Enhance Istio and Envoy Service Mesh with Intel Technologies

Explore Intel's innovative contributions to the Istio* and Envoy* open source community with a focus on achieving performance gains, strengthening security, and enhancing network functionalities through unleashing underlying hardware capabilities.

Fill a Kubernetes Gap: I/O Resource Scheduling & Isolation

This demo shows how to ensure optimal performance for critical workloads. Workloads aren't scheduled to a node whose available I/O resource can't meet the workload's requirements. Conversely, each workload's real-time disk I/O use is monitored, and to guarantee critical workloads' I/O performance, the noisy neighbor workloads' I/O resource boundary is shrunk dynamically.

AI Workload: HPA & VPA Confidential, Sustainable

It is challenging to orchestrate and scale AI workloads in cloud-native environments, especially for cross-domain uses like confidential and sustainable computing. Watch a containerized AI pipeline in a stateless microservice design as the significant workload demonstrate:

1. Horizontal Pod Autoscaler (HPA) and Vertical Pod Autoscaler (VPA) scaling by a Kubernetes* event-driven autoscaler (KEDA) operator.
2. Confidential computing using a confidential cloud-native primitive SDK.
3. Sustainable computing based on Kepler to collect a carbon footprint.

End-to-End Secured Large Language Model (LLM) Serving with Intel® TDX on Alibaba Cloud*

Secured LLM serving deploys and uses LLMs to ensure the confidentiality, integrity, and availability of models and data. This demo showcases end-to-end secured LLM pipelines (including multiturn chat and text completion) powered by BigDL Privacy Preserving Machine Learning (PPML) and Intel® Trust Doman Extensions (Intel® TDX) on Alibaba Cloud, which are secured by various security technologies.

Unleashing the Power of Generative AI on Intel Platforms

Bigdl-llm is an open source library for running LLMs on Intel laptops and GPUs using low-bit optimizations (int3, int4, nf4, int 5, int8) with very low latency. The library is built on top of various technologies and is optimized on Intel CPUs and GPUs. With bigdl-llm, you can build and run LLM applications using standard PyTorch* APIs (such as Hugging Face* Transformers and LangChain*). This demo showcases the experience of running LLMs on laptops using bigdl-llm.

Ad-Serving Showcase from DBApp Based on Secured Cloud Management Stack Around SGX

Secured Cloud Management Stack is an open source software solution to achieve an on-premise and hybrid multicloud management solution with optimized performance that is based on software guard extensions (SGX) and Intel TDX. DBApp is the network and information security leader in the People's Republic of China (PRC). This demo shows:

• SGX enablement in the cloud and how SGX is secure
• Internet precision marketing from DBApp, which combines multiparty data to see the ad-serving effect based on SGX

Build Trustworthy and Confidential LLM Inference with Intel TDX Shield

This demo shows an end-to-end, confidential LLM (Retrieval Augmented Generation [RAG]) reference solution that takes advantage of the Intel TDX shield to address privacy and security concerns. The secure design ingredients include an end-user prompt, LLM mode, RAG pipeline, and database protection with confidentiality and integrity in the full flow. This demo can be deployed in an Alibaba Cloud for Intel TDX instance with Intel® Advanced Matric Extensions (Intel® AMX) enabled to accelerate the RAG interference service.

Intelligent Traffic Searching on a Cloud-Native Edge Stack Enabled for a Private 5G Network

This session gives an in-depth introduction of a cloud-native software framework that helps end-to-end private 5G deployment through a cloud-native edge stack from Intel with intelligent traffic searching as a sample application. The format is a video recording with live Q&A.

Real-time 3D Human Pose Estimation

Estimate a 3D human pose from a single RGB camera without wearable devices in real time, based on container technology. This solution takes the video frames input from a single RGB camera and outputs the 3D human pose in real time. The solution can be readily employed for reconstructing virtual characters in the virtual 3D space by a rendering client, and could be widely used for motion capture, fitness teaching, sport scoring, and other digital twin applications.

A Better Total Cost of Ownership for Cloud Native Solutions through an Open Cloud from Intel

This reference cloud stack is optimized on Intel® architecture for a PRC enterprise cloud ecosystem around OpenStack* and Kubernetes to accelerate development and deployment of highly optimized and scalable cloud solutions. The demo shows how to accelerate the infrastructure and workload for cloud-native solutions with an Intel® Xeon® platform and technology from resource management, storage, and converged HPC and AI.