THE BEGINNINGS OF CENTRAL PROCESSORS – THE FIRST STEP OF A JOURNEY
Continuing from the Eniac in the technology history we see a whole new concept was born with the first large scale programmable computer. The technology of the times was large heat producing inefficient vacuum tubes. But people could barely measure work in seconds while computers promised thousands or more operations per second. Never could they imagine the capabilities of your mobile phone today. But we all have to start somewhere.
So it was time for people and companies to try to design electronic computers. This required two main components: a computer processing unit and memory to store the programs and data and perform the calculations. The basic concept of the Central Processing Unit was to take common mathematical operations that a person might perform at human speeds and now perform it thousands of times faster and without errors. But electronics also enabled operations that could be enormously beneficial in the larger scheme of mathematical requirements but would not be natural for people to perform. In other words the journey opens doors on totally new innovative approaches never before considered.
So the first task is to define the memory units, layout, how to define instructions and how to define data types (integer whole numbers, decimal place numbers, positive and negative numbers, operations, etc. It was extremely important to begin with a good design that one could continue to build upon. So companies and product families began to emerge. And in the beginning people had to learn the instructions, programming in machine code, and overcoming so many basic obstacles that it was extremely difficult to make great progress. And if you changed from one type of computer to anther you had to start all over from scratch and everything you did prior is not re-usable or transferable.
COMPUTER CHIPS BEGAN TO GET SMALLER AND MORE POWERFUL
In May 2005 Advanced Micro Devices (AMD) released their first dual core Central Processing Unit. In other words AMD had managed to shrink their CPU so much that they could now manufacture a 2 core CPU which was the equivalent of having 2 processors on one chip and a new concept of CORE was born.
In June 2011 AMD released their first Advanced Processing Unit. This was a Central Processing Unit (CPU) and a Graphics Processing Unit (GPU) together with shared memory on the same chip!. This was revolutionary because prior to this the CPU would create video content and have to move a ton of visual and auditory information to the GPU. Now the CPU could send a pointer, a simple address, to the shared memory and the GPU could instantly run with it. No latency, wait time, or lag waiting for the data to arrive. This would ultimately lead to AMD dominating the game console market with the Playstation and Xbox.
In June 2013 AMD released the first CPU with a clock speed exceeding 5 GHz (or 5 billion instructions per second.
HYPERTHREADING, SIMULTANEOUS MULTI-THREADING
At some point companies explored new methods of doing more work to compete with others on performance and resource utilization. The idea was that there are multiple instruction execution components that can operate in parallel (addition and multiplication without dependent data). So if the compilers and chips can identify operation that can be done in parallel and exploit available resources to get more work done at the same time, overall performance could be increased. This has a lot of challenges and requires a lot of compiler work but today it is very standard and expected.
MULTI-CORE MEETS CHIPLETS
At one point in time Intel told customers that no one would ever need more than 4 cores. But they went on to design and build an 8 core CPU and sell it to those who could afford it at $1050. At the same time Intel managed to also build a 10 core chip and then they would really gouge you for $1750.
In around 2015 AMD released a CPU colloquially called the Bulldozer which did not do well and nearly caused AMD to go bankrupt. AMD brought in Lisa SU and build a new world class design team who did a complete “clean sheet” design starting from scratch. They came up with a whole new concept in CPU’s called chiplets.
MONOLITHIC CHIPS VS CHIPLETS WITH INFINITY FABRIC
Intel typically builds what is called “MONOLITHIC” chips or very large chips. What happens here is the design tries to fit as many rectangular chips as will fit on one wafer (usually round and possibly 12 inches in diameter). But wafers always have some defects so if a chip fall on one of the defects it goes into the trash. Because of the large size of the Intel chips their normal yields are around 40%. But AMD had a new idea. Let’s make small chiplets with 4 cores that they could mix and match lego style and put them together as needed on a single chip. And to allow them all to function together efficiently they added a high speed bridge called Infinity Fabric. This started with 4 core chiplets and later became 8 core and soon will have 12 core chiplets. So first result of the tiny chiplets is that AMD’s yields reached 95% totally blowing Intel away on costs. Next they took 2 chiplets together to make an 8 core CPU and blew Intel away by releasing it at $499 to compete against Intel at $1050. The process continue to obliterate Intels entire product line up the stack into their most lucrative market – the Data Center. So Intel had 22 core at $4000 and 28 core CPUs at $10000. So a Data Center would have dual socket motherboards and with 2 28 core CPUS you were paying well over $20,000. But at the same time AMD had achieved a 64 core CPU and sell it for $7000 totally blowing Intel away on every metric of performance, efficiency, software licensing, computer computational density, air conditioning and operating costs, etc. Now AMD also has 86 and 128 core CPU and is on the way to 196 cores totally obliterating Intel. Furthermore today ALL SUPERCOMPUTERS are built on AMD processors. Intel only won 1 bid for the Aurora Supercomputer at Argonne Labs but could not deliver because Intel’s technology is broken and they cannot meet the required specifications. Today AMD has gone from 0.1% of the Data Center market in 2016 to over 50%, today. This is also complicated by all of Intel’s performance, overheating, and security vulnerability problems (eg Side Channel Attack and IME).
MORE INTEL PROBLEMS LOOM
Intel has been 3 to 4 generations behind AMD and TSMC for quite some time. Intel just lost its bid to compete with AMD in the game console market. The only way Intel can compete at all is by adding more power lines to the socket with each new CPU and over volting the processors. This has lead to processors running at well over 200 degrees and suffering massive failures where Intel does not stand behind their customers and does everything possible not to pay and replace the defective processors. The last Earnings Report was so bad that Intel announced that they care cancelling dividends and laying of some 20% of their work force globally. With everything that is happening to Intel and their massive problems, horrible work place environment, and lag behind TSMC I do not expect Intel to survive.