NVIDIA breaks AI reasoning record and promotes two new GPU products

NVIDIA's latest A30 and A10 GPU data center reasoning products for enterprise class servers completed their first performance show today and set a new reasoning record.
In mlperf benchmark, the industry recognized AI performance benchmark, NVIDIA GPU has set a new record in all categories.
NVIDIA refreshes AI reasoning record, promotes two new GPUs and six AI algorithms
Most of the systems participating in data center and edge reasoning test use NVIDIA GPU as AI accelerator, while a few use accelerators from other chip manufacturers such as Qualcomm, Xilinx, neuchips, Centaur, arm, etc.
NVIDIA's two new GPUs combine high performance with low power consumption, and can become the mainstream choice for enterprises in AI reasoning, training, graphics and traditional enterprise computing workload. Cisco, Dell technology, Huiyu, Inspur and Lenovo are expected to integrate the two GPUs into their highest capacity servers from this summer.
1、 NVIDIA has set a new record in six kinds of algorithm testing
Mlperf benchmark was launched in 2018 by David Patterson, winner of Turing prize, United Technology Company and the world's top universities. It is the industry's first set of general benchmark to measure the performance of machine learning software and hardware. It can show different performance of different CPU, GPU and accelerator combinations. Now mlperf has become one of the most mainstream standards to evaluate AI performance in the industry.
"As AI continues to bring changes to all walks of life, mlperf has become an important tool for enterprises to make wise decisions on it infrastructure investment." "Today, all major OEMs have submitted mlperf test results," said Ian buck, general manager and vice president of NVIDIA accelerated computing
In addition to the test results submitted by NVIDIA, Alibaba cloud, Dell technology, Fujitsu, Gigabyte technology, Huiyu, Inspur, Lenovo and ultramicro also submitted more than 360 test results based on NVIDIA GPU.
According to the latest test results, NVIDIA is the only company that delivers results for every test in the data center and edge categories, and shows leading performance in all mlperf workloads.
For example, in the benchmark test of data center, NVIDIA A100's performance is 17-314 times higher than that of CPU, and the newly released A10 and A30 also show good test results.
NVIDIA broke the AI reasoning record, promoted two new GPU products and six types of AI algorithms
Compared with cloud AI 100, NVIDIA A100 shows higher performance in resnet-50 and SSD large algorithms.
NVIDIA broke the AI reasoning record, promoted two new GPU products and six types of AI algorithms
NVIDIA A100 and Jetson series show the test results of all six kinds of algorithms in the edge benchmark.
NVIDIA broke the AI reasoning record, promoted two new GPU products and six types of AI algorithms
In addition, NVIDIA pioneered the multi instance GPU performance of NVIDIA ampere architecture, using seven MIG instances on a single GPU and running all seven mlperf offline tests at the same time. This configuration achieves almost the same performance (98%) as a single MIG instance.
These results demonstrate the performance and versatility of MIG. The infrastructure manager can configure the appropriate number of GPU computing for specific applications, so that each data center GPU can play the maximum utility.
Multiple submission results are also based on NVIDIA Triton inference server. The reasoning server supports models from all major frameworks, can run on GPU and CPU, and is optimized for different query types such as batch processing, real-time and serial transmission. It can simplify the complexity of AI deployment in applications, and ensure leading performance.
In the case of equivalent configuration, the performance achieved by using Triton's submission results is close to that achieved by the optimized GPU implementation and CPU implementation.
2、 A30 and A10 commercial this summer
NVIDIA can achieve such a good result thanks to the breadth of NVIDIA AI platform.
The AI platform includes many types of GPUs and optimized NVIDIA software, including tensorrt and NVIDIA Triton inference server. Microsoft, pinterest, postmates, T-Mobile, USPS, wechat and other enterprises have deployed NVIDIA AI platform.
A30 and A10 GPUs are the latest members of NVIDIA AI platform.
NVIDIA broke the AI reasoning record, promoted two new GPU products and six types of AI algorithms
A30 can provide common performance for industry standard servers, support a wide range of AI reasoning and mainstream enterprise computing workloads, such as recommendation system, conversational AI and computer vision.
A10 accelerates deep learning reasoning, interactive rendering, computer-aided design, and cloud games, enabling enterprises to support hybrid AI and graphical workloads based on a common infrastructure. By using NVIDIA virtual GPU software, the management can be improved, and the utilization rate and configuration of virtual desktop for designers, engineers, artists and scientists can be improved.
NVIDIA Jetson platform is based on NVIDIA Xavier system level module, which can provide server level AI performance at the edge and help realize more innovative applications in robotics, health care, retail and other fields. Based on NVIDIA's unified architecture and cuda-x software stack, Jetson is the only platform with compact design, capable of running all edge workloads and less than 30W power consumption.
In the past six months, NVIDIA's end-to-end AI platform has achieved 45% performance improvement in mlperf.
NVIDIA broke the AI reasoning record, promoted two new GPU products and six types of AI algorithms
NVIDIA A100 GPU is equipped with servers of leading server manufacturers, cloud of all major cloud service providers, and NVIDIA DGX system product portfolio (including NVIDIA DGX station A100, NVIDIA DGX A100 and NVIDIA DGX superpod).
A30 and A10 (165W and 150W respectively) are expected to be used in all kinds of servers from this summer, including NVIDIA certification system which has been strictly tested to ensure high performance under all kinds of workload.
NVIDIA Jetson AgX Xavier and Jetson Xavier NX system level modules have been supplied through global distributors.
NVIDIA Triton and NVIDIA tensorrt can be obtained through NVIDIA's software directory NGC.