HomeAI News
A quick 5-minute overview of the opening speech of Microsoft Technology Conference: The birth of the master of AI software and hardware map set
12

A quick 5-minute overview of the opening speech of Microsoft Technology Conference: The birth of the master of AI software and hardware map set

Hayo News
Hayo News
November 16th, 2023
View OriginalTranslated by Google

At 1 a.m. Beijing time on Thursday, Microsoft Ignite, the world's top technology conference, officially opened in Seattle. Microsoft CEO Satya Nadella gave an opening speech that lasted nearly an hour, introducing the latest progress made by the AI ​​leader.

Nadella said that at this week's conference, Microsoft will be able to come up with more than 100 updates related to AI , but as in the past, he will focus on five aspects in today's speech.

image

1. Peeling off the cocoon: What new technologies are there in Microsoft’s data center?

In the first part of the press conference, Nadella first mentioned that Microsoft has purchased a total of 19 gigawatts of renewable electricity since 2013, and is preparing to go further and cooperate with a number of wind power, photovoltaic, and nuclear fusion suppliers. , the company is moving toward the goal of using "zero-carbon emission" energy in all data centers by 2025.

image

In terms of network connections, in order to meet the requirements of AI and future workloads, Microsoft is producing a new generation of hollow core fiber (Hollow Core Fiber), which uses air as the light guide medium of the fiber to increase transmission speed by 47%. Of course, this is also related to Microsoft's "money capabilities". The company acquired the UK's Lumenisity, the leader in this track, at the end of December last year.

image
image

Taking a step further, Nadella announced that Azure Boost data center hardware will also be officially put into commercial use. The system is able to offload storage and network processes from the host to specialized hardware and software, thereby increasing speeds.

Next came the first climax of the conference, when Nadella officially released Azure Cobalt, the first CPU designed by Microsoft . This chip is based on Arm architecture and has up to 128 cores. Nadella also called it " the fastest chip among cloud computing vendors ."

image
image

While releasing the chip, Nadella also showed off the close cooperative relationship between Microsoft, Nvidia and AMD. The Azure cloud service will use the latest NVIDIA H200 chip , and the computing power provided by AMD's flagship chip MI300X will be available to some users starting from today. Nadella specifically mentioned that OpenAI’s GPT-4 is now running on MI300X.

image
image

After chatting with the two partners, Nadella took out another Microsoft self-developed chip - the Azure Maia AI accelerator chip for cloud training and reasoning . This chip uses a 5nm process and has 105 billion transistors.

image

Of course, having a powerful chip is not enough. Microsoft has also redesigned a complete cooling system for Azure Maia. Currently, this system is mainly responsible for handling Microsoft’s internal computing power needs, and will also be open to third-party loads in the future.

image

2. Microsoft❤OpenAI

Based on Microsoft's close relationship with OpenAI, the leader in AI software, Microsoft Azure cloud services will also provide users with OpenAI's latest products as soon as possible, such as GPT-4 Turbo and multi-modal capabilities, as well as fine-tuning based on GPT-4.

image

Microsoft also announced today the launch of the " MAAS" industry model (Models as a Service/Models as a Service). Users can directly call APIs, fine-tune and deploy various open source large models through Microsoft services. Microsoft has also reached cooperation with a number of large model developers such as Meta to launch well-known LLMs such as Llama 2 as a service.

image

Microsoft also released its latest open source pre-training model Phi-2 on Wednesday, with 2.7 billion parameters, which is significantly better than the previous generation product in terms of mathematics, logic and other aspects.

image

3. Huang Renxun’s appearance

Nadella also announced on Wednesday that Azure cloud will release Nvidia's AI foundry service . To this end, he also invited Huang Renxun to talk about the comprehensive cooperation between the two AI giants from hardware to software.

image

It is reported that the NVIDIA AI workshop covers three major elements: NVIDIA's AI basic model, NeMo framework and tools, and NVIDIA DGX Cloud AI supercomputing and services. Microsoft users will be able to use Nvidia's AI software to generate and deploy models on the Azure cloud.

Huang Renxun said that the first wave of AI originated from a group of startups such as OpenAI, and now the entire industry has entered the second wave, enterprise-level AI promoted by Microsoft Copilot . In his view, the third and largest wave will be the digitization of heavy industry throughout the world and benefiting from generative AI.

4. Reinventing data processing: Microsoft Fabric

Nadella announced on Wednesday that Microsoft Fabric, a "one-stop" data processing platform, is officially open to the public. The design concept of this platform is to aggregate all data assets of an enterprise and reshape the way existing enterprises process data.

image

At the same time, Fabric's AI assistant (Copilot) can also be integrated with Microsoft office ecosystems such as Office and Teams to provide functions such as "generate PPT with one click". As Microsoft's circle of friends further expands, Fabric will also open data channels with a number of SaaS companies.

5. GPTs are integrated into the Copilot system

Microsoft also took its vision of "Copilot for everyone" a step further on Wednesday, officially renaming Bing Chat to Copilot .

image

The Microsoft 365 service, which is more closely related to workers in the workplace, has also received a major update: in addition to supporting patches for the existing Copilot, the GPTs released by OpenAI last week will also be integrated into the Microsoft Copilot system.

image

Based on this vision, Microsoft also launched Copilot Studio on Wednesday to help users create GPTs and patches, customized workflows, etc., while also integrating software services from SAP and OpenAI GPTs into daily workflows.

image
image

Nadella revealed at the press conference that Microsoft’s office collaboration software Teams currently has 320 million users worldwide. As the basic platform for AI-driven workplace transformation, Microsoft also previewed the upcoming new features of Teams on Wednesday. For example, **"Yuanverse 3D Collaboration Platform" Mesh will be officially open to all users in January next year**.

image

It is worth mentioning that due to lack of time, the "Office work suite" that workers are most concerned about did not get the opportunity to appear at the CEO's press conference, and could only release updates silently on Microsoft's official website.

For example, Word is about to launch a "Understand and Comment" feature that allows users to quickly understand document revisions by asking questions such as "How can I see the changes that have occurred in this document?" to show the changes made by anyone who has accessed the document. changes and revisions.

PowerPoint will launch the "AI generated P-picture" function. Users can use corporate promotional materials (such as a tent) and then generate an image of a "tent placed outdoors" with one click.

image
image

Reprinted from 财联社View Original

Comments

no dataCoffee time! Feel free to comment