Daily Image
03-11-2022
Colloquium: GPUs in Radio Astronomy
| Submitter: | John Romein |
| Description: | The gray box in the image is our latest acquisition: an NVIDIA Jetson AGX Orin. It contains a highly energy efficient and tightly integrated GPU/CPU System-on-Chip. These small 11cm x 11cm systems are designed for edge computing (i.e., in the field) with GPUs. The PCIe slot in the image center contains a decent (dual 25 Gb/s Ethernet) network card, so that we can stream large amounts of data into the Jetson. On the right: a PowerSensor 3 prototype, with which we can monitor the Jetson's energy use at submillisecond time scales. The display in the bottom-right corner shows an instantaneous power use of 74 Watt (including network card), while the system is under heavy computational load. Join the colloquium this afternoon to know more about it: GPUs are extensively used for radio-astronomical signal and image processing, as they are fast and highly energy efficient. Today, I will present how new GPU technology enabled us to develop the tensor-core correlator: the fastest and most energy efficient GPU correlator worldwide. I will also discuss the next challenge: how to feed GPUs fast enough with data using next-generation network technology, which is necessary for wide-band instruments. Finally, I will show how we incorporate these GPU technologies to increase bandwidth, sensitivity, and greatly reduce latency in the new AARTFAAC processing pipeline. |
| Copyright: | John Romein |
| Tweet |  |