AMD is taking its Strix Halo-style silicon out of laptops and mini PCs, and pushing it into something much more industrial: robots.
The new Ryzen AI Embedded X100 lineup is built for physical AI systems, meaning robots, automation gear, industrial machines, smart cameras, and other devices that need local compute without depending fully on the cloud. For SEA, that matters more than it sounds. Malaysia, Singapore, Thailand, Vietnam, and Indonesia are all pushing harder into automation, logistics robotics, smart factories, and AI-assisted manufacturing. Chips like this are basically the “brain” layer for that future.
Three X100 chips, led by the X199
AMD’s X100 family has three models, roughly matching the earlier Strix Halo/Ryzen AI Max tiering.
The flagship X199 comes with 16 Zen 5 CPU cores and 40 RDNA 3.5 GPU compute units. Below that, the X188 offers 12 CPU cores and 32 GPU compute units, while the X168 has eight CPU cores with the same 32 GPU compute units.
AMD says the lineup can reach up to a 5.1GHz boost clock, support up to 128GB unified memory, and includes an XDNA 2 NPU rated up to 50 TOPS. The chips can run from 45W to 120W, and are designed for rougher environments with an operating range from -40°C to 105°C.
That last part is key. This is not just “powerful chip, put in robot, done.” Embedded customers care about stability and long support windows, so AMD is promising 24/7 operation and a 10-year embedded lifecycle. For companies building robotics platforms, kiosks, factory systems, or long-life commercial machines, that support window is a big deal.
AMD is clearly aiming at Intel and Nvidia
This launch is AMD’s answer to Intel’s physical AI push with Panther Lake SoCs, and also part of AMD’s broader attempt to pull developers away from Nvidia’s ecosystem.
The pitch is simple: instead of splitting CPU, GPU, AI accelerator, and memory across different components, put more of it together in one SoC. That can reduce latency and simplify system design, which is useful when a robot needs to process camera data, sensor input, AI models, and movement decisions quickly.
AMD shared benchmark claims for the X199 against Intel’s Core Ultra X7 358H. According to AMD, the X199 leads by 1.2x in Geekbench 6.1, 1.3x in PassMark, and 1.5x in an unofficial SPECrate 2017 integer workload. On graphics, AMD claims 1.4x faster Vulkan, 1.7x faster OpenGL, and 1.6x better Unigine Heaven Extreme performance.
For AI workloads, AMD also claims better Llama-bench results, including 1.4x faster Time to First Token and 3.5x higher tokens per second using a Vulkan backend at 45W.
But bro, don’t swallow the graphs whole.
The testing was not fully apples-to-apples. AMD used a Ryzen AI Max 395+ configured to reflect X199 specs on a reference board, while Intel’s chip was tested in an MSI Prestige 16 Flip AI+ with a 30W limit, then projected to 45W using scaling assumptions. Useful data point? Sure. Final verdict? Not yet.
Kria X100 gives developers a robot-ready board
Beyond the chips, AMD is also offering X100 through a Kria System on Module. The Kria X100 board measures 120mm x 120mm and follows the COM-HPC form factor.
AMD also has a full robotics developer platform using the X100 Kria SOM with a Spartan UltraScale+ FPGA baseboard, with connectivity aimed at cameras, industrial networking, and robotic sensors. Early access is available now, with full production planned for Q4 this year.
AMD also discussed comparisons against Nvidia’s Thor T5000, but again, the setup was indirect: Nvidia’s Jetson AGX Thor developer kit was compared against a GMKtech EVO-X2 AI mini PC using a Ryzen AI Max+ 395 configured to reflect X199 specs.
The bigger story is software. AMD says its HIPIFY tool can now handle around 70% to 80% of the effort when converting CUDA code to AMD’s HIP C++ path, based on 15 CUDA apps totalling 1,199 lines of code.
For Malaysian and SEA developers, the takeaway is straightforward: AMD wants to be taken seriously in robotics AI, not just gaming PCs. If pricing, developer support, and availability line up, X100 could become a practical option for local robotics labs, factory automation teams, and AI hardware startups.
Primary source: AMD
Reporting source: Tom's Hardware