Qualcomm Robotics RB5/RB6 - Why the RB5/RB6?
The Qualcomm RB5 (QRB5165) is an alternative embedded system for robotics and drone applications. It has certain advantages against the other popular platforms, particularly power consumption.
In this section, we will compare different embedded systems used in the market for similar applications, including NVIDIA and NXP. In particular, one of the best i.MX 8 (i.MX 8QM)[1][2][3], the cheapest Jetson Orin (Orin Nano) and a middle-end Jetson Orin (Orin NX)[4].
According to the comparison, the Qualcomm RB5 CPU performance is comparable to the middle-end NVIDIA embedded system, with an octa-core processor at 1.8 GHz. Moreover, the RB5 has 8GB LPDDR5 of RAM, comparable to the middle-end NVIDIA systems.
The GPU seems to be the weakest point of the RB5, providing an approximate performance of 1.3 TFLOP/s, compared to the Orin Nano with 10 TFLOP/s. Moreover, the CUDA ecosystem favours the NVIDIA platform. Nevertheless, the RB5 has AI acceleration, which outperforms the low-end NVIDIA Orin Nano and the i.MX 8 and a DSP unit, not included in any NVIDIA system but in i.MX 8.
The camera capture is another exciting comparison point. The RB5 is competitive against the middle-end NVIDIA system, supporting up to 7 concurrent cameras at 2.5 Gbps per lane with accelerated Video Processing Units and Image Signal Processing (ISP) units.
At the devkit level, the RB5 has sensors, like an accelerometer, gyroscopes, and pressure sensors, which are not often included in NVIDIA devkits. Moreover, it is equipped with Hardware Crypto Engines, which are absent in NVIDIA so far.
Finally, regarding power consumption, the RB5 provides similar capabilities to middle-end embedded systems like the Orin NX with almost four times less energy, consuming only 7W at maximum power, competing only against the i.MX 8, making it ideal for battery-powered systems like drones.
| Criteria | Qualcomm RB5 | NXP i.MX 8QM | NVIDIA Jetson Orin Nano | NVIDIA Jetson Orin NX |
|---|---|---|---|---|
| GPU | Adreno GPU 650 1.3 TFLOP/s |
2xGC7000XSVX GPU 256 GFLOP/s |
NVIDIA Ampere Orin Nano 8GB: 10 TFLOP/s |
NVIDIA Ampere Orin Nano 16GB: 25 TFLOP/s |
| CPU | Octa Core Armv8.2 1.8 GHz (lowest) |
Six Core Armv8 (1.2 GHz) |
Six Core Armv8.2 1.5 GHz |
Octa Core Armv8.2 2.0 GHz (in four cores) |
| AI Acceleration | Qualcomm NPU230 15 TOP/s |
None | None | 2x NVDLA 30 TOP/s |
| ISP Acceleration | Qualcomm Spectra 480 ISP | None | VIC 115 MHz 1x PVA v2.0 |
VIC 115 MHz 1x PVA v2.0 |
| Encoder Acceleration H.265 | Adreno VPU 8x1080p @60fps |
1x1080p @30fps H.264 | CPU 2x1080p @60fps |
Dedicated Unit 6x1080p @60fps |
| DSP Acceleration | Hexagon (also for image processing) | Tensilica HiFi 4 DSP | Via GPU | Via GPU |
| Camera Support | Up to 7 concurrent 2.5 Gbps/lane |
Up to 2 concurrent 1.5Gbps/lane (4 lanes each) |
Up to 4 Cameras 2.5 Gbps/lane |
Up to 6 Cameras ~2.5 Gbps/lane |
| Security Acceleration: cryptography and key provisioning | Crypto engine v5 SPU-240 |
None | None | None |
| Sensors | Accelerometer, Gyros, Pressure, etc… (devkit) |
None | None | None |
| Max. Power Consumption | 7W | 5W (based on stress measurements: VPU + CPU + GPU) |
15W | 40W |
References
- ↑ https://www.nxp.com/products/processors-and-microcontrollers/arm-processors/i-mx-applications-processors/i-mx-8-applications-processors:IMX8-SERIES
- ↑ https://developer.ridgerun.com/wiki/index.php/I.MX_8_Series_Applications_Processors
- ↑ https://www.nxp.com/docs/en/nxp/application-notes/AN12118.pdf
- ↑ https://developer.nvidia.com/embedded/jetson-modules