SGI
®
Many core computing
Solutions
Gábor Lehoczki
Silicon Computers Kft.
leho@silicon.hu
1
Nvidia Tesla series
Features
Number and Type of
GPU
Peak double precision
floating point
performance
Peak single precision
floating point
performance
M2075 M2090
K20
K20X
1 Fermi GPU
1 Fermi GPU
2 Kepler GK104s
515
665
190 Gigaflops
1.17
1.31
Gigaflops
Gigaflops
(95 Gflops per GPU)
Tflops
Tflops
3.52
3.95
Tflops
Tflops
208
250
GB/sec
GB/sec
5 GB
6 GB
2496
2688
1030
1331
Gigaflops
Gigaflops
Memory bandwidth
150
177
(ECC off)
GBytes/sec
GBytes/sec
Memory size (GDDR5)
6 GB
6 GB
448
512
CUDA cores
K10
4577 Gigaflops
(2288 Gflops per GPU)
320 GB/sec
(160 GB/sec per GPU)
8 GB
(4 GB per GPU)
3072
(1536 per GPU)
2
1 Kepler GK110
Intel® Xeon Phi™ Coprocessor
5110P
3120A
3120P
Code Name
# of Cores
Clock Speed
Peak double
precision floating
point performance
Peak single
precision floating
point performance
Cache
7120P
7120X
5120D
Knights Corner
60
1.053 GHz
57
1.1 GHz
57
1.1 GHz
61
61
60
1.238 GHz 1.238 GHz 1.053 GHz
1.011
TFlops
2.022
TFlops
30 MB
28.5 MB
28.5 MB
30.5 MB
30.5 MB
30 MB
Max TDP
225 W
300 W
300 W
300 W
300 W
245 W
Max Memory Size
8 GB
6 GB
6 GB
16 GB
16 GB
8 GB
5 GB/s
5 GB/s
5.5 GB/s
5.5 GB/s
5.5 GB/s
(16 channels)
Max Memory
Bandwidth
320 GB/s
3
SGI Development Suite
•
•
For Linux
software
development
Supports 2-4
developers
SGI
Performance
Suite
Intel
Composer XE
• Technical Computing
Performance for SGI
Systems
Rogue Wave
TotalView
Team
• Software development
tool suite tuned for
application performance
and code robustness
• Consists of SGI
Accelerate, SGI MPI,
SGI REACT, SGI
UPC
• Dynamic source code,
thread, and memory
debugging for C, C++
and Fortran HPC
applications
• Consists of C++ &
Fortran compilers, Math
Kernel Library,
Integrated Performance
Primitives, Threading
Building Blocks
• Tech support via SGI
Support contract
• Consists of TotalView,
MemoryScape,
ReplayEngine, and
CUDA debugging
• Includes 1 yr tech
support
• Includes 1 yr tech
support
4
4
Intel® Xeon Phi™ compared to a GPU
• Intel® Xeon Phi™ is a multi-core architecture with the
following characteristics :
– based on Pentium 4 in-order cores with a vector extension
– OpenMP programming
– “UV on a chip”
– PCIe x16 Gen 2 (~ 6.7 GB/s)
– Ease-of-Use
• NVIDIA Fermi GPU use streaming multi-processors with
currently 32 SIMD engines
– A very high number of threads (10s of thousands) hides latency to
memory
– Minimal context switching time between threads
– HW thread dispatcher
– PCIe x16 Gen (~ 11.3 GB/s) [Kepler 10]
5
GPU-ACCELERATED APPLICATIONS
01 Research: Higher Education and Supercomputing CASINO Code for performing Quantum Monte Carlo GPUs to determine protein binding sites
COMPUTATIONAL CHEMISTRY AND BIOLOGY
(QMC) electronic structure calculations for
Allows fast processing of large ligand
NUMERICAL ANALYTICS
finite and periodic systems
databases
PHYSICS
TBD Yes
Single only
06 Defense and Intelligence
CP2K Program to perform atomistic and
BUDE Molecular docking program Empirical Free
06 Computational Finance
molecular simulations of solid state, liquid,
Energy Forcefield Single only
07 Manufacturing: CAD and CAE
molecular and biological systems
Core Hopping Rapid screening of novel cores to
COMPUTATIONAL FLUID DYNAMICS
DBCSR (space matrix multiply library) Yes
improve
COMPUTATIONAL STRUCTURAL MECHANICS
GAMESS-UK The general purpose ab initio
drug properties
COMPUTER AIDED DESIGN
molecular
GPU accelerated application Yes
ELECTRONIC DESIGN AUTOMATION
electronic structure program for performing
FastROCS Molecule shape comparison application
09 Media and Entertainment
SCF-, DFT- and MCSCF-gradient
Real-time shape similarity searching/
ANIMATION, MODELING AND RENDERING
calculations
comparison
COLOR CORRECTION AND GRAIN
(ss|ss) type integrals within calculations
Yes
MANAGEMENT
using Hartree-Fock ab initio methods
Interactive Molecule
COMPOSITING, FINISHING AND EFFECTS
and density functional theory. Supports
Visualizer
EDITING
organics and inorganics.
Molecular visualization based on a raytracing engine
ENCODING AND DIGITAL DISTRIBUTION
Yes
Written for use only on GPUs Yes
ON-AIR GRAPHICS
GAMESS-US Computational chemistry suite used to Molegro Virtual Docker 6 Method for performing high
ON-SET, REVIEW AND STEREO TOOLS
simulate atomic and molecular electronic
accuracy
SIMULATION
structure
flexible molecular docking
WEATHER AND CLIMATE FORECASTING
Libqc with Rys Quadrature Algorithm,
Energy grid computation, pose evaluation
13 Oil and Gas
Hartree-Fock, MP2 and CCSD
and guided differential evolution
POPULAR GPU-ACCELERATED APPLICATIONS Yes
Single only
CATALOG | MAR14 | 01
Gaussian
PIPER Protein Docking Protein-protein docking
Research: Higher Education and Supercomputing
(In development)
program Molecule docking TBD
COMPUTATIONAL CHEMISTRY AND BIOLOGY
Predicts energies, molecular structures,
PyMol User-sponsored molecular visualization
Molecular Dynamics
and vibrational frequencies of molecular
system on an open-source foundation
APPLICATION DESCRIPTION SUPPORTED
systems
Lines: 460% increase
FEATURES MULTI-GPU SUPPORT
Joint NVIDIA, PGI and Gaussian
Cartoons: 1246% increase
ACEMD GPU simulation of molecular mechanics
collaboration
Surface: 1746% increase
force fields, implicit and explicit solvent
Yes
Spheres: 753% increase
Written for use only on GPUs Yes
GPAW Real-space grid DFT code written in C and Ribbon: 426% increase
AMBER Suite of programs to simulate molecular
Python
Single only
dynamics on biomolecule
Electrostatic poisson equation,
VEGA ZZ Molecular Modeling Toolkit Virtual logP,
PMEMD Explicit Solvent and GB Implicit
orthonormalizing of vectors, residual
molecular surface values Single only
Solvent
minimization method (rmm-diis)
VMD Visualization and analyzing large biomolecular
Yes
Yes
systems in 3-D graphics
CHARMM MD package to simulate molecular
Jaguar A high-performance ab initio package
High quality rendering, large structures
dynamics on biomolecule
for performing gas and solution phase
(100M atoms), analysis and visualization
Implicit (5x), Explicit (2x) Solvent
simulations
tasks, multiple GPU support for display of
via OpenMM. Native CUDA port in
TBD Yes
molecular orbitals
development
LATTE Density matrix computations CU_BLAS, SP2 Yes
Yes
Algorithm Yes
Bioinformatics
DESMOND High-speed molecular dynamics
MOLCAS Methods for calculating general electronic APPLICATION DESCRIPTION SUPPORTED
simulations of biological systems on
structures in molecular systems in both
FEATURES MULTI-GPU SUPPORT
conventional commodity clusters.
ground and excited states
BarraCUDA Sequence mapping software Alignment
The code uses novel parallel algorithms
CU_BLAS Single only
of short sequencing reads,
and numerical techniques to achieve high
MOLPRO Used for accurate ab initio quantum
alignment of indels with gap openings and
performance and accuracy
chemistry calculations
extensions
Yes
Density-fitted MP2 (DF-MP2), density fitted
Yes
DL-POLY Simulate macromolecules, polymers, ionic local correlation methods (DF-RHF, DFKS), DFT
CUDASW++ Open source software for Smithsystems, etc on a distributed memory
Yes
Waterman
parallel computer
MOPAC2013 Semiempirical Quantum Chemistry
protein database searches on GPUs
Two-body forces, Link-cell pairs, Ewald
Pseudodiagonalization, full diagonalization,
Parallel search of Smith-Waterman
SPME forces, Shake VV
and density matrix assembling
database
Yes
Single only
Yes
Folding@Home A distributed computing project that NWChem Calculations Triples part of Reg-CCSD(T), CUSHAW Parallelized short read aligner Parallel,
studies
CCSD and
accurate long read aligner for
protein folding, misfolding, aggregation, and
EOMCCSD task schedulers
large genomes
related diseases
Yes
Yes
Powerful distributed computing molecular
Octopus Used for ab initio virtual experimentation
04 | POPULAR GPU-ACCELERATED
dynamics system; implicit solvent and
and quantum chemistry calculations
APPLICATIONS CATALOG | MAR14
folding
Full GPU support for ground-state,
GATK The Genome Analysis Toolkit or GATK
Yes
real-time calculations; Kohn-Sham
is a software package developed at the
GPUGrid.net A distributed computing project that
Hamiltonian, orthogonalization, subspace
Broad Institute to analyse next-generation
uses
diagonalization, poisson solver, time
resequencing data. The toolkit offers a wide
GPUs for molecular simulations
propagation
variety of tools, with a primary focus on
High-performance all-atom biomolecular
yes
variant discovery and genotyping as well as
simulations; explicit solvent and binding
Q-CHEM Computational chemistry package designedstrong emphasis on data quality assurance.
Yes
for HPC clusters
Variant Calling; > 70X speedup Yes
GROMACS Simulation of biochemical molecules with Various features including RI-MP2 TBD
GPU-BLAST Local search with fast k-tuple heuristic
complicated bond interactions
QUICK QUICK is a GPU-enabled ab intio quantum Protein alignment according to BLASTP Single only
Implicit (5x), Explicit (2x) Solvent Yes
chemistry software package
mCUDA-MEME Ultrafast scalable motif discovery
HALMD Large-scale simulations of simple and
Running Hartree-Fock and DFT energy on
algorithm
complex liquids
GPU, Supports s, p, d, f orbitals on energy
based on MEME
Simple fluids and binary mixtures (pair
calculation, HF gradient with s,p,d orbital
Scalable motif discovery algorithm based
potentials, high-precision NVE and NVT,
support, GPU-based ERI generator
on MEME
dynamic correlations)
Yes
Yes
Single only
POPULAR GPU-ACCELERATED APPLICATIONS MUMmerGPU High-throughput local sequence
HOOMD-Blue Particle dynamics package written
CATALOG | MAR14 | 03
alignment
grounds
TeraChem Quantum chemistry software designed to program
up for GPUs
run on NVIDIA GPU
Aligns multiple query sequences against
Written for use only on GPUs Yes
Full GPU-based solution; Performance
reference sequence in parallel
LAMMPS Classical molecular dynamics package
compared to GAMESS CPU version
TBD
Lennard-Jones, Gay-Berne, Tersoff, and
Yes
NVBIO NVBIO is an open source C++ library
many more potentials
Materials Science
of reusable components designed to
Yes
APPLICATION DESCRIPTION SUPPORTED
accelerate bioinformatics applications using
NAMD Designed for high-performance simulation
FEATURES MULTI-GPU SUPPORT
CUDA.
of large molecular systems
gWL-LSMS Materials code for investigating the
Data structures, algorithms, and utility
Full electrostatics with PME and most
effects
routines useful for building complex
simulation features; 100M atom capable
of temperature on magnetism
computational genomics applications on
Yes
Generalized Wang-Landau method Yes
CPU-GPU systems
OpenMM Library and application for molecular
PEtot First principles materials code that
Yes
dynamics for HPC with GPUs
computes the behavior of the electron
NVBowtie A largely complete implementation of the
Implicit and explicit solvent, custom forces Yes
structures of materials
Bowtie2 aligner on top of NVBIO
Test Drive the World’s
Density functional theory (DFT) plane wave
Good coverage of Bowtie2 features and
Fastest Accelerator – Free!
pseudopotential calculations
comparable quality results
Take the GPU Test Drive, a free and easy way to
Yes
Yes
experience accelerated computing on GPUs. You
QMCPACK Solves the many-body Schrodinger
REACTA A modified version of GCTA with improved
can
equation
computational performance, support for
run your own application or try one of the preloaded for electronic structures using a quantum
Graphics Processing Units (GPUs), and
ones, all running on a remote cluster. Try it today.
Monte Carlo method
additional features. The purpose of REACTA
www.nvidia.com/gputestdrive
Main features Yes
is to quantify the contribution of genetic
02 | POPULAR GPU-ACCELERATED
Quantum Espresso/
variation to phenotypic variation for complex
APPLICATIONS CATALOG | MAR14
PWscf
traits.
Quantum Chemistry
An integrated suite of computer codes
GRM creation, REML analysis, Regional
APPLICATION DESCRIPTION SUPPORTED
for electronic structure calculations and
Heritability (including multi-GPU)
FEATURES MULTI-GPU SUPPORT
materials modeling at the nanoscale
Yes
Abinit Allows to find total energy, charge density
PWscf package: linear algebra (matix
SeqNFind SeqNFind® is a powerful tool suite that
and electronic structure of systems made of
multiply), explicit computational kernels,
addresses the need for complete and
electrons and nuclei within DFT
3D FFTs
accurate alignments of many small
Local Hamiltonian, non-local Hamiltonian,
Yes
sequences against entire genomes utilizing
LOBPCG algorithm, diagonalization/
VASP First principles materials code that
a unique hardware/software cluster system
orthogonalization. Contains BigDFT.
computes the behavior of electronic
for facilitating bioinformatics research in
Yes
structures based on quantum theory
Next Generation sequencing and genomic
ACES III Takes best features of parallel
Hybrid Hartree-Fock DFT functionals
comparisons.
implementations of quantum chemistry
including exact exchange
Hardware and softare for reference
methods for electronic structure
Yes
assembly, BLAST, SW HMM, and de novo
Integrating scheduling GPU into SIAL
Visualization and Docking Software
assembly
programming language and SIP runtime
APPLICATION DESCRIPTION SUPPORTED
Yes
environment
FEATURES MULTI-GPU SUPPORT
SOAP3 GPU-based software for aligning short
Yes
Amira 5 A multifaceted software platform
reads with a reference sequence.It can find
ADF Density Functional Theory (DFT) software
for visualizing, manipulating, and
all alignments with k mismatches, where k
package that enables first-principles
understanding Life Science and bio-medical
is chosen from 0 to 3
electronic structure calculations
data
Short read alignment tool that is not
Fock Matrix, Hessians Yes
3D visualization of volumetric data and
heurisitic based, reports all answers
BigDFT Implements density functional theory
surfaces
Yes
by solving the Kohn-Sham equations
Single only
SOAP3-dp SOAP3-dp: Ultra-fast GPU-based tool for
describing the electrons in a material
BINDSURF A virtual screening methodology that
short read alignment via index-assisted
DFT; Daubechies wavelets, part of Abinit Yes
uses
dynamic programming
Borrows-Wheeler Transformation,
only
calculations of a wide variety of pricing and
iray rendering
in Maximus supported GPU simulation using
Element 3D
Accuweather Storyteller Weather graphics Real-time
Dynamic Programming, Multi-GPU support
Intuvision Panoptes 3.0 Video Analytics Object
risk measures on a broad range of asset
Yes
CUDA
3D object based particle system Faster effects Yes Single only
Yes
recognition and change detection Yes
classes and derivatives
Inventor 3D mechanical design, documentation, and Single only
POPULAR GPU-ACCELERATED APPLICATIONS ASUCA Regional atmospheric model Entire model
UGENE Opensource Smith-Waterman for SSE/
LuciadLightspeed Geospatial Visualization and
Existing models code in C# supported
product simulation
Maxon Cinema 4D 3D modeling, animation, and
CATALOG | MAR14 | 11
Yes
CUDA, Suffix array based repeats finder and
Analysis Geospatial Situational Awareness Single
transparently, with minimal code changes,
Uses BIM for intelligent building
rendering Increased model complexity at interactive Video Copilot Optical
CAM - SE Global atmosphere model for climate
dotplot
only
Supports multiple backends including
components to improve design accuracy
rates
Flares
research
Fast short read alignment Yes
MotionDSP Ikena ISR Real-time Full Motion Video CUDA and OpenCL, Switches transparently
Single only
Single only
Lens flares plug-in for After Effects Faster effects
Dynamical core Yes
WideLM Fits numerous linear models to a fixed
and WAMI
between multiple GPUs and CPUS
Revit Building Information Modeling (BIM) for
NewTek Lightwave 3D modeling, animation, and
Single only
COSMO Regional atmospheric model Entire model
design and response
enhancement and analytics
depending on the deal support and load
structural engineering, and construction
rendering Increased model complexity at interactive Video Copilot Twitch Video effects plug-in for After Yes
Parallel linear regression on multiple
Real-time super-resolution-based video
factors
Modeling (BIM) to design, build, and
rates
Effects Faster effects Single only
GEOS-5 Global climate model Entire model Yes
similarly-shaped models
enhancement, filtering, mosaicing, video
Yes
maintain higher-quality, more energyefficient
Single only
EDITING
HOMME Dynamical core for global atmospheric
Yes
analytics, and transcoding
Manufacturing: CAD and CAE
buildings
Otoy Octane Render GPU Renderer CUDA-based APPLICATION DESCRIPTION SUPPORTED
model
POPULAR GPU-ACCELERATED APPLICATIONS Yes
COMPUTATIONAL FLUID DYNAMICS
Single only
GPU final-frame rendering Yes
FEATURES MULTI-GPU SUPPORT
Dynamical core Yes
CATALOG | MAR14 | 05
Nervve ViD SrX Video Analytics Object recognition APPLICATION DESCRIPTION SUPPORTED
NVIDIA Iray A ready-to-integrate, physically-based, Pixologic Sculptris 3D sculpting Increased model
Adobe Photoshop CC Image editing Over 30 effects HYCOM Ocean circulation model Dynamical core
NUMERICAL ANALYTICS
and tracking Yes
FEATURES MULTI-GPU SUPPORT
photorealistic rendering solution
complexity at interactive
for smoother image
Single only
APPLICATION DESCRIPTION SUPPORTED
NerVve Visual Search
Altair AcuSolve General purpose CFD software
Iray Interactive; Iray Photoreal; Iray
rates
manipulation in Mercury Graphics Engine
Meteo Earth Weather graphics Real-time Single only
FEATURES MULTI-GPU SUPPORT
Solution (NVSS)
Linear equation solver Yes
Cluster. Fast interactive ray tracing;
Single only
Single only
MITgcm Ocean circulation model Dynamical core
ArrayFire Comprehensive GPU function library
Video/Image Live and Forensic Search Video and
ANSYS Fluent General purpose CFD software
Physically-based, global-illumination
Redshift Renderer GPU-accelerated, biased renderer Adobe Premiere Pro CC Video editing Mercury
Single only
Hundreds of functions for math, signal/
Image content search Yes
Radiation heat transfer model, linear
rendering; Distributed cluster rendering
CUDA-based GPU final-frame rendering Yes
Playback Engine for real-time
NEMO Ocean circulation model Entire model (GYRE
image processing, statistics, and more.
OpCoast SNEAK Electromagnetic signals
equation solver
Yes
Side Effects Houdini 3D modeling, animation, and
video editing & accelerated rendering
config) Yes
Yes
propagation
Yes
Dassault Systemes CATIA
rendering Maximus-supported GPU simulation using Yes
NIM Dynamical core for global atmospheric
Mathematica Wolfram A symbolic technical
modeling for complex urban and terrain
Autodesk Moldflow Plastic mold injection software
- Live Rendering
OpenCL
Apple Final Cut Pro Video editing Faster effects
model
computing language
environments
Linear equation solver Single only
Photorealistic rendering Interactive. Fully integrated Single only
Single only
Dynamical core Yes
and development environment
Ray tracing, DTED and remote sensing
CPFD Barracuda-VR and
in CATIA V6.
The Foundry Mari 3D paint Increased model
Avid Media Composer Video editing Faster video
Weather Central Fusion
Development environment for CUDA and
inputs
Barracuda
Network rendering
complexity at interactive
effects, unique stereo 3D
Studio
OpenCL
Yes
Fluidized bed modeling software Linear equation
Yes
rates
capabilities
Weather graphics Real-time Single only
Yes
PCI Geomatics GXL Geospatial Visualization Image solver, particle
Bunkspeed Suite Easy to use photorealistic renderingSingle only
Single only
WRF Regional numerical weather prediction
MATLAB by Mathworks GPU acceleration for
orthorectification and additional
calculations
software
10 | POPULAR GPU-ACCELERATED
Grass Valley Edius Video editing Faster effects
model
MATLAB (high-level
image processing
Single only
Iray-based ray-tracing. Animation support.
APPLICATIONS CATALOG | MAR14
Single only
WSM5, WSM3, Ice Microphysics models Yes
technical computing language)
Yes
FluiDyna Culises for
Network rendering
COLOR CORRECTION AND GRAIN
Harris Velocity Video editing Faster effects Single
WSI TrueView Max Weather graphics Real-time
Support for 200+ of most used MATLAB
GAIA Multi-GPU, Multi-Machine distributed
OpenFOAM
Yes
MANAGEMENT
only
Single only
functions (incl. Signal Processing, Image
object store providing SQL style query
Solver library for general purpose CFD
RTT DeltaGen Redefines high-end 3D visualization APPLICATION DESCRIPTION SUPPORTED
Quantel Qube Broadcast video editing Faster video POPULAR GPU-ACCELERATED APPLICATIONS
Processing, Communications Systems, etc)
capability, advanced geospatial query
software
and
FEATURES MULTI-GPU SUPPORT
effects, unique stereo 3D
CATALOG | MAR14 | 13
Yes
capability,heatmap generation, and
Linear equation solvers Yes
realtime interaction. This latest version
Adobe SpeedGrade CC Color grading Real-time
capabilities
Oil and Gas
NMath Premium GPU-accelerated math and
distributed rasterization services
FluiDyna LBultra General purpose CFD software
gives users a broad suite of robust new
grading and finishing with
Single only
APPLICATION DESCRIPTION SUPPORTED
statistics for
Built for GPUs, scales to many machines
Lattice-Boltzmann solver Yes
features to truly revolutionize processes
Lumetri Deep Color Engine
Sony Vegas Pro Video editing Faster video effects FEATURES MULTI-GPU SUPPORT
.NET, automatically detects the presence
and many cards without end user query
Prometech Particleworks Particle-based CFD
and help increase visual quality, speed, and
Single only
and encoding Single only
Acceleware
of a CUDA-enabled GPU at runtime
modification or pre-meditation
software Implicit and explicit solvers Single only
flexibility.
Assimilate Scratch Color grading and finishing
ENCODING AND DIGITAL DISTRIBUTION
AxRTM
and seamlessly redirects appropriate
Yes
Vratis ARAEL General purpose CFD software based Interactive ray tracing and global
Accelerated debayering for real-time
APPLICATION DESCRIPTION SUPPORTED
AxKTM
computations to it.
Computational Finance
on
illumination. Integration with Siemens
digital finishing
FEATURES MULTI-GPU SUPPORT
Seismic Processing RTM, Kirchhoff, control source,
Automatically offloads computations to the
APPLICATION DESCRIPTION SUPPORTED
FVM with OpenFOAM compatibility
TeamCenter. Cluster support Realtime
Single only
Cinnafilm Tachyon Standards conversion Video
electromagnetism, forward modeling
GPU
FEATURES MULTI-GPU SUPPORT
Linear equation solver Yes
& Offline Production Process Integration
Blackmagic DaVinci
procession and encoding Yes
Yes
Single only
Aaon Benfield
Turbostream Ltd. CFD software for turbomachinery and scene building. Scene Analysis, Xplore
Resolve
Digimetrics Aurora Automated video and audio test CGGV GeoVation Seismic Processing Multiple
PHYSICS
Pathwise™
flows Explicit solver Yes
DeltaGen, SDK for DeltaGen
Color grading Real-time color correction, real-time
and
algorithms (RTM, etc) Yes
APPLICATION DESCRIPTION SUPPORTED
Specialized platform for real-time hedging,
Vratis SpeedIT extreme
Yes
denoising
measurment
ffA Geoteric Seismic Interpretation Attributes
FEATURES MULTI-GPU SUPPORT
valuation, pricing and risk management
for OpenFOAM
PTC Creo Parametric Parametric design solution
Yes
Video encoding and video processing Single only
calculations, geobodies
Chroma Lattice Quantum Chromodynamics (LQCD) Spreadsheet-like modeling interfaces,
Solver library for general purpose CFD
suite Anti-aliasing, better lighting and enhanced
Cinnafilm Dark Energy Application and plug-in for
Elemental Live Live streaming video processing and extraction
Wilson-clover fermions, Krylov solvers,
Python-based scripting environment and
software
shaded-with-edges mode. GRID Support
image
encoding
Yes
Domain-decomposition
Grid middleware
Linear equation solvers Yes
Single only
enhancement
Video encoding and video processing Yes
ffA SVI Pro Seismic Interpretation Attributes
Yes
Yes
RESEARCH CFD DEVELOPMENTS
Siemens PLM NX and
Image de-noising and restoration.
Elemental Server File-based video processing and calculations, geobodies
ENZO 3D block-structured AMR code for
Hanweck Associates Real-time options analytical
FEFLO (GMU-Lohner) General purpose CFD
Teamcenter
Simultaneous active and background
encoding Video encoding and video processing Yes extraction
cosmological structure formation
engine (Volera) Real-time options analytics engine software for
Product lifecycle management solutions
rendering
isovideo Viarte Video standards conversion CUDA- Yes
Accelerated magneto hydrodynamics
Yes
compressible and incompressible flows
from design to simulation to production to
Yes
accelerated video procession and
ffA SEA3D Pro Seismic Interpretation Attributes
solvers
Murex MACS Analytics
Implicit and explicit solver Yes
service
Digital Vision Nucoda Color grading Real-time color encoding
calculations, geobodies
Yes
Library
SD++ (StanfordJameson)
Design software, NX, and PLM viewer
correction Single only
Yes
extraction
GTC Simulates microturbulence and transport in
Analytics library for modeling valuation and
General purpose CFD software for
applications, TcVis and Active Workspace
Marquise Technologies
MainConcept CUDA
Yes
magnetically confined fusion plasma
risk for derivatives across multiple asset
compressible flows.
Client.
Rain
H.264/AVC Encoder SDK
GeoStar Seismic Suite Seismic Processing Multiple
Electron push and shift (accounting for
classes.
Explicit solver Yes
Single only
Color grading CUDA-based real-time color correction H.264 video encoder Video encoding and video
algorithms (RTM, etc) Yes
>80% of run time)
Market standard models for all asset
S3D (Sandia and Oak
POPULAR GPU-ACCELERATED APPLICATIONS Single only
processing Yes
Headwave Suite Seismic Interpretation Attributes
Yes
classes paired with the most efficient
Ridge NL)
CATALOG | MAR14 | 09
Quantel Pablo Rio Color grading and finishing Real Sorenson Squeeze Video transcoding application andcalculations, Volume Rendering Yes
GTS Simulates microturbulence and the motion
resolution methods (Monte Carlo
Direct numerical solver (DNS) for turbulent
ELECTRONIC DESIGN AUTOMATION
time color correction Yes
plug-In Video encoding and video processing Yes
HUE HUEspace Seismic Interpretation Interpretation
of charged particles and interactions in
simulations and Partial Differential
combustion
APPLICATION DESCRIPTION SUPPORTED
Red Digital Cinema
Snell Alchemist on
development platform Yes
fusion plasma
Equations)
Chemistry model Yes
FEATURES MULTI-GPU SUPPORT
REDCINE-X
Demand
OpenGeo Solutions
Push and shift for both electron and ion
Yes
DualSPHysics SPH-based CFD software SPH model Agilent Technologies ADS Simulation tool for design Color grading CUDA-accelerated debayering Single Video standards conversion Video procession and
OpenSeis
dynamics
Numerical Algorithms
Yes
of RF, microwave
only
encoding Yes
Seismic Processing Spectral Decomposition Yes
Yes
Group (NAG)
NASA FUN3D General purpose CFD software Linear and high speed digital circuits
SGO Mistika Color grading and finishing Real-time Telestream Vantage Video transcoding and
Panorama Tech Seismic Processing, Modeling
MILC Lattice Quantum Chromodynamics (LQCD)
Random number generators, Brownian
equation solver Single only
Signal integrity simulation Single only
color correction and finishing Single only
processing Video encoding and video processing Yes Multiple algorithms (RTM, etc) Yes
codes simulate how elemental particles
bridges, and PDE solvers
08 | POPULAR GPU-ACCELERATED
Agilent Technologies
The Pixel Farm PFClean Image restoration and
ON-AIR GRAPHICS
Paradigm Echos RTM Seismic Processing RTM
are formed and bound by the “strong force”
Monte Carlo and PDE solvers Single only
APPLICATIONS CATALOG | MAR14
EMPro
remastering CUDA-based image processing
APPLICATION DESCRIPTION SUPPORTED
algorithm Yes
to create larger particles like protons and
RMS Catastrophic risk modeling for FSI
COMPUTATIONAL STRUCTURAL MECHANICS
Modeling and simulation environment for
acceleration
FEATURES MULTI-GPU SUPPORT
Paradigm SKUA Reservoir Modeling Faults, Horizons
neutrons
(earthquakes, hurricanes, terrorism,
APPLICATION DESCRIPTION SUPPORTED
analyzing 3D EM effects of high speed and
Single only
Avid On-air Graphics Motion Graphics Real-time
and Flow Simulation Grid Yes
Staggered fermions, Krylov solvers,
infectuous diseases)
FEATURES MULTI-GPU SUPPORT
RF/Microwave components
COMPOSITING, FINISHING AND EFFECTS
graphics rendering Single only
Paradigm Geophysical
Gauge-link fattening
Risk analytics Yes
Abaqus/Standard Simulation and analysis tool for
FDTD solver Yes
APPLICATION DESCRIPTION SUPPORTED
Brainstorm Estudio Virtual Sets and motion graphics VoxelGeo
Yes
Tanay ZX Lib (Fuzzy
structural
ANSYS Nexxim Circuit simulation engine for
FEATURES MULTI-GPU SUPPORT
Real-time graphics rendering Single only
Seismic Interpretation Volume Rendering, Horizon
PIConGPU A relativistic Particle-in-Cell code that
Logic)
mechanics
RF/analog/
ABSoft Neat Video Video noise reduction plug-in
Chyron HyperX On-air graphics Real-time graphics Flattening Yes
describes the dynamics of a plasma by
Financial analytics and data mining library Monte
Direct sparse solver Yes
mixed-signal IC design; IBIS-AMI analysis
CUDA-based acceleration Single only
rendering Single only
Roxar RMS Reservoir Modeling Multi GPU
computing the motion of electrons and ions
Carlo simulations, pricing of vanilla
ANSYS Mechanical Simulation and analysis tool for speedup with GPU computing.
Adobe After Effects CC Motion graphics and effects Chyron LEX On-air graphics Real-time graphics
capabilities via HUEspace Yes
subject to the Maxwell-Vlasov equation.
and exotic options, fixed income analytics,
structural
AMI analysis Single only
3D raytracing engine based on NVIDIA
rendering Single only
Seismic City Prestack
Simulation of laser-wakefield acceleration
data mining
mechanics
ANSYS HFSS Simulation tool for modeling 3-D full- OptiX
Harris Inscriber On-air graphics Real-time graphics Interpretation
of electrons
Yes
Direct and iterative solvers Yes
wave
Yes
rendering Single only
Seismic Processing Multiple algorithms (RTM, etc)
Yes
SciComp, Inc Derivative pricing (SciFinance) Monte Impetus Afea Predicts large deformations of
electromagnetic fields in high-frequency
Autodesk Flame
Monarch dScript 3D Virtual sets Real-time graphics Yes
QUDA Library for Lattice QCD calculations using
Carlo and PDE pricing models Single only
structures
and high-speed electronic components.
Premium
rendering Single only
SpectraSeis Seismic Processing Full elastic waveGPUs
POPULAR GPU-ACCELERATED APPLICATIONS and components exposed to extreme
Transient solver Yes
Finishing and color grading Post-production
Monarch Twister Pro Virtual Sets and motion
equation imaging and
CUDA supports the following fermion
CATALOG | MAR14 | 07
loading conditions
CST Microwave Studio
integrated end-to-end
graphics Real-time graphics rendering Single only
analysis of microseismic fracture data
formulations: Wilson,Wilson-clover,Twisted
Xcelerit SDK Software Development Kit (SDK) to
Linear equation solver Yes
(MWS)
toolset for 3D VFX, editorial, and color
Monarch Virtuoso Virtual Sets and motion graphics Yes
mass,Improved staggered (asqtad or HISQ)
boost
LS-DYNA Implicit Multiphysics simulation package 3D EM modeling and simulation Transient solver
grading
Real-time graphics rendering Single only
Stoneridge Technologies
and Domain wall
the performance of Financial applications
used Linear equation solver Yes
Yes
Single Only
Pixel Power Clarity On-air graphics Real-time
GAMPACK
Yes
(e.g. Monte-Carlo, Finite-difference) with
MSC Nastran Simulation and analysis tool for
Delcross Savant Simulation tool for installed antenna Autodesk Smoke Finishing and editing Faster effects graphics rendering Single only
Reservoir Simulation GPU Algebraic MultiGrid
RAMSES Simulates astrophysical problems on
minimum changes to existing code
structural
performance and antenna-to-antenna
Single only
RT Software tOG On-air graphics Real-time graphics Package Yes
different scales (e.g. star formation,
C++ programming language, crossplatform (back-endmechanics
coupling
Boris FX Continuum
rendering Single only
Terraspark Insight Earth Seismic Interpretation
galaxy dynamics, cosmological structure
generates CUDA and
Direct sparse solver Yes
High-frequency solver Yes
Complete
Vizrt Viz Engine On-air graphics Real-time graphics Horizon orientation attributes; automated
formation)
optimized CPU code), supports Windows
MSC Marc Simulation and analysis tool for structural EMSS FEKO 3D EM modeling and simulation MoM Visual effects plug-in Faster effects Single only
rendering Single only
fault extraction, Curvature Attributes
CUDA acceleration is applied for radiative
and Linux operating systems.
mechanics
solver Single only
Cinnafilm Dark Energy
Wasp 3D Beehive On-air graphics and virtual sets
Yes
transfer for reionization, and the
Yes
Direct sparse solver Yes
Remcom XFdtd 3D EM modeling and simulation
Plug-in
Real-time graphics rendering Single only
Tsunami RTM Seismic Processing RTM algorithm
hydrodynamic solver using AMR
Synerscope’s Synerscope
OptiStruct Simulation and analysis tool for structural FDTD solver Yes
Color management All features accelerated with
12 | POPULAR GPU-ACCELERATED
Yes
Yes
Data Visualization
mechanics
SPEAG SEMCAD-X 3D EM modeling and simulation CUDA Yes
APPLICATIONS CATALOG | MAR14
WesterGeco Omega2
Chemora Chemora is a system for performing
Visual big data exploration and insight tools Graphical Direct and iterative solvers Yes
FDTD solver Yes
CoreMelt complete Visual effects plug-in Faster
ON-SET, REVIEW AND STEREO TOOLS
RTM
simulations of systems described
exploration of large network
DS Exsight Uses Abaqus Standard for GPU
Rocketick RocketSim Verilog simulation Verilog
effects Single only
APPLICATION DESCRIPTION SUPPORTED
Seismic Processing Multiple algorithms (RTM, etc)
by differential equations running on
datasets including geo-spatial and
computing Direct sparse solver Single only
simulation Yes
eyeon Fusion Effects and compositing Faster effects FEATURES MULTI-GPU SUPPORT
Yes
accelerated computational clusters
temporal components
DS DesignSight Uses Abaqus Standard for GPU
Media and Entertainment
Single only
3ality Intellicam 3D stereo camera adjustment CUDA-For more information on GPU-accelerated
Chemora embeds the equations’
Single only
computing Direct sparse solver Single only
ANIMATION, MODELING AND RENDERING
GenArts Monsters GT Visual effects plug-in Faster based imaging Single only
applications please visit,www.nvidia.com/teslaapps
computational kernels into dynamically
QuantAlea’s Alea.cuBase
PAM-CRASH Implicit Multiphysics simulation
APPLICATION DESCRIPTION SUPPORTED
effects Single only
BlueFish 4K Review Review and approval of 4K
© 2014 NVIDIA Corporation. All rights reserved.
compiled loop nests shaped for input size
F#
package used Linear equation solver Single only
FEATURES MULTI-GPU SUPPORT
GenArts Sapphire Visual effects plug-in Faster
content Real-time Single only
NVIDIA, the NVIDIA logo, and CUDA, are
and GPU structure
F# package enabling a growing set of F#
NX Nastran Simulation and analysis tool for structural Autodesk 3ds Max +
effects Single only
Colorfront On-Set Dailies Review and approval of 4K trademarks and/or registered trademarks of
Yes
capability to run on a GPU.
mechanics
NVIDIA iray
ROBUSKEY Chroma keyer plug-in Faster effects
content Real-time Yes
NVIDIA Corporation. All company and product names
06 | POPULAR GPU-ACCELERATED
F# for GPU accelerators Yes
Linear equation solver Single only
3D modeling, animation, and rendering iray
Single only
eyeon Dimension 3D stereoscopic workflow Realare trademarks or registered trademarks of the
APPLICATIONS CATALOG | MAR14
Altimesh’s Hybridizer C# Multi-target C# framework COMPUTER AIDED DESIGN
interactive, photorealistic and
Neat Video Open FX Video noise reduction plug-in time Single only
respective owners with which they
Defense and Intelligence
for data parallel
APPLICATION DESCRIPTION SUPPORTED
physically correct rendering
Faster effects Single only
Lightcraft Previzion On-set workflow Real-time Single are associated. Features, pricing, availability, and
APPLICATION DESCRIPTION SUPPORTED
computing.
FEATURES MULTI-GPU SUPPORT
Yes
NewBlueFX Video
only
specifications are all subject to change without
FEATURES MULTI-GPU SUPPORT
C# with translation to GPU or Multi-Core
AutoCAD 2D and 3D CAD design, drafting, modeling, Autodesk Maya 3D modeling, animation, and
Essentials
Tweak RV Review and approval of 4K content Real- notice. MAR14
DigitalGlobe Advanced
Xeon
architectural drawing, and engineering
rendering Increased model complexity, larger scenes,Video effects plug-in Faster effects Single only
time Single only
Ortho Series
Yes
software. Supports Open GL. Native DWG™
windowed stereo
NewBlue Titler Pro Video titling plug-in Faster effects SIMULATION
Geospatial Visualization Image orthorectification Yes MISYS Global Risk Regulatory compliance and
support.
Single only
Single only
APPLICATION DESCRIPTION SUPPORTED
Eternix Blaze Terra Geospatial Visualization 3D
enterprise wide
Surface, mesh, and solid modeling tools,
Autodesk Motion Builder Character animation and
Pixelan AnyFX Video effects plug-in Faster effects FEATURES MULTI-GPU SUPPORT
visualization of geospatial data Yes
risk transparency package
model documentation tools, parametric
motion capture Increased model complexity at
Single only
3DAliens Glu3d SPH fluid simulation Faster
Exelis (ITT) ENVI Geospatial Visualization Image
Risk Analytics Yes
drawing capabilities. Native DWG™
interactive
Re:Vision Effects Visual effects plug-in Faster effects simulation Single only
orthorectification
MiAccLib High Speed Multi-Algorithm Search Engine support.
rates
Single only
Blastcode Kilton/
(custom builds only)
library providing high speed text string
Single only
Single only
Red Giant Effects Suite Visual effects plug-ins FasterMegaton
Yes
search with scalability of searching text
AutoCAD Design Suite AutoCAD 2014 software, plus Autodesk Mudbox 3D sculpting Increased model
effects Single only
Physics-based simulation plug in Faster simulation
MrGeo Geospatial Visualization Terrain analytics Yes and/or keywords on hundreds millions of
tools to create,
complexity at interactive
Red Giant Magic Bullet
Single only
GeoWeb3d Desktop Geospatial Visualization 3D
records and/or text data.
capture, connect, and showcase designs
rates
Looks
Jawset TurbulenceFD Physics-based simulation plug
visualization of geospatial data Yes
Exact Text match Search, Approximate\
2D/3D display of designs, interactive 3D
Single only
Color and finishing tools Faster effects Single only
in Faster simulation Single only,
Incogna GIS Geospatial Visualization Image
Similarity Text Search, Wild Card
presentation with realistic materials,
Cebas finalRender GPU Renderer CUDA-based GPUThe Foundry HIERO Shot management, conform
multi-GPU
processing on Tesla cloud servers
Text Search, Proximity & Percentage
rendering-ray tracing
Rendering Yes
and review
available in
Object recognition
Text Search, MultiKeyword and
Single only
CentiLeo GPU Render GPU Renderer CUDA-based timeline
late Q2/Q3
Yes
MultiColumnMultiKeyword Text Search,
Autodesk 3ds Max 3D animation creative toolset for GPU Rendering Single only
Better interactivity Single only
WEATHER AND CLIMATE FORECASTING
Intergraph Motion Video
RadixSort Text
modeling,
Chaos V-Ray RT GPU Renderer CUDA interactive The Foundry NUKE and
APPLICATION DESCRIPTION SUPPORTED
Analyst
Yes
animation, simulation, and rendering for
and final-frame GPU
NUKEX
FEATURES MULTI-GPU SUPPORT
Video filters and mosaic’ing - Geo-fuses
SunGard Adaptiv
games, film, and motion graphics
Rendering
Compositing tools with 3D tracker Faster effects
Accuweather Cinemative
FMV analytics with intelligence data
Analytics
3D modeling, mesh and surface modeling,
Yes
Single only
HD
Full motion video ortho mosaic processing Single
A flexible and extensible engine for fast
improved Nitrous viewport performance,
Jawset TurbulenceFD Physics-based simulation plug-Video Copilot Software
Weather graphics Real-time Single only
6
Software Development Ecosystem
for Intel XeonOpen
PhiSource Commercial
Compiler
gcc (kernel build only, not
for applications),
Python
Intel Parallel Studio XE (C++ & Fortran)
Intel Cluster Studio XE (C++ & Fortran)
CAPS HMPP compiler (beta)
Debugger
gdb
Intel Debugger
Rogue Wave TotalView (beta)
Allinea DDT
Libraries
TBB (in Intel Studio XE)
MPICH2
FFTW
NetCDF
Intel MKL, Intel MPI, OpenMP, Intel IPP,
Cilk™ Plus (in Intel Studio XE products),
NAG
Rogue Wave IMSL
Profiling & Analysis
Tools
Intel VTune Amplifier XE
Intel Trace Analyzer & Collector
Intel Inspector XE
Workload Scheduler
Altair PBS Professional,
Adaptive Computing Moab
System Management
SGI Management Center
Bright Cluster Manager (beta)
For more information:
http://software.intel.com/en-us/articles/intel-and-third-party-tools-and-libraries-available-with-support-for-intelr-xeon-phitm
7
Next generation XEON: Haswell
8
Next-next generation HPC XEON
PHI
9
“Pyramid 1U”
Model
C1104G-RP5
Chassis Profile
1U standard-depth
Servers/System
One dual-socket
Chipset
Intel® C600
Max. Processors
Two Intel® Xeon® processor E5-2600
series
Max. CPU TDP
115W
Memory Slots
8 DIMM slots
Memory Type
1600/1333/1066/800 MHz DDR3 ECC
Reg
Max. Hard Disk Drives
4 x 2.5" drives
Expansion Slot
Three PCIe 3.0 x16
double-width slots
One external PCIe 3.0 x8 low-profile slot
Networking (Onboard) Dual-Port GigE controller (Intel® I350)
NVIDIA GPU accelerator support:
• 3x K10, K20, K20X
NVIDIA graphics GPU support:
• Quadro Plex 7000
Intel accelerator support:
• 3x Phi
10
IPMI Remote
Management
Integrated IPMI 2.0
Power Supply
1800W Redundant* Platinum Level
10
“Pyramid 2U”
Model
C2110G-RP5
Chassis Profile
Servers/System
2U standard-depth
One dual-socket
Chipset
Intel® C600
Max. Processors
Two Intel® Xeon® processor E5-2600 series
Max. CPU TDP
Memory Slots
130W
8 DIMM slots
Memory Type
1600/1333/1066/800 MHz DDR3 ECC Reg
Max. Hard Disk Drives
10 x 2.5” drives
Expansion Slot
NVIDIA GPU accelerator support:
• 4x K10, K20, K20X
NVIDIA graphics GPU support:
• Quadro Plex 7000
Intel accelerator support:
• 4x Phi
Four internal PCIe 3.0
x16 double-width
(optional riser card for 2 additional external
PCIe 3.0 x16 double width slots),
One external PCIe 3.0 x8 low profile and one
PCIe 2.0 x4 full height
Networking (Onboard)
Dual-Port GigE controller (Intel® I350)
IPMI Remote Management
Integrated IPMI 2.0
Power Supply
1800W Redundant** Platinum Level
11
11
SGI UV 2000
Within a Single Standard 19” Rack
Up to:
• 64 socket/512 cores/1024 threads
®
®
- Or 34 CPU + 30 Intel Xeon Phi™
- Or 36 CPU + 28 NVIDIA® Tesla®
K20,K20X, or K40 (2 partitions)
• 16TB memory
• 63 x16 PCIe Gen 3 Links
12
12
SGI UV and Scale-out Compared
Feature
Standard Scale-out Servers
SGI UV
Architecture
Reference Terms
Scale Out
Cluster
Distributed Memory
Scale Up
Single System Image
Shared Memory
System Limit
16 cores, 0.5 TB memory
2048 cores, 64 TB memory
CPU
x86
Intel® Xeon® or AMD Opteron™
X86
Intel® Xeon®
Memory, Storage,
networking
Industry standard
Industry Standard
Interconnect or
System Fabric
Ethernet or InfiniBand
SGI NUMAlink
Hardware Package or
Building Block
Blades or Rackmount, 19” rack
Blades (UV2000) or
Rackmount(UV20), 19” rack
Software
Off the Shelf
Off the Shelf
Applications
Small scale single apps
cluster or “MPI” apps
Small or large single apps
Cluster or “MPI” apps also
Cost
Lowest cost x86 architecture
20-75% > scale-out,
~1/3 the cost of ‘Big Iron’.
13
13
SGI® Value Propositions
• Long history of working with accelerators
•
•
•
“Home-brewed” – Geometry Engine, TPU (Tensor Processor Unit)
FPGA’s (RASC™ technology), GPUs, SOCs
50 application experts focused on it
• Accelerators in both a scale-up and scale-out environment
•
•
32 Intel® Xeon® Phi™ coprocessors in SGI® UV™ 2000 (COSMOS and
TGAC)
SGI: the only vendor able to deliver hybrid scale-out and scale-up solutions
• Everything you need in a powerful solution
•
•
•
Factory-integrated, tested, rack level delivery – plug in and go
Starter kits
Worldwide customer support
• Fully-managed, with SGI Management Center and Performance
Suite software
14
14
SGI® Integrated Clusters for Intel
Xeon Phi
A complete, managed GPU solution of software and
hardware, all you need for a powerful deployment
• Hardware stack
•
Rackable™ C2108 head node
•
Either/or Rackable C1104G or C2110G compute nodes
•
Infiniband NICs and Switch
•
Ethernet switch for management
•
Add SGI InfiniteStorage solutions, either direct-attached to the head node or SGI
NAS
Software stack
•
SGI Management Center software
•
Altair PBSpro Load Management software
Complete Factory Test & Integration
15
SGI Management Center
®
Premium Edition offers Accelerator Monitoring and Management!
• Single system management console with remote server monitoring
and control
• Ease of use, full system management including
GUI and
CLI
• Policy driven, fine grained power load management
for green
computing
• Advanced fault, event,
and alert management
for
improved reliability
• Advanced capabilities including Accelerator management, BIOS
management, and high availability
16
SMC Premium Edition– Accelerator Monitoring
Detailed GPU
reporting
enables high
performance
systems to
remain at
their peak
performance
17
Optimized for HPC & Big Data
Solutions
• SGI ICE X
• SGI UV™ 2000
®
™
18
SGI ICE X Compute Node with GPU
• Two Intel® Xeon® processor E5-2600 series
• Two
NVIDIA® TESLA® K20
• FDR InfiniBand - single or dual plane
• Four DDR3 DIMMs per socket @ 1600 MT/s
• Up to one 2.5” SATA HDD/ SSD drives
• Liquid cold sinks
One dual socket node w/
two coprocessors in
one blade slot!
FDR + 1:1 processor to coprocessor ratio
= Balanced Throughput
19
19
SGI ICE X Compute Node with Phi
• Two Intel® Xeon® processor E5-2600 series
• Two Intel® Xeon Phi™ 5120D Coprocessor
• FDR InfiniBand - single or dual plane
• Four DDR3 DIMMs/ socket @ up to 1600 MT/s
•
Expect 1866 MT/s support w/1 DPC populated (w/ Ivy
Bridge-EP)
• Up to one 2.5” SATA HDD/ SSD drive
• Liquid cold sinks
One dual socket node w/
two coprocessors in
one blade slot!
FDR + 1:1 processor to coprocessor ratio
= Balanced Throughput
Balanced
20
20
throughput
SGI Cold Sink Technology – Twin Blades
21
SGI Meets All Accelerator Needs!
• SGI has the longest history with
accelerators
• Accelerators have gone mainstream, we are
supporting them across our product line
• The only vendor to be able to deploy scaleup, scale-out and hybrid landscapes with
accelerators everywhere.
22
22
SGI ICE: Sample ICE Customers
BAW - Bundesanstalt für
Wasserbau
Bridgestone
Central Research Inst of
Electric Pwr Ind
CSIR - Centre for
Mathematical Modeling and
Compute
Exeter University
GENCI
HLRN Berlin
HLRN Hannover
ICHEC
ICR
Idaho National Laboratory
IFREMER
Imperial College
IMSc
INMET
JAMSTEC
Korean Air Force
KU Leuven
Mazda Motor Corporation
McLaren Motor Racing Ltd
Mercedes-Benz GP
MPO
NASA Ames
NASA Langley
NIIFI
NIMS
NMCAC
NOAA (CSC)
NTNU
Onera
ORNL
Pontificia Universidade
Catolica – PUC-Rio
Rosshydromet
Scientific Analysis Group
Semiconductor Energy Labs
23
Sikorsky
Skoda Auto
TI-09
Tokyo University – ISSP
Total
Toyota
U.S. Air Force - Arnold AFB
U.S. Navy/NRL
Universidade Catolica
Universite Paul Sabatier –
CICT/Calmip
University of Arizona
University of Hyderabad
University of Oxford
University of São Paulo –
USP/IAG
University of Rio de Janeiro –
NACAD
University of Rio Grande do
Sul – CESUP
US Army TACOM
Hungary 2011-2014
NIIFI supercomputers for non-profit customers:
• Cluster:
SGI ICE
Debrecen (18TFlops, 1536 core; 6TB ram)
• Cluster + GPU: HP
Szeged (14TFlops, 2304 core; 5,6TB ram, 6xM2070)
• SMP/ccNUMA: SGI UV
Pécs (10TFlops, 1152 core, 6TB ram)
http://www.niif.hu/szolgaltatasok/szuperszamitastechnika/altalanos_ismerteto
24
Hungary 2015
• Cluster:
200+ Tflops
(200pc GPU or PHI)
• Cluster:
30+ GPU or PHI
• Cluster:
SGI ICE
Debrecen (18TFlops, 1536 core; 6TB ram)
• Cluster + GPU: HP
Szeged (14TFlops, 2304 core; 5,6TB ram, 6xM2070)
• SMP/ccNUMA: SGI UV
Pécs (10TFlops, 1152 core, 6TB ram)
25
©2012 SGI
26
26
