Customized compute acceleration in the datacenter is key to the wider roll-out of applications based on deep neural network (DNN) inference. In this article, we investigate how to maximize the performance and scalability of field-programmable gate array (FPGA)-based pipeline dataflow DNN inference accelerators (DFAs) automatically on computing infrastructures consisting of multi-die, network-connected FPGAs. We present Elastic-DF, a novel resource partitioning tool and associated FPGA runtime infrastructure that integrates with the DNN compiler FINN. Elastic-DF allocates FPGA resources to DNN layers and layers to individual FPGA dies to maximize the total performance of the multi-FPGA system. In the resulting Elastic-DF mapping, the accelerator may be instantiated multiple times, and each instance may be segmented across multiple FPGAs transparently, whereby the segments communicate peer-to-peer through 100 Gbps Ethernet FPGA infrastructure, without host involvement. When applied to ResNet-50, Elastic-DF provides a 44% latency decrease on Alveo U280. For MobileNetV1 on Alveo U200 and U280, Elastic-DF enables a 78% throughput increase, eliminating the performance difference between these cards and the larger Alveo U250. Elastic-DF also increases operating frequency in all our experiments, on average by over 20%. Elastic-DF therefore increases performance portability between different sizes of FPGA and increases the critical throughput per cost metric of datacenter inference.

@en

The Spanish Central System is an intraplate mountain range that divides the Iberian Inner Plateau in two sectors – the northern Duero Basin and the Tajo Basin to the south. The topography of the area is highly variable with the Tajo Basin having an average altitude of 450–500 m and the Duero Basin having a higher average altitude of 750–800 m. The Spanish Central System is characterized by a thick-skin pop-up and pop-down configuration formed by the reactivation of Variscan structures during the Alpine orogeny. The high topography is, most probably, the response of a tectonically thickened crust that should be the response to (1) the geometry of the Moho discontinuity, (2) an imbricated crustal architecture, and/or (3) the rheological properties of the lithosphere. Shedding some light on these features is the main target of the current investigation. In this work, we present the lithospheric-scale model across this part of the Iberian Massif. We have used data from the Central Iberian Massif Deformation (CIMDEF) project, which consists of recordings of an almost-linear array of 69 short-period seismic stations, which define a 320 km long transect. We have applied the so-called global-phase seismic interferometry. The technique uses continuous recordings of global earthquakes (>120∘ epicentral distance) to extract global phases and their reverberations within the lithosphere. The processing provides an approximation of the zero-offset reflection response of a single station to a vertical source, sending (near)-vertical seismic energy. Results indeed reveal a clear thickening of the crust below the Central System, resulting, most probably, from an imbrication of the lower crust. Accordingly, the crust–mantle boundary is mapped as a relatively flat interface at approximately 10 s two-way travel time except in the Central System, where this feature deepens towards the NW reaching more than 12 s. The boundary between the upper and lower crust is well defined and is found at 5 s two-way travel time. The upper crust has a very distinctive signature depending on the region. Reflectivity at upper-mantle depths is scattered throughout the profile, located between 13 and 18 s, and probably related to the Hales discontinuity.@en