Artificial Intelligence It can be said that his teenage stages are clumsy. After its appearance on the market, AI solutions are growing everywhere due to advances in computing power and the use of rich data.
Mr. Gaddy Singer, vice president of Intel's Artificial Intelligence Product Group, said, "Extensive learning has begun from the prominent childhood period in 2014 and 2015 to 2019 and 2020.
As part of AI's current growth, it includes a wide range of chip solutions, ranging from traditional chip makers such as Intel to cloud computing companies like Google and numerous start-up companies. Mr. Singer argues that as the market develops over the next few years, it will evolve to Intel's strengths.
"As we moved to reasoning, extensive deployment, real-time latency solutions, and total cost of ownership, these aspects were all really appealing to Intel for the past years.
Singer explained how Intel was considering the AI chip market,
Three changes in AI that form the foundation of Intel's strategy
Mr. Singer pointed out that three major trends were recognized in the industry, as the singer remembered around 2014. First, detailed learning develops richer feature sets and gathers richer data. For example, image recognition is moving from identifying cats to identifying potentially malignant cells in 3D images.
The organization then proofs the conceptual development to training and reasoning. With this transition, Tinger adds more weight to total cost of ownership, he says.
The third trend is the emergence of a deep learning framework. Currently, dominant executives such as Caffe and TensorFlow did not exist or were very early a few years ago. This allowed us to move from our own solution to a large integrated solution.
Part 1: Build new talent and technology
In anticipation of the deeper development of detailed learning, Intel has been doing a few things, starting with introducing more talent and new technology. In recent years, Intel acquired Nervana, Movidius, Mobileye for investment in Altera.
Through these acquisitions, Intel is focusing on hardware and software development, Singer said. This means that at the substance level, we must continue to "stimulate Zeon as the basis of AI".
Mr. Singer pointed out that Boosting Xeon also has relevance to software. "Over the past two years we've been trying to optimize the software stack and exploit the parallelism inherent in Xeon hardware.
Make a case for Xeon:
Singer said Xeon plays an important role even if it introduced a GPU and a specialized accelerator.
On the other hand, cloud service providers and companies are efficient only looking at the total cost of ownership. Intel also enhanced Xeon with extensions such as AVX and other extensions to support concurrent workloads. By combining neural network calculations with more traditional calculations, the company continues to improve chips through "machine learning functions that are particularly effective for mixed workloads".
As an example, singer refers to the "attention" function of machine translation and refers to a large dataset when translating a more targeted dataset. This function says "If you try to understand images and words in the image inside your brain, you actually imitate it," singer says.
"The calculation of this part will work very well, depending on the general purpose of the processor," he said.
Overall, Xeon says that it is competitive because "flexibility and real workload are mixed."
Part 2: Creating diverse product lines
Intel is building a variety of solutions to meet very different needs. "Sometimes performance, sometimes energy efficiency, sometimes latency"
In addition to Xeon, this includes Movidius VPUs and FPGAs that support Microsoft Azure. The first Nervana commercial processor will be released next year.
How about a special chip manufactured by a cloud company?
Some of Intel's major customers are designing their own chips to meet their own needs. For example, Google owns a TPU. Ultimately, this development makes sense. After all, who knows better the necessity of AI treatment than Google?
Mr. Singer said that developing the best solution for Intel's product portfolio to meet the cloud provider's current computing needs and its new needs is "to understand computational trends and leadership We are convinced that we can create solutions and work with cloud service providers and the industry to do this.
In addition, Intel can provide close system integration, says Mr. Singer. "This brings great value to cloud service providers and we will see how our internal solutions match the products and solutions our company offers, which will be developed over the next few years.
Part 3: System integration
System integration is the third part of Intel's strategy to win the AI chip war.
"If you say that most tasks are GPU and accelerator executing tasks, there really is a workload to solve them," says Mr. Singer. Intel focuses on "how to create the best system solution, because it will evolve to ultimately provide concrete solutions to customers"