AMD Ryzen AI Embedded P132i socket: FP8

What the Ryzen AI Embedded P132i is

The Ryzen AI Embedded P132i carries the 6-core P132 specification into industrial temperatures. Two Zen 5 cores plus four Zen 5c cores give 12 threads with boost to 4.5 GHz, cache is 6 MB of L2 with 8 MB of L3, graphics are 4 RDNA 3.5 compute units to 2800 MHz, and the XDNA 2 NPU offers up to 50 TOPS. Default TDP is 28 W in the 15 to 54 W range on TSMC 4 nm.

The i grade moves the junction floor to minus 40 degrees while holding the 105 degree ceiling, which admits cold start cabinets, vehicles and outdoor enclosures. Performance figures match the standard AMD Ryzen AI Embedded P132 exactly, so only thermal validation changes between the grades. Memory validation on industrial boards caps LPDDR5X at 8000 MT/s.

The 10-year lifecycle with last time buy in 2036 covers both grades. The 6-core parts launched with the P100 range in January 2026 and reached production in the second quarter, so the platform is mature.

Socket and board compatibility

The chip solders down in FP8 with a permanent 25 × 40 mm joint, so board and module choice is the compatibility story. Vendors such as Advantech, congatec and Kontron carry production modules, and every build is qualified for its SKU across the industrial span.

Firmware must name the SKU, so reused carriers need updates that list the P132i. Package sharing with Phoenix through Gorgon Point helps hardware reuse, but qualification stays per model.

Inside the industrial range the choice is core count, with 8-core and larger i models above on the same package. Beyond FP8 only the Halo class FP11 offers more.

Memory, PCIe and cooling

Memory is DDR5-5600 or soldered LPDDR5X validated to 8000 MT/s on industrial boards, dual channel with ECC and Link-ECC options. Capacity is fixed at manufacture on soldered layouts.

Expansion gives 14 PCIe Gen4 lanes, two USB4 ports, USB 3.2 and 2.0, four display outputs to 7680x4320, and 10-gigabit Ethernet with time sensitive networking for deterministic links. Lanes serve NVMe, networking and vision per board manual. The 4 compute unit graphics engine covers cockpit and HMI display.

Cooling designs to the configured TDP from 15 to 54 W with materials rated minus 40 to 105. The 28 W default suits fanless industrial enclosures. No boxed cooler exists.

Platform, software and lifecycle

AMD targets the 6-core industrial SKU at HMIs, gateways and in-vehicle controllers with real graphics in harsh air. Linux, Xen, FreeRTOS and ROCm span control, separation, real-time and GPU math, and the 10-year supply to 2036 protects long programs. The 4-to-6-core P100 SKUs also carry automotive AEC-Q100 qualification for cockpit programs, which pairs the quality paperwork vehicle makers require with the industrial temperature grade of the i models.

Families often validate one board for both grades, keeping software and spares common across indoor and outdoor installs.

Upgrade paths

Soldered parts cannot be upgraded. New builds around smaller or larger industrial P100 SKUs are the only moves inside FP8, and Halo needs FP11. Desktop Gorgon class silicon uses AM5.

Size SKU and memory for lifetime duty at project start, since the field allows no changes.