3e2c3934/NCCL_BANDWIDTH_PROFILING.md#L59",
    "source_document_sha256": "293d119b6374d8073f040fe6dd277fe42f50f5f9baaadfdfa0d2120e65f12568"
  }
]
```