InfiniBand Enables the Most Efficient Machine Learning Platforms
Machine learning is a pillar of today’s technological world, offering solutions that enable better
and more accurate decision-making based on the great amounts of data being collected. Machine
learning encompasses a wide range of applications, ranging from security, finance, and image and
voice recognition, to self-driving cars, healthcare and smart cities.
InfiniBand accelerates all popular frameworks such as TensorFlow, CNTK, Paddle, Pytorch
and Apache Spark with RDMA, and continues to innovate and accelerate solutions for fastest and most
scalable distributed execution of training large and powerful models.
By providing low latency, high bandwidth, high message rate, and smart offloads, InfiniBand solutions are the most deployed high-speed interconnect for large-scale machine learning
- for both training and inferencing systems.
The Selene Supercomputer
NVIDIA’s DGX SuperPOD with NVIDIA Mellanox HDR 200Gb/s InfiniBand Deployment
One of the fastest and most efficient supercomputer on the planet
built in under one month
Maximizing Data Center Storage and Network IO Performance with NVIDIA Magnum IO
Magnum IO utilizes storage IO, network IO, in-network compute, and IO management to simplify and speed up data movement, access, and management for multi-GPU, multi-node systems.
Magnum IO supports NVIDIA CUDA-X™ libraries and makes the best use of a range of NVIDIA GPU and NVIDIA networking hardware topologies to achieve optimal throughput and low latency.
Learn More
SETTING A NEW BAR IN MLPERF
NVIDIA training and inference solutions deliver record-setting performance in MLPerf, the leading industry benchmark for AI performance.
NVIDIA RAPIDS Accelerator and InfiniBand Supercharges Apache Spark 3.0
As data scientists shift from using traditional analytics to AI applications with Apache Spark 3.0, InfiniBand provides the fastest path to move data between remote Spark nodes and is tightly coupled to the Spark Shuffle operations leveraging the UCX communication framework.
Learn MoreSHARPv2 Delivers Highest Performance for AI
An autoencoder
is a neural network that learns to copy its input to its output. It has an internal
(hidden) layer that describes a code used to represent the input, and it is
constituted by two main parts: an encoder that maps the input into the code, and a
decoder that maps the code to a reconstruction of the original input.
Variational autoencoders, use a variational approach for latent representation
learning and allow for the design of complex generative models of data, and fit
them to large datasets. Some interesting examples include the generation of images
of fictional faces, perhaps for on-line customer service or professors in an online
educational institute, or event high-resolution digital artwork.
Mellanox SHARPv2 technology improves the scalable performance of VAE significantly.
SHARPv2 Performance Advantage
Many parallel
applications will require accessing the reduced results across all processes rather
than the root process. In a similar complementary style of MPI_Allgather to
MPI_Gather, MPI_Allreduce will reduce the values and distribute the results to all
processes.
Latency as well as network bandwidth is important for many AI workloads including
NLP. InfiniBand with Mellanox SHARP technology dramatically reduces latency for
messages that are typically larger in size. Even RDMA over Converged Ethernet is no
match for the in-network computing capabilities of SHARP.
SHARPv2 Delivers Highest Performance for AI
Google Neural Machine Translation (GNMT) is a neural machine translation (NMT)
system developed by Google that uses an artificial neural network to increase
fluency and accuracy in Google Translate.
GNMT improves on the quality of translation by applying an example-based (EBMT)
machine translation method in which the system "learns from millions of examples“
The tightly coupled integration between the new NCCL® + SHARP plugin from HPC-X™ demonstrates exceptional performance for GNMT
10X Performance with NVIDIA Mellanox GPUDirect® RDMA
Designed specifically for the needs of GPU acceleration, GPUDirect RDMA provides direct communication between NVIDIA GPUs in remote systems. This eliminates the system CPUs and required buffer copies of data via the system memory, resulting in 10X better performance.
NVIDIA Mellanox Unified Fabric Management
Maximizing Performance for Distributed Machine and Deep Learning with SHARP
- NVIDIA Mellanox Product DocumentationView
- NVIDIA Mellanox UFM PlatformDownload
- NVIDIA Mellanox MetroX-2 SystemsDownload
- QM8700 Mellanox Quantum™ HDR Edge SwitchDownload
- QM8790 Mellanox Quantum™ HDR Edge SwitchDownload
- CS8500 Mellanox Quantum™ HDR Modular SwitchDownload
- ConnectX-6 HDR 200Gb/s VPI AdapterDownload
- LinkX® InfiniBand Cables and TransceiversView
- BlueField-2Download
- BlueFieldDownload
- Innova SmartNICDownload