poster contact name p5286 nicolas delbosc: mnnd leeds.ac...
TRANSCRIPT
hot aircreated bythe serverscold air
coming fromthe floor vent
air is exitingat the CRAC unit
servers
(and other engineering applications)
MotivationsFluid Simulation
User Inputs(keyboard, mouse, motion sensors...)
BoundaryConditions
FastVisualisation(3D rendering,graphs, VR...)
Coupled Model
We gratefully acknowledge the support of NVIDIA with the donation of the Tesla K40 GPU used forthis research.
Turbulence Model: Smagorinsky LES
Saving Energy in Data Centers Using Real‐Time SimulationN. Delbosc, A. Khan, J.L. Summers1 2 1
1School of Mechanical Engineering, School of Civil Engineering, University of Leeds, UK2
Lattice Boltzmann Method
Context: up to 40% of the energy spentin Data Centres is on cooling.Objective: to develop a real-time indoorairflow simulation tool allowing: optimisation of ventilation system design, control and prediction of thermal loads. Benchmark Data Center
Real‐Time CFD Model
Performances
Results (live demo)
Tricks for the Tesla K40
Other Applications
References
Acknowlegments
CPU
Tesla C2070
Tesla K40
GTX Titan
GTX Titan Z
GTX 780 Ti
440200
192720
10-20
Brix(GTX 760)
7431515
1480740
7861580
5561785
500 1000 1500double precision single precision
disable ECC: (14% faster)
disable GPU boost: (13% faster)
combined speed increase: +27%
microscopicscale
mesoscopicscale
macroscopicscale
LBM Navier‐StokesParticle
Methods
velocity, pressuredistribution functionsparticles position and momentum
(higher is better)MLUPS
12
3
4
5
6
Velocity Temperature
D3Q19 D3Q6
12
3
4
5
6
advection
Boussinesq
Boussinesq forcing term:
7
8 9
10
11
1213
1415
16
17
18
variableIT loadpowers
flow ratestemperatures
...
Optimized implementation graphics processing unit towards real-time fluid simulation. (CAMWA 2013)http://dx.doi.org/10.1016/j.camwa.2013.10.002http://www.efm.leeds.ac.uk/~mnnd/index.php
nvidia-smi -e 0
nvidia-smi -ac 3004,875
The framework can be used for other applications.
example: hospital
CONTACT NAME
Nicolas Delbosc: [email protected]
P5286
CATEGORY: COMPUTATIONAL PHYSICS - CP20