NVIDIA announces the availability of its new A100 Ampere-based accelerator. The GPU has the same specifications as the A100 SXM, and holds 6912 Shader cores. It's a PCI Express 4.0 compatible product wit 40 GB of HBM2e video memory.
The product holds 432 Tensor cores, there are no RT cores which are not needed anyway as the product is for deep learing and the likes. The switch to a PCIe model also dropped the TDP, now at 250 Watt. The A100 is based on a partially disabled GA100 GPU. In these cards thus have 6912 shader cores and yeah, 54 billion transistors. It is produced at 7 nm.
There is no big difference in other basic specifications, FP64 is at 9.7 TFLOPS, FP64 Tensor Coresrun 19.5 TFLOPS, FP32 performs at 19.5 TFLOPS, Tensor Float 32 has 156 TFLOPS, and Bfloat 16 has 312 TFLOPS. The video memory is Samsung HBM2e 40GB, the memory bandwidth is 1.6TB/sec, and the cooler uses a passive cooler that uses fans installed in the server for cooling.
-- NVIDIA -- More than 50 A100-powered servers from leading vendors around the world — including ASUS, Atos, Cisco, Dell Technologies, Fujitsu, GIGABYTE, Hewlett Packard Enterprise, Inspur, Lenovo, One Stop Systems, Quanta/QCT and Supermicro — are expected following last month’s launch of the NVIDIA Ampere architecture and the NVIDIA A100 GPU. Availability of the servers varies, with 30 systems expected this summer, and over 20 more by the end of the year.
“Adoption of NVIDIA A100 GPUs into leading server manufacturers’ offerings is outpacing anything we’ve previously seen,” said Ian Buck, vice president and general manager of Accelerated Computing at NVIDIA. “The sheer breadth of NVIDIA A100 servers coming from our partners ensures that customers can choose the very best options to accelerate their data centers for high utilization and low total cost of ownership.”
The first GPU based on the NVIDIA Ampere architecture, the A100 can boost performance by up to 20x over its predecessor — making it the company’s largest leap in GPU performance to date. It features several technical breakthroughs, including a new multi-instance GPU technology enabling a single A100 to be partitioned into as many as seven separate GPUs to handle varying compute jobs; third-generation NVIDIA® NVLink® technology that makes it possible to join several GPUs together to operate as one giant GPU; and new structural sparsity capabilities that can be used to double a GPU’s performance.
NVIDIA also unveiled a PCIe form factor for the A100, complementing the four- and eight-way NVIDIA HGX™ A100 configurations launched last month. The addition of a PCIe version enables server makers to provide customers with a diverse set of offerings — from single A100 GPU systems to servers featuring 10 or more GPUs. These systems accelerate a wide range of compute-intensive workloads, from simulating molecular behavior for drug discovery to building better financial models for mortgage approvals.
Server manufacturers bringing NVIDIA A100-powered systems to their customers include:
NVIDIA is expanding its portfolio of NGC-Ready™ certified systems. Working directly with NVIDIA, system vendors can receive NGC-Ready certification for their A100-powered servers. NGC-Ready certification assures customers that systems will deliver the performance required to run AI workloads.
NGC-Ready systems are tested with GPU-optimized AI software from NVIDIA’s NGC™ registry, which is available for NVIDIA GPU-powered systems in data centers, the cloud and at the edge.
NVIDIA A100 Optimized Software Now Available
NVIDIA A100 is supported by NVIDIA Ampere-optimized software, including CUDA 11; new versions of more than 50 CUDA-X™ libraries; NVIDIA Jarvis, a multimodal, conversational AI services framework; NVIDIA Merlin, a deep recommender application framework; the RAPIDS™ suite of open source data science software libraries; and the NVIDIA HPC SDK, which includes compilers, libraries and software tools to maximize developer productivity and the performance and portability of HPC applications.
These powerful software tools enable developers to build and accelerate applications in HPC, genomics, 5G, data science, robotics and more.