To meet the growing demand for 10x growth in deep learning, artificial intelligence, and big data analytics, Supermicro has expanded its GPU-optimized system line.
NVIDIA’s Tesla V100 PCI-E and V100 SXM2 Tensor Core GPU accelerators will be showcased by Super Micro Computer, Inc. (NASDAQ: SMCI), a world leader in virtualization, storage, networking solutions, and green computing technology, at its Platinum Sponsor Booth at the GTC Taiwan 2018 Taipei Marriott Hotel on May 30-31.
New 4U system with NVIDIA NVLinkTM interconnect technology targeted for an optimum acceleration of highly parallel applications, including AI, deep learning, self-driving vehicles, smart cities, health care, big data, high-performance computing (HPC), virtual reality, and more. A GPU-accelerated server platform from NVIDIA, the SuperServer 4029GP-TVRT offers eight NVIDIA Tesla V100 32GB SXM2 GPU accelerators for cluster and hyper-scale applications with maximum GPU-to-GPU connectivity. Thermal zones for the GPU and CPU have been separated to ensure uncompromising performance and stability under very demanding workloads, using NVIDIA NVLink technology with over five times the bandwidth of PCIe 3.0.
Chief executive officer and president of Supermicro Charles Liang said the 4029GP-TVRT system could produce 5,188 frames per second on ResNet-50 workloads and 3,709 images per clock cycle on InceptionV3 workloads in preliminary internal benchmark tests. Scaling to many systems using GPU Direct RDMA also yields amazing, practically linear performance gains.” These new GPUs with 2X memory integrated into performance-optimized 1U and 4U systems via next-generation NVLink allow our clients to expedite their applications and innovations to help tackle some of the world’s most complex and challenging issues.
“Supermicro’s high-density servers tailored for NVIDIA Tesla V100 32GB Tensor Core GPUs will benefit enterprise customers,” said Ian Buck, vice president and head of NVIDIA’s accelerated computing. Up to 50%, faster results on complicated deep learning and biomedical applications can be achieved because V100s double the memory, which reduces the need to optimize for memory.
As a result, the Tesla P4 can be used in any scalable server with the support of Supermicro GPU systems. It is possible to integrate deep learning into the video transcoding pipeline thanks to Tesla P4’s hardware-accelerated transcode engine, enabling a new class of intelligent video applications. To train ever-more sophisticated neural networks on ever-larger datasets, we must use deep learning, a computer model unlike any other. Using powerful Supermicro GPU servers, these models can give the most throughput possible for inference workloads.
NVIDIA SCX-E3-class GPU-Accelerated Server Platforms, including the speed 4U, SuperServer 4029GR-TRT2 system, can run up to 10 PCI-E NVIDIA Tesla V100 accelerators using Supermicro’s revolutionary and GPU-optimized slightly more complex PCI-E design, significantly enhances GPU peer-to-peer communication performance. With the new SuperServer 1029GQ-TVRT, you can have four NVIDIA Tesla V100 SXM2 32GB GPU accelerators in 1U of rack space, while the older SuperServer 1029GQ-TRT only supports two. GPU-accelerated server platforms of the NVIDIA SCX-E2 family include both of the 1029GQ servers.
Now that big data analytics and deep learning have converged, as have NVIDIA GPU architectures and the latest machine learning techniques, deep learning applications necessitate GPU networks with more processing power and better communication. The NCCL P2PBandwidthTest shows that Supermicro’s single-root GPU technology allows many NVIDIA GPUs to interact efficiently to decrease latency and maximize throughput.
An all-in-one machine learning solution from Supermicro.
Canonical and Supermicro have teamed up to offer TensorFlow machine learning solutions.
It is constructed and proven using high-performance, high-reliability, high-quality, and scalable Supermicro SuperServers, SuperStorage systems, and Supermicro Ethernet switch technology.
Ubuntu’s maker, Canonical, assists businesses in making the most of the operating system. Pure upstream Kubernetes (CDK) has been tested on many cloud platforms. Canonical also provides a broad ecosystem of tools, libraries, services, current metrics, and monitoring tools to make CDK easier to consume so that you can innovate quicker. Canonical.com.
On top of a Kubernetes cluster, Kubeflow provides convenient Machine Learning (ML) resources. TensorFlow’s installation is made simpler with Kubeflow, which also provides ways for utilizing GPUs attached to the base host during the execution of ML workloads. Open source TensorFlow is a high-performance numerical calculation software package. Because of its adaptable design may be used on a wide range of hardware platforms (including PCs, servers, mobile devices, and the edge), including CPUs, GPUs, and TPUs.
Rackmount Workstations from Supermicro are the best in the busines.
NVIDIA A100TM Tensor Core GPUs are used in Supermicro’s highest-performance, fastest-to-market systems. Customized configurations for the new HGXTM A100 8-GPU and HGXTM A100 4-GPU platforms are available from Supermicro. The latest NVIDIA® NVLinkTM and NVIDIA NVSwitchTM technologies enable these systems to perform up to 5 PetaFLOPS in a single 4U system.
4U system with HGX A100 8-GPU HGX A100
It’s ideal for sizeable deep learning exercises and neural network modeling projects. Supermicro’s unique AIOM support for the AIOM support of the new 4U GPU design systems, the NVIDIA HGX A100 8-GPU baseboard, six NVMe U.2 and two NVMe M.2 I/O, 10 PCI-E 4.0 x16 I/O, and NVIDIA NVLink and NVSwitch, GPUDirect RDMA, GPUDirect Storage and NVMe-oF on InfiniBand for 8-GPU communication and data flow between systems.
A system with HGX A100 4-GPUs in a 2U form factor
NVIDIA A100 Tensor Core GPU and the HGX A100 4-GPU baseboards are included in the AS -2124GQ-NART server. Fast CPU-GPU connections and high-speed networking extension cards are supported by PCI-E Gen 4 on this system.