“Intel Processor Giant Trips”
By Scott Hamilton
If you look at the full spectrum of Intel’s service offerings, the company is still in solid shape as the giant in computer processor technology. However, they recently announced a delay in the release of their next generation processor, due out this fall. You might say the Intel giant has tripped. This leaves us to question, will the giant fall?
Intel has seen new challenges in the last few years that it has not had to deal with in about a decade. There are two new competitors in the processor market, both taking double-digit shares of the market from the once dominant Intel. The second major impact to Intel is the end of Moore’s Law. Moore’s law stated that the number of transistors in a single processor die would double every twelve months. This law held true for nearly four decades, but slowed significantly as the chip manufacturing technology resulted in transistors nearly the size of an atom.
Up until this year Intel was the leader in developing the next technology to make the transistor smaller. Last year they released their 10-nanometer technology, which means that the smallest feature size on chip die is 10-nanometers wide. They have been running a continuous cycle nicknamed tick-tock for the last several decades. In the “tick” they develop a new processor with new features, and in the “tock” they reduce the manufactured size of the technology. This year was supposed to be the “tock” year, reducing the size of their processor, allowing for more processor cores per chip and increasing performance.
At the end of July, Intel announced that there was a problem with their seven-nanometer processor manufacturing technique that would require some major efforts to overcome. This has resulted in a delay for the general release by between 12-18 months of their seven-nanometer processors. This announcement came on the tail of a previous blow to Intel by competitor AMD.
AMD announced the general release of its Epyc line of processors, with capabilities which overshadow Intel in both performance and price. AMD is still using the older 14-nanometer manufacturing techniques which are less expensive and easier to maintain consistent manufacturing. Due to the more streamlined design of their processor, they are able to achieve a massive 64-cores per die, in comparison to Intel maxing out at 32-cores. The AMD 64-core processor is also around 40 percent less expensive than Intel.
One of the major differences in technology that will help to keep Intel from failing is the interconnect technology called the Universal Processor Interconnect (UPI). UPI is a method of allowed multiple chips in the same computer board to communicate with each other at high speeds. Intel has an eight way UPI link on their enterprise class processors, but due to the internal design of the AMD processor, they are limited to only two UPI links between chips.
So what does the number of UPI links mean for server manufacturers? First of all you can connect eight processors together in tight configuration with eight links, but you can only connect two processors in a single server with two links. This means the largest number of cores in a single AMD server is limited to 128, where Intel can supply servers with 256-cores. These larger servers will keep the processor giant from failing, unless AMD can figure out a way to increase the number of UPI links.
At the same time, the other processor competitor, ARM, released its latest server processors with much lower power consumption than any competitor. ARM has been the leader in overall processor sales for over a decade, but not in the server market. Nearly every cellular phone in production is utilizing ARM processors. Next week I will talk more about ARM and where it is heading.
Stay safe and learn something new.