HUAWEI CONNECT 2023

Finance Foundation Model Solution

Powered by an Ascend AI cluster and a foundation model that is trained with 100 billion parameters, our Finance Foundation Model Solution is designed for credit, risk control, customer services, and other such scenarios. Bringing intelligence to financial services helps boost efficiency by more than 50%.

Credit approval for MSMEs: 2 weeks → 1 day

Enterprise risk identification accuracy: ↑38%

Government Foundation Model Solution

Our Government Foundation Model Solution helps cities build AI computing infrastructure and go intelligent more easily. Empowering the digital economy, this solution makes government services more accessible and efficient and urban governance more precise.

A major engine of growth for the digital economy with an ROI of more than 300%

Intelligent assistant for government services

Multi-modal, all-domain sensing

Smart Factory Solution

Intelligent Factory Solution will create an AI-powered industrial platform that features cloud-edge-device synergy, realizing intelligent transformation of quality inspection, equipment maintenance, logistics, scheduling, and other services， and helping manufacturers continuously boost quality and efficiency while reducing costs.

Product defect identification ratio: >99%

Production issue response speed ↑40%

Average delivery times ↓20%

AI-Based New Energy Power Prediction Solution

In order to cope with the volatility and unpredictability of new energy power generation, enterprises need to forecast the generated power to provide key support for power dispatch.
1. Improving the accuracy of new energy power prediction hangs on improving the accuracy of meteorological element prediction.
2. The traditional solution is based on x86 CPUs and uses numerical calculation. Power prediction takes a long time and it is tough to increase accuracy by any margin.

Based on the Ascend AI computing power center, Huawei's new energy power prediction solution is the first minute-level rolling prediction algorithm. Huawei worked with partners to build a new energy power prediction solution with cloud-edge synergy to provide accurate prediction service for all stations, improving power generation efficiency and power grid connection and consumption capabilities.

Faster power prediction

Huawei's innovative AI algorithm shortens the time taken to predict power from hours to minutes, improving efficiency by 140 times.

'One station, one policy': accurate prediction service

Based on the cloud-edge synergy architecture featuring 'unified group training + inference at stations', accurate prediction models can be provided for stations, improving prediction accuracy by 3 to 5%.

Intelligent E&P Solution

Our Intelligent E&P Solution is built upon high-performance computing, massive data storage, and an AI exploration and production (E&P) model. Oil and gas companies using this solution will see a spike in E&P efficiency along with increased reserves.

Earthquake and well surveying modeling times: ↓80%

Forecast accuracy: >95%

AI Weather Forecasting Solution

Based on Pangu-Weather, the first AI model to outperform conventional NWP methods in terms of accuracy for short and medium-range weather forecasts, Huawei has built a cloud-based AI weather forecasting solution powered by high-precision Atlas compute clusters. Integrated training and inference are supported. Furthermore, Huawei has developed an ensemble forecast solution powered by large AI models. This solution supports customized AI applications for short and medium-range weather forecasts for weather service agencies as well as enterprises, for example, km-level weather forecasts for specified regions and areas.

AI algorithm innovations include: 1) 3D Earth-specific transformer (3DEST), which is better at processing spatially uneven 3D meteorological data and can learn weather patterns more quickly and accurately. 2) Hierarchical temporal aggregation, which involves training a series of models with increasing forecast lead times. This reduces the number of iterations needed to generate a longer-range forecast, hence the cumulative errors that would result from these iterations.

Accurate

• In upper-air forecasting, Pangu-Weather has 10%+ lower RMSEs than the ECMWF Integrated Forecasting System (IFS).
• Pangu-Weather can generate kilometer-level high resolution forecasts for specified regions.

Fast

• Pangu-Weather can generate a 7-day global weather forecast with a horizontal resolution of 25 km in 10 seconds.
• It can generate an ensemble forecast with over 1,000 members.

Cost-effective

Pangu-Weather can generate a weather forecast within seconds using a few Atlas servers.，with which an ensemble forecast with over 1,000 members is also achievable.

Intelligent Mining Solution for Underground Mines

Our Intelligent Mining Solution for Underground Mines comes packed with an industrial Internet network architecture, low-frequency intrinsically safe 5G, an intrinsically safe IP ring network, the Pangu mine model, and the industry's first MineHarmony OS. By bringing intelligence to mining equipment and applications, this solution will lead to safer and more productive mining.

World's first ever foundation model for the mining industry

Time to service of new algorithms: A few days

AI applications in 20+ scenarios powered by the Pangu mine model

Intelligent Transportation System (ITS) Solution

Our Intelligent Transportation System (ITS) Solution employs technologies such as software-defined cameras (SDCs), radar- and video-based trajectory generation, edge computing, cloud, and big data, to realize the intelligent detection of traffic accidents and provide the integrated display of transportation elements, including both operations and scheduling. This solution will make transportation management easier and faster.

• Traffic accidents detected within 10 seconds
• Secondary car accident risk: ↓70%

• 10+ types of traffic violations
• Transportation efficiency ↑15%+

Train Freight Detection System (TFDS) Solution

Our Train Freight Detection System Solution, or TFDS Solution, is supported by Huawei's Pangu railway model. As it covers all TFDS scenarios, the solution makes intelligent analysis of fault images and integrated delivery possible, thus increasing detection efficiency by 200%.

67 freight train models、430+ faults

Fault detection accuracy: >99.3%，Near-zero missed inspections of critical faults

OceanProtect X3000 Backup Storage

Huawei OceanProtect X3000 backup storage system is designed based on end-to-end acceleration and Active-Active reliable architecture. It features ease of use, ultimate efficiency and reliability. This industry-leading performance and reduction backup storage helps users achieve efficient backup and recovery with significant TCO savings and is widely used in IT environments such as small and medium businesses and remote/branch offices.

Efficient

All I/O paths are optimized end-to-end, with max physical and logical backup bandwidths of 6 TB/hour and 10 TB/hour, respectively, 40% higher than the industry benchmark.Multi-layer inline variable-length deduplication and feature-based compression tech help achieve a max data reduction ratio of 72:1, 30% higher than the industry benchmark.The deduplication function helps slash the replication bandwidth between sites by up to 98%.

Reliable

Industry's only active-active reliable architecture supports failover in seconds in event of a single controller failure and ensures zero service interruption. Ransomware protection technologies — encryption (protocol, replication link, arrays), secure snapshot, WORM, and air gap — ensure the security and availability of backups.

Easy to use

The 2U chassis is only 488 mm long, energy-saving.Three-step resource allocation using the initial configuration wizard greatly improves resource allocation efficiency.Provides a graphical user interface and supports mobile intelligent O&M for easy management.

OceanStor A800 High-Performance AI Knowledge Repository Storage

Large AI model parameters have grown exponentially in scale, from the hundreds of billions to trillions. Training sets have also evolved from unimodal to multi-modal (images, videos, etc.) paradigms, resulting in a data explosion that has increased volumes by thousands, and brought unprecedented challenges.
First, poor loading performance among massive numbers of small files has undermined training efficiency.
Second, large model parameters are frequently being optimized, which has led to unstable training platforms, with service interruptions occurring every two days on average, and frequent, time-consuming read and write of TB-level checkpoints.
Third, training 100-billion-parameter models requires massive GPU resources, since individual GPUs are not efficient enough, and ensuring millisecond-level real-time inference for conversational applications is quite challenging.

OceanStor A800 delivers a staggering 24 million IOPS and 500 GB/s bandwidth within a single controller enclosure, serving as the high-performance AI knowledge repository storage solution. With four times the performance of the next runner-up, a single storage system is all you need for small file loading of training sets and high-bandwidth-based resumable training. You'll also benefit from access to a leading intrinsic vector knowledge repository, which supports large model inference applications with over 250,000+ QPS, achieving accelerated vector retrieval and millisecond-level inference response.

Ultra-high performance

Our groundbreaking data-control separation architecture facilitates the direct flow of data to disks, reducing CPU usage and delivering 24 million IOPS per controller enclosure. Training set loading efficiency is 4x as high as the industry's next best.

Rapid recovery

Cutting-edge technologies like disk-controller collaboration and NFS+ parallel client allow for multi-link parallel access. 500 GB/s bandwidth per controller enclosure enables 3x faster resumable training from checkpoints than the industry's next best.

Robust inference

Our leading intrinsic vector knowledge repository eliminates AI hallucinations and boosts vector retrieval speeds to 250,000+ QPS, with millisecond-level inference response.

AI Data Lake Storage OceanStor A310

Powered by an AI-native design, a single storage system covers the entire AI process. The all-flash product model houses up to 8 nodes and 96 disk slots in a 5 U chassis and adopts system-wide PCIe 5.0 to deliver industry-leading hybrid performance of 400 GB/s bandwidth and 12 million IOPS, helping you easily eliminate I/O bottlenecks in AI process.

Multi-protocol interworking

It provides a unified foundation that enables data access across multiple protocols such as NFS, CIFS, S3, HDFS, and POSIX, and allows one data copy to be invoked in multiple phases, eliminating data copying and conversion throughout the AI process.

Data fabric

Based on a global file system (GFS), it provides a globally unified data view and scheduling across systems, regions, and clouds to help users access AI raw data scattered in different regions and simplify the process of aggregating massive amounts of AI data.

Near-memory computing

With embedded powerful computing power, it provides near-data, precise processing capabilities that allow you to filter, combine, and process AI training data on the storage side and avoid unnecessary data transmission between storage and compute for 30% higher training efficiency.

Excellent scalability

The fully symmetric scale-out architecture can be smoothly expanded to 4096 nodes anytime and anywhere, helping you easily cope with ever-increasing data generated in the entire AI process.