The world and technology is fast moving. Just 10-20 years ago Network companies would use the same CPU/memory/ASIC composition for years and only change a few technical or physical aspects of a router or switch.
Today is a lot different. While you think the basic operations of routers are just looking up networks in the routing tabel and forward packets, it's become a lot more complicated. The sheer amount of packets have changed from maybe thousands or hundreds of thousands to millions or even billions on the internet backbone and on top of that, the demand for more bandwidth has exploded. Routing tables have become even larger and local networks can span for thousands of miles across the world.
New services have been introduced. MPLS routers need to support full QoS (Quality of Service), Multicast has changed from nice to have, to need to have. Netflow has been perfected over the years and is used today to give realtime insights into how end users experience delay/throughput/jitter/lag etc. Everything can be emulated and this costs A LOT of CPU power.
Sadly threat control has been an everyday thing. Hackers will try to compromise security and/or try to shut down companies using DoS/DDoS attacks. But the attacks have also become more clever. An attack can be emulated to look like "normal" traffic or user usage, which ultimately will break any system. Therefore, to counter these forms of attacks, routers have been equipped with the ability to see patterns in network traffic. It doesn't mean packet inspection as this becomes more and more impossible, but network patterns is ways of analyzing how attacks evolve and how they are used. Again, this costs CPU.
There are a lot of other things, that i've probably forgotten.
A small comparison:
In the early 2000's Cisco released the Cisco 12000 Series Router as a Service Provider Core. These routers used the same sort of design as some of the most powerful supercomputers in the world, a design that ties many different processors together with a series of extremely fast switches. The 12000 series used 200-MHz MIPS R5000 processors, the same type of processor used in the workstations that generate much of the computer animation and special effects used in older movies. The largest model in the 12000 series, the 12016, used a series of switches that could handle up to 320 billion bits of information per second and, when fully loaded with boards, move as many as 60 million packets of data every second. Beyond the computing power of the processors, these routers can handle so much information because they are very highly specialized. Relieved of the burden of displaying 3-D graphics and waiting for mouse input, modern processors and software can cope with amazing amounts of information.
Today though, The Cisco Network Convergence System (NCS) 6000 Series Router, for instance, provides 16 Tbps of full-duplex network bandwidth through 8 line cards. Each card delivers up to 2 Tbps throughput using a mix of 10-Gbps and 100-Gbps interfaces per card. The Cisco NCS 6008 system also provides modular optics options to meet a wide range of distance requirements. In a multi-chassis configuration, the Cisco NCS 6008 system can be expanded to support up to 256 Tbps of full-duplex forwarding throughput.
The Cisco NCS 6000 Series is powered by the Cisco nPower Network Processor Units (NPU), an innovative programmable forwarding application-specific integrated circuit (ASIC). The Cisco nPower X1 is designed to deliver the industry’s first zero packet loss (ZPL) and zero topology loss (ZTL) In-Service Software Upgrade (ISSU) capability.
Oh and the NPU is:
- First true 400 Gigabits-per-second (Gbps) throughput from a single
chip, to enable multi-terabit network performance. All packet
processing, traffic management and input/output functions are
integrated on a single nPower X1 and operate at high performance and
- Highest-performing programmable control designed to seamlessly handle
hundreds of millions of unique transactions per second. The nPower's
industry-leading processing architecture is purpose built for
machine-driven events and ultra-high-definition video applications.
- With 4 billion transistors on a single chip, the nPower X1
implementation achieves unprecedented levels of performance,
functionality, programmability, and scale for a network processor.
- Enables solutions with eight times the throughput and one quarter the
power per bit compared with Cisco's previous industry-leading network
The same story goes for switches. More features, technology, throughput, faster ASIC's etc. All upgraded to meet the demands of today networks.