Silicon restructure

As a consumer, if you are in the market for laptops, desktops you are going to compare the mac to PC and linux. If you are checking the hardware and compare new M1 mac laptops, mini desktops to PC, linux, chrome offerings, immediate question is going to be where are my extra physical cores, neural engine (dedicated processor for machine learning needs), low power consumption benefits of the processor for battery life, reduced heat and state of the art IO for the same price. M1 is killing the benchmarks in app performance. You can argue app end user experience, speeds on different OS, but reviews and numbers for M1 macs are great. This question becomes new consumer expectation. And this new consumer expectation is going to become a shareholder issue for others.

Computer consumption areas have evolved over the years from servers, desktops, laptops to hand-held, wearables. Adoption of smart phones since their introduction has been phenomenal. The technology that goes into them is challenging the traditional laptops and desktops in forms. It’s possible any product you buy will have some form of embedded computers in the future.

Let us look at today’s consumption areas of computers. 

Even smaller devices (e.g IOT) | Smaller Devices (e.g. watch) | Phones | Tablets/Pads | Laptops | Desktops | Gaming Desktops | Servers | Custom Servers

Need from the processor on phones and smaller devices end is reduced power consumption, reduced heat, reduced size. It’s not clock speed and processor multl-threading. Computer forms were adjusted to CPU, GPU, SOC offerings in the past but it is changing the other way. Now this is getting popularly referred as coming age of custom silicon. 

It’s not just clock speed even for laptops and desktops. That segment seems to be the need for gamers and power hungry scalers. M1 benchmarks for single threads are starting the debate of physical cores vs virtual threads as benchmarks looks great for single threads. So, when it comes to CPU it comes down to packaging density of transistors (10nm,7nm,5nm..), cores (more physical ones), machine learning processors (edge) and SOCs (all of them in a single unit).

The speed of IO connecting the processor to disks, displays and devices is increasing rapidly. M2 solid state disk speeds over pc-ie 3.0 are 32 gb/s and it’s small form compared to hard drives are amazing. This speed is expected to increase with pc-ie 4.0 and 5.0. Compare this to 6gbps SATA that your old hard drive were connected to. When it comes to connecting displays, it’s displayPort, thunderbolt and hdmi for 4K, 6k, 8k displays. Displayport 1.0 speeds is at 10.8 gb/s. Displayport 2.0 is at 80 gb/s. HDMI 1.4 speeds is at 10.2 gb/s. HDMI 2.0 is at 18 gb/s. HDMI 2.1 is at 48 gb/s. Thunderbolt 2 is at 20 Gbps. Thunderbolt 3 is at 40 gb/s. Compare to this old DVI at 9.9 gb/s. Network IO was 1 gb/s now it has come to 10 gb/s. All this means data moves at a faster rate which is needed for software performance. Forms of connectors will matter in final products. When it comes to software performance, it’s not just the processor but latest IO as well. That’s what the consumers see and want.

Needs from the processor in each of consumption areas are different, so should the silicon and the kernels. So, what does that mean to the companies from processor designers to computer manufacturers. My opinion - not just reinvent the chip and products for edge but restructure for consumption areas.