Azure Announces ONNX Integration
Microsoft Azure announced at the beginning of last week a preview of Open Neural Network Exchange’s Runtime, or ONNX Runtime, support for NVIDIA’s TensorRT. ONNX is an open format for deep learning, machine learning, and artificial intelligence model exchange that was co-developed by Microsoft, Facebook, and AWS.
ONNX allows developers the freedom to choose the training framework they deem right for the task, running on the training hardware of their choice, and then to pick a different set of hardware to deliver a trained model as a service, called inference. Trained models can be exported to ONNX from popular frameworks such as PyTorch and TensorFlow. ONNX lets developers focus on optimizing the inferencing performance of their trained models separately from the training process.
The company says that this move is another step towards open and interoperable AI. Microsoft and NVIDIA worked jointly to integrate the two systems and have validated NVIDIA TensorRT support for all the ONNX Models in the “model zoo”. This will automatically optimize ONNX models for NVIDIA GPUs in Azure. Azure cites that it has seen up to 2x improved performance using the TensorRT execution provider on internal workloads from Bing Multimedia services.
Google Goes Gaming
Google revealed at the Game Developers Conference (GDC) held in San Francisco on March 19 that it is launching its Stadia cloud gaming service. Stadia will stream games from the cloud to the Chrome browser, Chromecast, and Pixel devices. Google will be powering this streaming using its own data centers, which it says will “ensure servers are as close to players around the world as possible;” delivering lower latency to stream games effectively across the internet.
Google CEO Sundar Pichai, along with Phil Harrison, a former Sony and Microsoft executive, got on stage at GDC to introduce Stadia during a special keynote. Pichai began with saying he plays FIFA 19 “quite a bit” and continued by saying Stadia will be the platform for everyone. Harrison says Google will amplify its game streaming service by using YouTube and many of the platforms creators that already create game clips on the service.
Google previously tested the service as Project Stream in recent months, allowing Chrome users to stream games in their browser, such as Assassin’s Creed Odyssey. Google didn’t reveal an exact date for release, just saying it would be available sometime this year. The company also didn’t put a price tag on the service or the number of games that will be available at launch but is promising more details to come in the summer.
AWS Gives a Sneak Peek at G4 Instances
In a blog post for AWS, Jeff Barr, Chief Evangelist of AWS, gave a sneak peek at the cloud company’s newest GPU-equipped instance, the G4. G4 is designed for machine learning training and inferencing, video transcoding, and other similar applications. He states that the G4 instances will be available in multiple sizes and in bare-metal form.
Barr gave a list of specs that he noted are still being fine-tuned for customers to look forward to, such as: AWS-custom Intel CPUs, up to 8 NVIDIA T4 Tensor Core GPUs, and up to 100 Gbps networking. The new NVIDIA T4 GPUs feature 320 Tensor cores, 2,560 CUDA cores, and 16 GB of memory.
The T4 GPUs also include RT Cores for real-time ray tracing and can provide up to 2x the graphics performance of the preceding NVIDIA M60. That’s all that Barr writes now, but he says he will have a lot more to say about the instances in the future.
Alibaba Unveils PolarDB
At its Beijing Summit Alibaba unveiled several new product launches that it hopes will help facilitate cloud adoption in China, including the service PolarDB. PolarDB is a cloud-native relational database service that is designed for enterprise-grade database applications and is compatible with Oracle, MySQL, and PostgreSQL.
The company also announced X-dragon Super-Computing Cluster Instance SCC-GN6, which is the most powerful super-computing bare-metal server instance the company has launched to date. The new instance has the capability of enhancing the cluster performance by more than 100%. SCC-GN6 supports supercomputing scenarios such as autonomous driving, machine interpretation, natural language processing, and recommendation systems.
Cloud computing is becoming the main business focus of Alibaba group and the company’s continued development of products and services in the arena is only proof of that. At the summit, Jeff Zhang, President of Alibaba Cloud and Chief Technology Officer of Alibaba Group stated, “Today, Alibaba Cloud not only provides infrastructural support to the entire Alibaba economy but has also developed proven technologies to empower millions of customers in China and worldwide, significantly lowering cloud adoption barriers, ensuring inclusive access, and enabling collaborating throughout the ecosystem.”
Tencent’s Earnings Report is In
In final news, Tencent reported their earnings last Thursday. The Chinese investment conglomerate stated their Cloud revenue has increased by over 100% to 9.1 million RMB, or $136 billion, for the year 2018. The company’s paying customers more than doubled year-on-year in the fourth quarter of 2018.
Tencent Cloud’s global infrastructure covered 25 regions and operated 53 availability zones as of the end of 2018, posting an increase of 4 more regions and 17 availability zones from the end of 2017.
That’s a wrap for this week’s Liftr Cloud Look Ahead. Has your business made major strides using cloud? We want to hear from you! Email us at ideas@liftrcloud.com.
See you next week!