TESLA

  • What is GPU Computing
  • GPU Applications
  • Servers and Workstations
Subscribe
GPU Applications

ANSYS

Computer-aided engineering (CAE) simulations enable engineers to design more virtual prototypes, thereby spending less time building physical ones. More iterations lead to a higher quality product. GPUs further improve productivity by accelerating simulations, thus helping reduce the product development times resulting in a competitive advantage for businesses.


ANSYS® and NVIDIA® work collaboratively to ensure the fastest performance for simulations running on ANSYS parallel software. NVIDIA GPUs support the following products:
  • ANSYS® Mechanical® supports a full feature set.
  • ANSYS® Fluent® FLUENT® offers GPU support for pressure-based coupled solver and for radiation heat transfer models. ANSYS Polyflow supports models for polymer and glass processing with as many as 10 million elements through use of GPUs since version 16.0.
  • ANSYS HFSS and HFSS SBR+ solvers offer GPU support for transient electromagnetic and antenna performance simulations.
  • ANSYS® Nexxim™ supports a full feature set; ANSYS® HFSS™ supports transient solver in 15.0.

Accelerating ANSYS simulations with NVIDIA GPUs have never been easier. Since version 15.0, all HPC license products (HPC, HPC packs, HPC workgroups, and HPC enterprise) enable GPUs. Specifically, each GPU socket is treated the same as a single CPU core with respect to license requirements. As a result, more simulations are possible with existing HPC licenses combined with NVIDIA GPUs for dramatically increasing simulation productivity.

ANSYS® Mechanical™ 17.0 provides significant performance speedups when using NVIDIA Quadro K6000 or the Tesla K40 active GPUs in workstations. Similar performances are seen with Tesla K80 boards in servers. See the performance results on a standard ANSYS Mechanical benchmark below. To learn more, please read the ANSYS Tech Tip article "Accelerating Mechanical Solutions with GPUs"


ANSYS Mechanical
Distributed ANSYS Mechanical 17.0 on 2x Haswell-EP (Xeon E5-2698 v3 2.3 GHz) 16-core CPU, Quadro K6000 GPU, 256 GB RAM; Static, Nonlinear Structural Analysis of a Turbine, 3.2M DOF; double-precision direct sparse solver
"ANSYS Mechanical users can get higher throughput on large models with an NVIDIA Tesla K40 in their workstations. Models as large as 12M degrees of freedom can be accelerated with a single K40 and provide 2X performance gains over just CPUs."

Ray Browell
Lead Product Manager
ANSYS, Inc.

ANSYS® Fluent® 17.0 features multi-GPU support to deliver higher productivity in CFD simulations. This performance enhancement is the result of innovative GPU-accelerated solvers developed by NVIDIA in collaboration with ANSYS (called AmgX) and the license scheme in ANSYS. The implementation in Fluent accelerates pressure-based coupled flow solver, speeding up the flow portion of CFD simulations. This benefit can simply be unlocked by adding Tesla K80 GPUs to any existing HPC infrastructure. It is designed to run across multiple nodes with multiple GPUs in a cluster configuration, just like CPU systems. Please refer to the NVIDIA application note on Accelerating ANSYS Fluent 15.0 Using NVIDIA GPUs for more details on using GPUs in Fluent 15.0.

The chart below compares the performance of Tesla K80 GPU boards(2 K80 GPUs per board) versus CPUs in ANSYS Fluent 17.0 of a landing gear model run on a single node server.


ANSYS Mechanical
ANSYS Fluent 17.0 with AmgX on 2x Haswell-EP (Xeon E5-2698 v3 2.3 GHz) 16-core CPU, Tesla K80 boards, 256 GB RAM; External flow over aircraft landing gear with LES & acoustics, 15m cells; unsteady, double-precision Pressure-based coupled solver
 

RECOMMENDED PROFESSIONAL PRODUCTS


The powerful GPU computing capabilities in ANSYS was developed on NVIDIA Professional GPUs. A CUDA-capable NVIDIA GPU (CUDA v4.2 or above) should be used. Suggested GPUs from the NVIDIA® Tesla® Kepler™ product family or NVIDIA® Quadro® Kepler product family are recommended.

NVIDIA GPU computing products are designed to deliver the highest computational performance to accelerate computations in ANSYS software.

Tesla Benefits

Highest Computational Performance
> High-speed double precision operations
> Large dedicated memory
> CUDA-Accelerated math libraries
> High-speed bi-directional PCIe communication
> NVIDIA GPUDirect™ with InfiniBand
Most Reliable
> ECC memory
> Rigorous stress testing
> ANSYS QA testing
Best Supported
> Professional support network
> Large and growing developer community
> OEM system integration
> Long-term product lifecycle
> 3 year warranty
> Cluster & system management tools
   (server products)
> Windows remote desktop support
 

Recommended Configurations

Servers/Clusters
> New System Purchase

ANSYS Fluent:
Each node with
    > 2 CPU sockets (up to 16 CPU cores each); recommend 2 Tesla K80 boards per CPU socket
    > Total 128 to 256 GBs of CPU memory
    > Disk with minimum 2000 GB (scratch)
ANSYS Mechanical:
Each node with
    > 2 CPU sockets (8-32 CPU cores total); recommend 1 Tesla K80 board per CPU socket
    > Total 128 to 256 GBs of CPU memory
    > Disk with minimum 2000 GB (scratch)
Workstations
> New System Purchase

ANSYS Mechanical:
    > 2 CPU sockets (8-32 CPU cores total); recommend 1 GPU per CPU socket
    > Total 48 GBs of CPU memory or more;
    > Disk with minimum 500 GB
    > SSD with minimum 1000 GB
> GPU OPTIONS
    > Tesla K40
    > Quadro K5200/M6000 for pre/post

    OR a single GPU

    > Quadro K6000 for compute and pre/post

NVIDIA Tesla and Quadro products are available from all major professional workstation OEMs. Only Tesla GPU computing products are designed and qualified for compute cluster deployment.

BUY OPTIMIZED TESLA SYSTEMS


NVIDIA partners with system vendors to provide optimal solutions that accelerate your HPC workloads. Buy now and enjoy all the benefits of GPU-acceleration on ANSYS.

For a complete list of Tesla Preferred Providers, go to the Tesla Where to Buy page.

Useful Links:

ANSYS 17.2 capabilities chart
ANSYS 17.2 GPU Accelerator and co-processor capabilities
GPUs speed the solution of complex electromagnetic simulation with ANSYS HFSS
Installed Antenna Performance Modeling on Electrically Large Platforms