Resources
White papers and data briefs.
White papers
Five to ten pages each on one problem in AI network design, newest first.
Chakra Workloads Bring AI Network Simulation to Life
The platform runs both captured and synthetic Chakra workloads, the MLCommons format for describing how AI training and inference jobs use the network.
October 2025 · 6 pages · PDF, 2.5 MB
Ultra Ethernet Transport: Accelerate AI with UET
Comparing RoCE and Ultra Ethernet Transport on the same simulated fabric, using the platform’s UET NIC model.
September 2025 · 7 pages · PDF, 4.8 MB
Simulating Scale-Up: A Critical Step in AI Infrastructure Design
Why scale-up fabrics such as NVLink and UALink belong in a network simulation, and what including them changes for network, infrastructure and silicon teams.
July 2025 · 6 pages · PDF, 128 KB
Simulating the Future: The Role of PFC in AI Data Center Network Accuracy
How Priority-based Flow Control shapes simulated results for lossless Ethernet and RDMA traffic, and how the platform models it.
June 2025 · 5 pages · PDF, 594 KB
Beyond GPU Speed: Understanding Network Bottlenecks in Distributed LLM Training
Using Chakra traces to see where an LLM training job loses time to the network, and what ML researchers, infrastructure teams and planners each take from that.
May 2025 · 10 pages · PDF, 5.0 MB
10 AI Challenges Simulation Can Answer
Ten questions AI data center teams cannot settle in a lab or on a small testbed, from topology changes to hardware failures, and how simulation answers each.
March 2025 · 7 pages · PDF, 2.2 MB
The Strategic Importance of Chakra Workloads in Modern Network Simulation
What Chakra workloads are — graphs of the computation and communication in real AI training — and why simulating with them differs from simulating generic traffic.
February 2025 · 6 pages · PDF, 568 KB
The Importance of Simulating Large Language Model Workloads with Network Simulations
Why LLM performance depends on network behavior, and how simulation helps tune routing, transport and bandwidth allocation for LLM traffic.
January 2025 · 5 pages · PDF, 128 KB
Simulating AI at Datacenter Scale
How networking companies have used ns-3 on the Scala Compute Platform to simulate clusters of 64k nodes and beyond.
December 2024 · 7 pages · PDF, 1.0 MB
Data briefs
Two-page summaries of what the platform does.
SCP for ns-3
The platform for ns-3: Chakra workload support, UET and RoCE configuration, scale-out and scale-up topologies, the consortium, and the public API.
January 2026 · 2 pages · PDF, 3.2 MB
Network Simulation with UEC Transports and OEM Hardware Models
The UEC NIC model for ns-3, built to the UEC 0.75 transport specification, with simulation up to 64k nodes.
November 2024 · 2 pages · PDF, 2.5 MB
Public API
An overview of programmatic access to the platform. The full reference is in the docs.
Public API Summary
What the network simulation API covers — workspaces, configurations, models, simulation runs, traces and templates — and what ships with it: an OpenAPI 3.1 specification and a Python client.
January 2026 · 7 pages · PDF, 564 KB
Earlier solution briefs
Written in 2022 about the Scala Compute Platform.
Accelerating Innovation with the Scala Compute Platform
Simulating and emulating datacenter networks at scale, for operators and networking equipment makers, before the infrastructure is built.
June 2022 · 2 pages · PDF, 507 KB
Accelerating Distributed Machine Learning Training
Simulating the cluster network for distributed ML training, to compare strategies, topologies and hardware before deploying them.
July 2022 · 2 pages · PDF, 380 KB
Want to see how a result is structured?
Tell us what you are trying to evaluate and we will walk you through the readout structure directly, including the sections that are still being finalized.