• Aucun résultat trouvé

Brian2GeNN: accelerating spiking neural network simulations with graphics hardware

N/A
N/A
Protected

Academic year: 2021

Partager "Brian2GeNN: accelerating spiking neural network simulations with graphics hardware"

Copied!
13
0
0

Texte intégral

Loading

Figure

Table 1.  Configurations used for benchmarking. * Maximum performance in GFLOPS.
Figure 1.  Benchmark of the net simulation time on a 12 core CPU with a single thread (dark gray) or using  OpenMP with 24 threads (light gray), compared to a consumer GPU (TITAN Xp) and an HPC model (Tesla  V100)
Figure 2.  Benchmarking of the net simulation time for different GPU models. Measurements were taken  separately for the MBody model (top) and COBAHH model (bottom) for double precision floating point (left)  and single precision (right)
Figure 4.  Minimal biological runtime after which the total simulation time, including preparations such as  code generation and compilation (cf
+3

Références

Documents relatifs

Onil Nazra Persada Goubier et al.: Wireless Sensor Network-based Monitoring, Cellular Modelling and Simulation UBO/LABSTICC developed PickCell / NetGen, a basic tool allowing

Unité de recherche INRIA Rennes : IRISA, Campus universitaire de Beaulieu - 35042 Rennes Cedex (France) Unité de recherche INRIA Rhône-Alpes : 655, avenue de l’Europe - 38334

In this pa- per, a finite element model with enhanced kine- matics for the simulation of structures with em- bedded anchors is presented.. An elemental en- richment is added to

In particular, a steady solution was obtained, even if in [19] experimen- tal measurements showed an unsteady behavior of the cavitating flow. This was due to the choice of

During all the access area, the request can be asynchronously treated because the simulation description ensures that the data will not be modified. At the end of

photons Collimator Detected photons Emitted photons ARF plane ARF-histo Emitted ARF-nn or One count in one energy window Detection probability in all energy windows Image

In addition, based on the observation that the full range of floating point accuracy may not be always needed, we propose and implement an initial design of the tolerance value

This garbage comes from the original clauses at the start of the DPLL (T ) rewriting sequence, which might have been simplified in later steps of DPLL (T ) but which remain unchanged