NU4000 provides computing power exceeding a total of 8 Terra OPS, said to be the most powerful vision processor available:
- 3 Vector Cores that provide 500 Giga OPS
- A dedicated CNN processor that exceeds 2 Terra OPS enabling deep neural networks such as VGG16 reaching 40 frames (ROIs) per second at 10 times less power of the equivalent GPU, DSP or FPGA implementations
- 3 Powerful CPU Cores that provide more than 13,000 CoreMark
- Depth processing engine that delivers a throughput of 120Mp/s and supports multiple simultaneous streams of stereo and structured light
- SLAM engine enabling accurate key point extraction at 120fps from 2 cameras simultaneously
- Advanced Time-Warp HW engine that reduces the Motion-to-Photon latency to 1msec for extensive VR and MR use cases
- More than 3MB of on-chip SLAM servicing the vision cores
- High throughput LPDDR4 interface that reduce external memory access bottlenecks
- Connects to 6 cameras and 2 displays
- Chip area 7 x 8mm2 in 12nm process
Now you just need someone to write the non reusable code for a proprietary platform which you have no clue will be around in a year.
ReplyDelete