TOPS (Tera Operations Per Second)

The standard unit for AI chip compute: how many trillion operations it can do per second.

TOPS is the most commonly advertised spec for AI chips; 1 TOPS means one trillion operations per second, usually measured at INT8 (8-bit integer) precision, though some vendors quote it at lower precisions like INT4 or FP4, which inflates the number. It's not the same as TFLOPS (trillion floating-point operations per second), so it's important to check which precision a number refers to before comparing chips. TOPS is a theoretical peak; how fast a model actually runs also depends on memory bandwidth, the software stack, and operator support, so it should only be used as a rough guide. When choosing a robot's onboard compute platform, TOPS gives a first-pass sense of whether it can run a vision model or a VLA on the robot itself.

ExampleNVIDIA's Jetson AGX Orin 64GB is rated at 275 TOPS (INT8, sparse), and it's a common choice as the onboard compute platform for humanoid robots.

Also called

Trillion Operations Per Second

Related

Onboard Compute Platform · System on Chip (SoC) · FLOPS / TFLOPS (Floating-Point Operations per Second) · Numerical Precision Formats (FP32 / FP16 / BF16 / FP8 / INT8 / INT4) · NVIDIA Jetson Orin · On-Device / Edge Deployment

Sources

NVIDIA Jetson Orin