The new family of processors squeezes up to 16 cores and 32 threads into a 4 nm chiplet, hitting 5.7 GHz clocks and packing as much as 128 MB of L3 cache. AMD is pitching the parts as ideal for latency-sensitive jobs such as network packet processing, industrial controls and even AI inference thanks to full AVX-512 support.

The lineup comes with six SKUs, ranging from 65W to 170W TDP, letting customers pick between efficiency and raw grunt. AMD says it can deliver 2.8 times the crypto-mining throughput of comparable Xeons, which will not amuse Troubled Chipzilla’s marketing department.

Security and resilience are baked in, with DRAM ECC, on-chip parity checks and PCIe error detection. Infinity Guard is along for the ride to keep the bad guys at bay, while DDR5-5600 support and 28 lanes of PCIe Gen 5 give plenty of bandwidth for networking, storage and accelerators.

Perhaps the biggest selling point is longevity. AMD is promising seven years of planned manufacturing availability, aiming squarely at embedded buyers who hate surprise obsolescence. To make integration easier, the chips use the same AM5 socket that underpins multiple EPYC generations, so board designers can swap parts without reinventing the wheel.

With the EPYC 4005, AMD is signalling it wants to dig deeper into embedded and edge deployments, offering what it calls “future-ready designs.” In practice, it is another move to erode Chipzilla’s stranglehold on industrial compute, one firewall and factory floor at a time.