AMD Introduces X100 Chips on Zen 5 for Robots with 24/7 Operation and 10-Year Lifespan

AMD unveiled specialized versions of its Strix Halo hybrid chips, branded as the X100 series, for physical AI systems in robotics. These chips share similar specifications with the client Ryzen AI Max but are designed for continuous 24/7 operation and a 10-year service life. Three models are available: X199 with 16 Zen 5 cores and 40 RDNA 3.5 compute units, X188 with 12 cores and 32 compute units, and X168 with 8 cores and 32 compute units. They feature frequencies up to 5.1 GHz, support up to 128 GB of unified memory, include an XDNA 2 NPU with 50 TOPS performance, configurable TDP from 45 to 120 W, and an operating temperature range of -40°C to 105°C. AMD claims the flagship X199 outperforms Intel Core Ultra X7 358H by 1.2x in GeekBench 6.1, 1.3x in PassMark, and 1.5x in SPECrate 2017 integer tests, as well as 1.4x in Vulkan and 1.7x in OpenGL graphics benchmarks. In physical AI tasks, it achieved 1.4x lower time-to-first-token and 3.5x higher tokens per second in a Llama benchmark with Vulkan backend at 45 W TDP. However, the comparison was not entirely fair: AMD used its Ryzen AI Max 395+ configured as X199 on a reference board, while Intel's chip was tested at 30 W TDP with performance scaled to 45 W using public benchmark scaling factors. AMD also offers the X100 chips as part of the Kria System on Module (SOM), a 120x120 mm integrated platform compliant with COM-HPC form factor, including an AMD Spartan UltraScale+ FPGA and interfaces for cameras, industrial networks, and sensors. In benchmarks commissioned by Open Navigation and Mimix, the AMD X100 Kria was compared to Nvidia Thor T5000, but the comparison used a Jetson AGX Thor developer kit versus a mini-PC with a similarly configured Ryzen AI Max+ 395. AMD promotes its HIPIFY tool to convert CUDA code to AMD HIP C++, claiming 70-80% automation for porting efforts, successfully porting 15 CUDA applications with 1199 lines of code. AMD plans to build a comprehensive solution for humanoid robots using Spartan UltraScale+, Zynq UltraScale+, and Versal AI Edge Gen 2 FPGAs.