Studying the Impact of CPU and Memory Controller Frequencies on Power Consumption of the Jetson TX1

Abdelhafez, Hazem A.; Ripeanu, Matei

doi:10.1109/fmec.2019.8795334

Cited by 2 publications

(4 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Holly, et al [19] correlate CPU and GPU frequencies and the number of CPU cores with the latency, power and energy for inferening on the Jetson Nano. Some [2] examine the effect of these on power consumption for stream processing workloads. We focus on the impact of such configurations, including storage media and DNN framework settings, on the end-to-end time and energy consumed for DNN training workloads on the Jetson AGX Xavier.…”

Section: Related Workmentioning

confidence: 99%

“…We use the following default configurations in our experiments based on best practices from literature [1,2], unless stated otherwise. The default power mode is the highest rated for all devices: MAXN for the AGX and Nano, and 15𝑊 for NX (modes 𝑔, 𝑀𝐴𝑋 𝑁 and 15𝑊 in Table 3).…”

Section: Default Configurationmentioning

confidence: 99%

“…So, we get caching benefits only in a sweet spot, when: (1) the training data is small enough to fully fit in the cache, i.e., within the available RAM after loading the DNN model, and yet (2) the training data also is large enough that the fetch I/O time dominates over the pre-processing time. 5.1.3 Pipelining reduces the stall time.…”

Section: Pipelined Training and Disk Cachingmentioning

confidence: 99%

“…Hence, we only report data for the first 15ℎ. 2 In our experiments, we notice slight variations in frequencies even with DVFS off, e.g., when the CPU is set as 2265𝑀𝐻𝑧 we notice readings of 2263𝑀𝐻𝑧 and 2262𝑀𝐻𝑧. For simplicity, we consider all frequencies beyond the 95 𝑡ℎ percentile of the expected value to be the static value.…”

Section: Effect Of Dvfsmentioning

confidence: 99%

See 3 more Smart Citations

Characterizing the Performance of Accelerated Jetson Edge Devices for Training Deep Learning Models

S.K

Kesanapalli

Simmhan

2022

Proc. ACM Meas. Anal. Comput. Syst.

View full text Add to dashboard Cite

Deep Neural Networks (DNNs) have had a significant impact on domains like autonomous vehicles and smart cities through low-latency inferencing on edge computing devices close to the data source. However, DNN training on the edge is poorly explored. Techniques like federated learning and the growing capacity of GPU-accelerated edge devices like NVIDIA Jetson motivate the need for a holistic characterization of DNN training on the edge. Training DNNs is resource-intensive and can stress an edge's GPU, CPU, memory and storage capacities. Edge devices also have different resources compared to workstations and servers, such as slower shared memory and diverse storage media. Here, we perform a principled study of DNN training on individual devices of three contemporary Jetson device types: AGX Xavier, Xavier NX and Nano for three diverse DNN model--dataset combinations. We vary device and training parameters such as I/O pipelining and parallelism, storage media, mini-batch sizes and power modes, and examine their effect on CPU and GPU utilization, fetch stalls, training time, energy usage, and variability. Our analysis exposes several resource inter-dependencies and counter-intuitive insights, while also helping quantify known wisdom. Our rigorous study can help tune the training performance on the edge, trade-off time and energy usage on constrained devices, and even select an ideal edge hardware for a DNN workload, and, in future, extend to federated learning too. As an illustration, we use these results to build a simple model to predict the training time and energy per epoch for any given DNN across different power modes, with minimal additional profiling.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Default Configurationmentioning

confidence: 99%

Section: Pipelined Training and Disk Cachingmentioning

confidence: 99%

Section: Effect Of Dvfsmentioning

confidence: 99%

See 2 more Smart Citations