How did you improve the architecture?

Question

Intel

Hello. In the process of reading Wikipedia: https://ru.wikipedia.org/wiki/FLOPS
I came across interesting numbers:
Intel Pentium 4 2.5-2.8 GHz (2004) - up to 5-5.6 gigaflops
Intel Core i7-4930K (Ivy Bridge), frequencies 3.7-4.2 GHz, 6 cores (2013) - 130-140 gigaflops
For the sake of interest, I decided to bring them to some one value.
For example, Core i7 has 1.3 times the frequency, and 6 cores (not one).
That is, if we translate the performance to parrots of the Pentium 4 system, then we get: 130 / 6 / 1.3 = 16.6 (gigaflops)
Question: how can two processors at the same frequency and one core have a performance that differs three times?
I understand that the question is too abstract, but a certain number of keywords for searching would greatly simplify my communication with Google :)
Well, a separate question: are these parrots too abstract? Google chrome slows me down equally on both processors, despite the difference in performance by an order of magnitude;)

Answer 1

FLOPS are achieved by blocks running on floating point numbers (only with this data). And there are several of them in one core. The rest is vector operations, where arithmetic operations are performed in parallel. It's called SIMD - one command, several data. Intel has several such technologies, but the most used are SSE and AVX of various versions. For example, AVX512 allows you to process 8/16 numbers at the same time. Some commands are double. For example, multiplication with addition. Therefore, it is considered to multiply the performance by a factor of two. Therefore, an application that does not use such technologies will still perform at about the same speed as it did a few years ago. Kernels allow you to parallelize execution / computation, but no more. A single-threaded application will definitely not gain any speed on multi-core processors.

Answer 2

For example - hardwareguide.ru/%D0%BF%D1%80%D0%BE%D1%86%D0%B5%D1...

If we compare the norms of 65 nm and 45 nm, then ... a 30% reduction in power dissipation during switching, as well as a 20% increase in the switching speed of the transistor.