Anisha

• Public cloud providers, such as Amazon Web Services, Microsoft Azure, and Google Cloud, offer different services that allow users to request instances on a pay-as-you-go basis.
• Users can choose between EC2 and Azure Virtual machines, or SageMaker, a fully managed machine learning service.
• The second option, AI-as-a-Service, uses services provided by Amazon that do not require the user to create any models or algorithms.
• The third option is to use Amazon's AI-as-a-Service, which offers the user access to predefined functions and services.
• Users can choose between different processors and software, depending on their needs.
• Hardware requirements for AI workloads depend on the specific instruction set that the processor supports.
• Intel DL Boost is a feature that is available on certain 2nd and 3rd Gen Intel Xeon Scalable processors.
• It enables you to achieve a four times speedup for any workload when you move from 32-bit to 8-bit compute.
• This instruction set is only available in 2nd and 3rd Gen Intel Xeon Scalable processors.
• You can check whether an existing instance has Intel DL boost exposed by running a command called lscpu.
• If you only need small to medium machine learning inference-based workloads, you don't need more than four virtual CPUs.
• For anything involving training, Intel offers the Habana Gaudi DL1 instances, which is specific to deep learning training on AWS.
• Different models have different requirements on resources, which impacts how you choose the cloud instance.
• On the left is CNN, which is computation intensive. On the right is the operational intensity, which is measured in flops per byte. CNN is the highest.
• There are different types of deep learning models with different characteristics.
• GPUs have more compute resources, but fewer memory resources.
• A CPU has much larger memory space than a GPU, and a CPU bus has much higher bandwidth and lower latency than a PCIe bus.
• Data is another important factor to consider when choosing a model.
• Real-time inference is important for applications where data movement is expensive.

Test your knowledge on cloud computing services, AI-as-a-Service, hardware requirements for AI workloads, and deep learning model characteristics. This quiz covers topics such as public cloud providers, machine learning services, processor features, and resource considerations for different models.