NVIDIA sets Six MLPerf AI Benchmark Records

NVIDIA sets Six MLPerf AI Benchmark Records

NVIDIA achieved six new MLPerf AI performance records with the release of the industry’s first broad set of AI benchmarks. Tensor Core GPUs achieved the best performance across every MLPerf benchmark submitted.

MLPerf consortium was formed to build a common set of benchmarks to enable the machine learning (ML) field to measure performance for both training and inference from mobile devices to cloud services. Google, Intel, Baidu, NVIDIA, Facebook, Microsoft and more than dozens of technology leaders support MLPerf.org. It covers AI fields like computer vision, language translation, personalized recommendations, and reinforcement learning tasks.

NVIDIA submitted results for six categories such as image classification, object instance segmentation, object detection, non-recurrent translation, recurrent translation, and recommendation systems. NVIDIA did not submit results for the seventh category, Reinforcement Learning as this test category does not take advantage of GPU acceleration.
NVIDIA, MLPerf Categories

According to NVIDIA announcement, "MLPerf scores are computed by first measuring time to train to the specified target accuracy, then normalizing to the time taken by an unoptimized implementation on a reference platform. The goal of normalization is to have a similar magnitude of scores across different benchmarks since benchmarks take different amounts of time to train. Both the time-to-train and MLPerf scores are published on the MLPerf website, and so we present time to train in minutes."

NVIDIA used NVIDIA DGX systems, including NVIDIA DGX-2 featuring 16 fully connected V100 Tensor Core GPUs and claims that it is the world’s most powerful AI system. Results for NVIDIA time to convergence, both for single-node and at-scale implementations are depicted in the image below.

Vice president and general manager of Accelerated Computing at NVIDIA, Ian Buck, said, “The new MLPerf benchmarks demonstrate the unmatched performance and versatility of NVIDIA’s Tensor Core GPUs. Exceptionally affordable and available in every geography from every cloud service provider and every computer maker, our Tensor Core GPUs are helping developers around the world advance AI at every stage of development.”

The 18.11 release of the NVIDIA GPU Cloud (NGC) deep learning containers with the complete software stack and the top AI frameworks, optimized by NVIDIA that achieved outstanding MLPerf performance are available for download from NVIDIA GPU Cloud (NGC) container registry.

The details of benchmark results are available at https://mlperf.org/results/.


Note: We at TechSutram take our ethics very seriously. More information about it can be found here.
Mandar Pise Opinions expressed by techsutram contributors are their own. More details

Mandar is a seasoned software professional for more than a decade. He is Cloud, AI, IoT, Blockchain and Fintech enthusiast. He writes to benefit others from his experiences. His overall goal is to help people learn about the Cloud, AI, IoT, Blockchain and Fintech and the effects they will have economically and socially in the future.

No comments:

Post a Comment

    Your valuable comments are welcome. (Moderated)