Unprecedented Acceleration at Every Scale
The NVIDIA A100 Tensor Core GPU delivers unprecedented acceleration at every scale for AI, data analytics, and high-performance computing (HPC) to tackle the world’s toughest computing challenges. As the engine of the NVIDIA data center platform, A100 can efficiently scale to thousands of GPUs or, with NVIDIA Multi-Instance GPU (MIG) technology, be partitioned into seven GPU instances to accelerate workloads of all sizes. And third-generation Tensor Cores accelerate every precision for diverse workloads, speeding time to insight and time to market.
A100 is part of the complete NVIDIA data center solution that incorporates building blocks across hardware, networking, software, libraries, and optimized AI models and applications from NGC™. Representing the most powerful end-to-end AI and HPC platform for data centers, it allows researchers to deliver real-world results and deploy solutions into production at scale.
BERT Training
BERT pre-training throughput using Pytorch, including (2/3) Phase 1 and (1/3) Phase 2 | Phase 1 Seq Len = 128, Phase 2 Seq Len = 512; V100: NVIDIA DGX-1™ server with 8x V100 using FP32 precision; A100: DGX A100 Server with 8x A100 using TF32 precision.
AI models are exploding in complexity as they take on next-level challenges such as accurate conversational AI and deep recommender systems. Training them requires massive compute power and scalability.
NVIDIA A100’s third-generation Tensor Cores with Tensor Float (TF32) precision provide up to 20X higher performance over the prior generation with zero code changes and an additional 2X boost with automatic mixed precision and FP16. When combined with third-generation NVIDIA® NVLink®, NVIDIA NVSwitch™, PCI Gen4, NVIDIA Mellanox InfiniBand, and the NVIDIA Magnum IO™ software SDK, it’s possible to scale to thousands of A100 GPUs. This means that large AI models like BERT can be trained in just 37 minutes on a cluster of 1,024 A100s, offering unprecedented performance and scalability.
NVIDIA’s training leadership was demonstrated in MLPerf 0.6, the first industry-wide benchmark for AI training.
A100 introduces groundbreaking new features to optimize inference workloads. It brings unprecedented versatility by accelerating a full range of precisions, from FP32 to FP16 to INT8 and all the way down to INT4. Multi-Instance GPU (MIG) technology allows multiple networks to operate simultaneously on a single A100 GPU for optimal utilization of compute resources. And structural sparsity support delivers up to 2X more performance on top of A100’s other inference performance gains.
NVIDIA already delivers market-leading inference performance, as demonstrated in an across-the-board sweep of MLPerf Inference 0.5, the first industry-wide benchmark for inference. A100 brings 20X more performance to further extend that leadership.
BERT Large Inference
BERT Large Inference | NVIDIA T4 Tensor Core GPU: NVIDIA TensorRT™ (TRT) 7.1, precision = INT8, batch size = 256 | V100: TRT 7.1, precision = FP16, batch size = 256 | A100 with 7 MIG instances of 1g.5gb: pre-production TRT, batch size = 94, precision = INT8 with sparsity.
Throughput for Top HPC Apps
Geometric mean of application speedups vs. P100: benchmark application: Amber [PME-Cellulose_NVE], Chroma [szscl21_24_128], GROMACS [ADH Dodec], MILC [Apex Medium], NAMD [stmv_nve_cuda], PyTorch (BERT Large Fine Tuner], Quantum Espresso [AUSURF112-jR]; Random Forest FP32 [make_blobs (160000 x 64 : 10)], TensorFlow [ResNet-50], VASP 6 [Si Huge], | GPU node with dual-socket CPUs with 4x NVIDIA P100, V100, or A100 GPUs.
To unlock next-generation discoveries, scientists look to simulations to better understand complex molecules for drug discovery, physics for potential new sources of energy, and atmospheric data to better predict and prepare for extreme weather patterns.
A100 introduces double-precision Tensor Cores, providing the biggest milestone since the introduction of double-precision computing in GPUs for HPC. This enables researchers to reduce a 10-hour, double-precision simulation running on NVIDIA V100 Tensor Core GPUs to just four hours on A100. HPC applications can also leverage TF32 precision in A100’s Tensor Cores to achieve up to 10X higher throughput for single-precision dense matrix multiply operations.
Customers need to be able to analyze, visualize, and turn massive datasets into insights. But scale-out solutions often become bogged down as these datasets are scattered across multiple servers.
Accelerated servers with A100 deliver the needed compute power—along with 1.6 terabytes per second (TB/sec) of memory bandwidth and scalability with third-generation NVLink and NVSwitch—to tackle these massive workloads. Combined with NVIDIA Mellanox InfiniBand, the Magnum IO SDK, and RAPIDS suite of open source software libraries, including the RAPIDS Accelerator for Apache Spark for GPU-accelerated data analytics, the NVIDIA data center platform is uniquely able to accelerate these huge workloads at unprecedented levels of performance and efficiency.
BERT Large Inference | NVIDIA TensorRT™ (TRT) 7.1 | NVIDIA T4 Tensor Core GPU: TRT 7.1, precision = INT8, batch size = 256 | V100: TRT 7.1, precision = FP16, batch size = 256 | A100 with 1 or 7 MIG instances of 1g.5gb: batch size = 94, precision = INT8 with sparsity.
A100 with MIG maximizes the utilization of GPU-accelerated infrastructure like never before. MIG allows an A100 GPU to be partitioned into as many as seven independent instances, giving multiple users access to GPU acceleration for their applications and development projects. MIG works with Kubernetes, containers, and hypervisor-based server virtualization with NVIDIA Virtual Compute Server (vCS). MIG lets infrastructure managers offer a right-sized GPU with guaranteed quality of service (QoS) for every job, optimizing utilization and extending the reach of accelerated computing resources to every user.
Ultimate performance for all workloads.
Highest versatility for all workloads.
* With sparsity ** SXM GPUs via HGX A100 server boards, PCIe GPUs via NVLink Bridge for up to 2-GPUs
Join this webinar to learn what's new with the NVIDIA Ampere architecture and its implementation in the NVIDIA A100 GPU.