On December 14th, Intel hosted an event where they presented their products that had artificial intelligence (AI) capabilities. Today I will break down the event so that people who are new to the semiconductor industry can understand what took place.
Often times, presenters at these product release events assume the audience is extremely knowledgeable about the industry, so they use a lot of industry jargon which makes it difficult for casual viewers to understand what is going on. Below are time markers of when certain terms were used during the event, along with their layman term explanation.
Disclaimer: The video and slides below do not belong to us. They are all from Intel.
Full Intel AI Everywhere Event Video
7:59
Xeon 5th Generation is Intel's computer chip for data centers. Customers who are using Intel's 4th Generation Xeon chips can easily upgrade to the 5th Generation platform.
Data Centers: A facility that contains massive computers that store, process, and send data to different locations.
8:26
Intel's Xeon chip can do AI training and inference. Training use massive amounts of data to teach AI models how to recognize patterns within the data. Once the model has been trained to recognize patterns in the data, the model can now be used to make predictions about new data. The better an AI model is trained, the better predictions it can make.
Training AI models is expensive and time consuming. As of today, NVIDIA makes the best chips to train large, complex AI models. The Xeon chips can train less complex models and is well equipped to perform inference/predictions.
Xeon 5th Generation is on Intel 7 technology.
9:20
AI PC is a personal computer with AI capabilities.
Intel's Core Ultra line has AI capabilities because the computer not only has a CPU, which is the brains of the computer that receives and sends instructions, it also has a Neural Processing Unit (NPU) that can handle AI related tasks. The NPU reduces the amount of work that the CPU needs to do.
Intel Core Ultra is on Intel 4 technology and Xeon is on their 7 technology.
20:47
Intel Core Ultra Specs
3D Performance Hybrid Architecture with Foveros is a chip format developed by Intel where they combine Performance and Efficiency CPU's (cores) along with a System on Chip (SOC). The Performance Cores (P-Cores) handle power intensive tasks like gaming. The Efficiency Cores (E-Cores) are smaller than the Performance cores and they handle tasks that require less power than the Performance Core. In addition to the Performance & Efficiency Cores, the 3D Hybrid Architecture also has a Low Power Island Efficiency Core (LPE) which deals with even less power intensive tasks than the standard Efficiency Core such as the computer's camera, memory, and NPU. The Low Power Efficiency Core components are shown below.
Intel's 3D Performance Hybrid Architecture with Foveros makes the computer run more efficiently because computer tasks are directed to the appropriate Performance or Efficiency Cores. Properly sending tasks to the right Performance or Efficiency Cores can improve a computer's battery life.
Foveros is Intel's way of designing their computer chips where one chip is stacked on top of the other. They stack their chips vertically instead of flat.
Intel 4 Compute Tile is an Intel chip that contain Performance and Efficient Cores.
Up to 6x P- Cores represents Intel Core Ultra's 6 Performance Cores
8x E- Cores represents Intel Core Ultra's 8 Efficient Cores
2x LPE Cores represents Intel Core Ultra's 2 Low Power Efficient Cores. These cores handle tasks that require less energy than the standard Efficient Cores.
22 Threads describe Intel Core Ultra's ability to process 22 computer tasks at the same time.
Up to 5.1 GHZ describes the speed of the computer. In a very general sense, higher gigahertz (GHZ) leads to faster computer speeds. Different components within a computer can make this relationship break down though.
Built in Arc GPU with up to 8 Xe Cores means Intel Core Ultra has 8 cores to process graphics.
Dedicated NPU with n-Stream Execution means there is a chip dedicated to processing AI related tasks. I don't know what n-Stream Execution means???
Up to 64GB means the computer's memory can store 64 gigabytes of data. 1 gigabyte is 1 billion bytes of data.
LPDDR5 (x)- 7467/ 96GB DDR% - 5600: LPDDR5 means Low Power Double Data Rate 5th Generation and it refers to the memory being able to operate with low amounts of power. The 7467 & 5600 refer how fast the memory can transfer data.
Thunderbolt 4 is a USB type connection that transfers data extremely fast. It can transfer a maximum of 40 Gigabytes per second.
Integrated Wi-Fi 6E (Gig+) is the enhanced version of Wi-Fi's 6th generation. Each generation is faster than the previous one.
21:15
Performance-Per-Watt measures how energy efficient a computer system is.
Core i7 1370P was released in the 1st quarter of 2023 and it was made on Intel 7 technology.
Ryzen 7840U was released 5/3/2023 by AMD.
Intel Core Ultra 7165H was released 12/14/23.
22:28
28W Envelope is the maximum amount of heat a component or computer chip can disperse.
23:44
Video Game Performance
Core i7 1370P - 25 FPS at 29.61 W vs Ultra 7 165H - 83 FPS at 31.41 W show Intel's latest chip (located on the right) can display more frames per second (FPS) in a video game than their previous chip. A higher FPS means better graphics and a smoother gameplay experience.
Xess means Xe Super Sampling which helps increase the Frames Per Second in a game.
25:35
Microsoft Copilot is an AI tool that helps people become more productive when they're using a Microsoft program.
Windows Studio Effect is software created by Microsoft that applies filters and effects to the computer's camera and audio.
26:15
OpenVino is a platform created by Intel to help people develop their own AI models.
FP8 & FP16 refer to Floating Point numbers???
Int8???
27:03
ISV stands for Independent Software Vendor. This is someone who creates and sells their own software.
27:55
Models are used to build artificial intelligence programs.
Quantization Tools transform things with infinite values into things that can be measured. For example, although light has a wide range of values, we can describe one unit of light as a photon.
Frameworks contain the tools necessary to build an AI model.
Applications are computer programs.
31:48
OEM stands for Original Equipment Manufacturer. OEM's are companies that make products that are then used in the products of other companies. For example, HP makes laptops, but they need Intel's computer chips to operate.
34:45
Advanced Matrix Extensions is a tool created by Intel to speed up the development of AI Models.
34:57
Deep Learning is AI that learns the way humans do. The more experience the model gains, the better it can solve problems.
35:57
CSP stands for Cloud Service Provider. CSP's store data for customers in a data center. The customer gains access to their data over the internet by paying a fee to their Cloud Service Provider.
36:52
1.5 Billion Parameters describe the amount of variable data points that are used to train the BioGPT AI model. AI models vary in the amount of data points needed for training.
SLA means Service Level Agreement and it is the contract between a service provider and their customer.
37:04
Faster Natural Language Processing refers to training a machine so it can understand human language.
Network is the connection between different devices so data can be exchanged between them.
Faster User Plane Functions ???
Edge refers to data being processed close to the user of the data, instead of the data being sent and retrieved from a far off data center. When data is stored close to the user, it can be sent and retrieved faster.
Faster Object Classification deals with detecting objects in an image.
38:06
Foundation Models are AI models that can be used to create different programs. Foundation models save people a lot of time because they don't have to create their AI model from scratch.
Generative AI is a model that can produce text, images, and audio. ChatGPT is one example of generative AI.
Machine Learning is teaching a machine to use data so it can improve its accuracy and performance over time.
38:23
Open Data Lakehouse Architecture is a database that contains both raw and processed data.
Query Performance is how well a system responds to a user's request for information.
Query Throughput???
39:10
TCO stands for Total Cost of Ownership, and it calculates the total cost to own something. The calculation includes the Purchase Price of the object + the costs to Maintain that object.
39:45
Hyperscalers are companies that operate large data centers like Amazon, Microsoft, and Amazon.
40:11
Workload Optimized Infrastructure is the process of combining software and hardware in a way that produce the best results from a computer system.
Cloud Native Workloads???
41:10
Google Cloud Virtual Machines is a digital version of a physical computer.
41:45
High Throughput is the amount of data that can be sent over a network within a certain time frame.
Low Latency refers to data that is processed with little delay.
45:25
MLPerf: "ML" stands for machine learning and "Perf" stands for Performance. MLPerf is a benchmark that measures the performance of the software, hardware, and services used in AI.
45:55
Supercomputer is a powerful computer that can analyze massive amounts of data using AI.
HPC stands for High Performance Computing, and it refers to using multiple, powerful computers, to solve complex problems that regular computers cannot handle.
47:03
API stands for Application Programming Interface. It is software that allows different apps to communicate with each other.
Open Software is software that can be changed by its user to fit their specific needs.
Heterogenous Silicon???
49:03
ML Ops stands for Machine Learning Operations, and it is the process of efficiently training, deploying, and monitoring an AI model.
50:41
Production AI is the use of artificial intelligence to make the process of manufacturing products more efficient.
51:39
Intel TDX???
55:54
PyTorch was created by Meta to build artificial intelligence programs.
Conclusion
I hope that my explanation of the terms used during Intel's event gave clarity to those who are new to the semiconductor industry. As you saw, some terms I skipped because I really did not understand them. I still included them with question marks because I wanted to highlight that learning is a process. Although I am still new to industry, I have way more knowledge about it now than when I first started writing about it in January.
This whole process is amazing to me because I never thought that I would be the one writing about technology and enjoying it along the way. For years, I was Jonah and technology was Assyria. This experience is constantly reminding me that nothing is impossible. Through continuous study, we can learn any topic. Don't let anything stop you from tackling the impossible.
Stay strong, stay blessed, and God willing, I will see you next week.
But Jesus looked at them and said, “With man this is impossible, but with God all things are possible.”
Matthew 19:26 ESV
Comments