What's with the CPU speed thing?

[gazz]

Whats the deal with processors nowadays... years ago it was total speed, a bit like the car engine size thing, bigger is faster.
my first pc having a 333mhz cpu... then it was all out faster and faster speeds.
till it got to about 3.5ghz and it seemed to stall.

Nowadays they seem to even be going backwards. i'd have expected us to be on like terrrahertz cpu's by now.

I know there's the core thing.. but again that's stuck on like 6 or 8 cores it seems.

What do we have to look for to tell if the latest 3.5ghz 6 core processor is any better than a 5 year old 3.5ghz 6 core processor?

i remember there being things about the die size, and i imagine the speed of data transfer through the cores etc, but that is not clear when processors are still sold using things like '3.5ghz, 6 core'

So how do i read the other specs to determine the better processor?

 

[ExRayTed]

What you're talking about is IPC improvement - https://www.tomshardware.co.uk/ipc-cpu-definition,review-34520.html for a rudimentary introduction

IPC is just one area CPUs have improved, there'll be many more that will have had an impact over the last few years but I do agree improvements from one generation to the next have slowed dramatically.

 

[EsaT]

Problem of trying to increase clock speeds is in speed of signaling and transistor switching.
Besides causing errors in logical operation when pushed too high, those also increase power consumption.
And compromising everything else for designing for maximal clock speed simply doesn't work.
High Instructions Per Clock performance with average clocks is lot better for general purpose CPU performance and energy efficiency.
Just like Intel's NetBurst/Pentium 4 (originally visioned to reach 10GHz) and AMD's "Fail"dozer becoming fiascos only good for converting electricity to heat showed.

Since Pentium 4 architectural improvements have made more and more instructions to be executed in single clock cycle.
Also over the years new instructions have been added, doing more complex things in one cycle.

But lately Intel really hasn't done much anything on that.
Sinche "4th generation" Haswell there's been only one architectural improvement: 6th gen Skylake.
Intel CPUs since that have bene about tweaking clock speed higher and then adding more cores, when Ryzen stopped Intel from milking us with decade old cour cores.
"7th" gen Kaby Lake was just higher clocked Skylake.
And "8th" gen Coffee Lake just increased core count without affecting to efficiency/performance of individual core.
Coming 9000 serie is again increase of core count.


Really except for adding new instructions, increasing IPC of CPU core has become exponentially harder and incremental.
And additional complexity from new instruction execution units increases "bureaucracy" of CPU, adding also some power consumption.
So for general purpose CPU adding new instructions has to be considered carefully, especially when those don't help any with existing code/software.

Also with materials/manufacturing tech limiting clock speeds there's no easy path ahead in that.
Time when power consumption could be allowed to constantly increase in chase of higher clock speeds is far behind.

So increasing core count is lot easier performance increase path for anything which can be multithreaded.

 

[StarShock]

clock speeds was always a bit of marketing thing anyway, was a nice simple number that you could point to and go "this ones faster cos the number is bigger" but for the most part it only guaranteed to work when comparing like for like. If you look back at the old Athlon 64s they were numbered for the equivalent intel clock speeds for example the Athlon 64 3800+ ran at 2.4 ghz but was meant to run equivalent to a Pentium 4 at 3.8 ghz, and just about managed it. At the time AMD had a huge IPC advantage.

 

[HangTime]

To be honest there is no easy way to compare, you just have to look at some benchmarks. I have no idea how a 'novice' would go about choosing a processor these days, but at least I guess the internet is widely used now so research options are available rather than getting stitched up in PC World.

To be fair, IPC improvements have been around for ages so you'd have 486>Pentium>P2>P3 etc being faster than predecessors even without clockspeed increases (even within those they would release revisions with more cache, higher bus speed etc). P4 was a bit of an anomaly as it focused on clockspeed over IPC (P3 was faster clock-for-clock than the original P4 iirc) but as mentioned above at the same time AMD was pushing much slower clocked cpus with better IPC.