The fleet is already mixed
“Microsoft therefore maintains many FPGA parts and families in production simultaneously, supporting multiple board generations for a decade or longer. Currently, Azure has 5 generations of FPGA boards actively in deployment.”
Five generations means five timing models, five sets of vendor tools and five ideas of what "healthy" means. Add a second vendor and the number doubles. A dashboard built on any one of those models cannot say whether a 28 nm card in one rack is in better shape than a 16 nm card in the next.
What a fleet metric has to be
- Measured, not modeled. The number describes this card, today, not the family's worst case.
- Comparable. Measured slack as a share of the clock period means the same thing on every part, so one threshold applies everywhere.
- Per element. A card-level average hides the one weak element that will fail first.
- Continuous. A number from acceptance day is a history lesson. About once a second is a health signal.
| Vendor | Family | Process node | Status |
|---|---|---|---|
| AMD | 7-series | 28 nm | Supported |
| AMD | UltraScale | 20 nm | Supported |
| AMD | UltraScale+ | 16 nm | Supported |
| Altera | Agilex 7 | 10 nm class | Select devices |
| Altera | Stratix 10 / Arria 10 | 14 / 20 nm | In development |
| AMD | Versal AI Core | 7 nm | In development |
Two layers, so the view never changes
What your team sees and acts on is the same on every vendor: the health metric, the telemetry schema, the alert thresholds and the aging model. What differs between vendors, integration with each toolchain and calibration for each family and node, is handled below that line by Fluid Silicon. When a new family is qualified, it appears in the same dashboard with the same number.
One set of rules for change
A fleet metric is only useful if the actions it drives are as uniform as the data. Read-only is the default. Voltage scaling, frequency scaling and repair are enabled separately, by policy. Changes run only in maintenance windows you schedule, can be coordinated per device, per server or across the fleet, and every one of them is handshake-gated and measured again within a second.
Where this shows up
Fleet Deployment & Operations covers the telemetry, the change control and the dashboards. Device support lists the families and their status.
Cited for context. Fluid Silicon is not affiliated with Microsoft.