Bringing up Cosmo

31 snips

Apr 24, 2025

Nathanael Huffman, an electrical engineer at Oxide Computer Company, joins the discussion to unveil the challenges and triumphs behind their next-generation server, Cosmo. The team shares fascinating tales of hardware design, including the complexities of integrating FPGAs and troubleshooting nuanced issues. They also reflect on the significance of collaboration in overcoming design hurdles, from circuit board rework to power system optimizations. Humor and camaraderie pepper their insights, making the technical intricacies accessible and engaging.

Ask episode

AI Snips

Chapters

Transcript

Episode notes

INSIGHT

FPGA Improves Power Sequencing

Using FPGA for power sequencing added flexibility and easy updates compared to fixed discrete logic.
FPGA-based sequencing hardware supports complex, adaptive power and reset control for AMD Turin processors.

ANECDOTE

Grapefruit Enables Early Risk Reduction

Oxide developed Grapefruit, an OCP DC-SCM form-factor board, as a test platform for service processor and FPGA design before Cosmo tape-out.
Grapefruit enabled months of FPGA development and risk reduction, proving SPD, eSpy, and other critical subsystems early.

ANECDOTE

Pin Mismatch Fixed with Jumper Rework

Initial FPGA communication was blocked due to miswiring of an FMC bus pin, requiring a quick physical jumper rework.
The rework was minor and performed via jumper wires under the BGA SP chip, demonstrating nimble bring-up fixes.

Get the Snipd Podcast app to discover more snips from this episode

Get the app

Oxide is bringing up its next generation server. To discuss the (amazingly smooth) bringup process, Bryan and Adam were joined by members of the oxide team. Tales of adversity, re-work, un-re-work, and triumph!

In addition to Bryan Cantrill and Adam Leventhal, we were joined by Oxide colleagues Nathanael Huffman, Ian Sobering, Matt Keeter, and Aaron Hartwig.

We mentioned quite a few terms! Here's a helpful guide:

Cosmo - Oxide’s next-generation sled (currently in development) with an AMD Turin CPU
Gimlet - Oxide’s current-generation sled with an AMD Milan CPU
Turin - AMD Epyc 9005 Series
Milan - AMD Epyc 7003 Series
Genoa - AMD Epyc 9004 Series (Oxide chose to skip this generation)
Sequencing - the precise control of when power rails are energized throughout a PCB
Sled - One of the (max 32) computers in an Oxide rack; a custom form-factor optimized for power and cooling efficiency
IBC - Intermediate Bus Converter (Our 54VDC -> 12VDC converter)
RoT - Root of Trust
SP - Service Processor, the small computer (running Hubris) that allows for low-level control
Ignition - An even lower-level control network for power management (including power of the SP)
Ruby - The AMD reference platform (Oxide has used this to prepare Cosmo software in advance of bringup)
DC-SCM - https://www.opencompute.org/documents/ocp-dc-scm-spec-rev-1-0-pdf and OpenCompute standard form factor.
Grapefruit - OCP DC-SCM form-factor board with our SP, RoT, and FPGA on it, used to replace the OCP DC-SCM baseboard management controller in the Ruby platform.
Cadence - Software Oxide previously used for PCB design
Altium - Software Oxide now uses for PCB design
Hubris - Oxide’s embedded operating system, run on the SP and RoT
Humility - The Hubris debugger
PLM - Product Lifecycle Management – a class of software used for managing hardware BOMs
BOM - Bill of Materials – the components required to build a hardware product
RFK - Our colleague, Robert Keith (to distinguish him from our other colleague, Robert, and our former colleague, Keith)
FPGA - Field Programmable Gate Array – Also referred to as “soft logic” – effectively programmable hardware
ILA - Integrated Logic Analyzer
JTAG - A debugging interface for various processors
UART - A serial port or connection

For previous tales from the bringup lab:

If we got something wrong or missed something, please file a PR! Our next show will likely be on Monday at 5p Pacific Time on our Discord server; stay tuned to our Mastodon feeds for details, or subscribe to this calendar. We'd love to have you join us, as we always love to hear from new speakers!