Why are all cpus built with cores all at the same speed?

[jiminy]

In general if you are buying a new desktop pc you would determine which processor to buy by what your expected workload will be. Performance in games tends to be determined by single core speed, whereas applications like video editing are determined by number of cores.

 

In terms of what is available on the market - all the CPUs seem to have roughly the same speed with the main differences being more threads or more cores.

Two examples from each architecture:

• Intel Core i5 7600k, 3.80 GHz, 4 cores
• Intel Core i7 7700k, 4.20 GHz, 4 cores, 8 threads
• AMD Ryzen 1600x, 3.60 GHz, 6 cores, 12 threads
• AMD Ryzen 1800x, 3.60 GHz, 8 cores, 16 threads

 

Why are there no variants with two different 'sized' cores?

 

For examples sake consider the following scenarios:

1.  CPU1 - 4 Cores @ 4.0GHZ
2.  CPU2 - 2 Cores @ 4.0GHZ & 4 Cores @ 2.0GHZ
3.  CPU3 - 8 Cores @ 2.0GHZ

Each core would have the same instruction set, and each would have a base clock of 100mhz so the difference is in the frequency multiplier.  They would also be able to lower the frequencies up and down according to load.

 

Then: 

• CPU1 should excel at workloads which are single-threaded
• CPU3 should excel at workloads which are multi-threaded
• CPU2 is a happy medium of two extremes (provided you are able to make the right thread hit the right core)

 

I know that ARM did a similar thing with their big.LITTLE tech, but their focus was on battery-saving, which is not of concern here.

 

I think it basically comes down to the question: Is it possible to make two cores with a given speed take up less space than a single core which is twice as fast?

 

 

 

[Megah3rtz]

First off most AMD Ryzen CPUs have lower clockspeed due to their architecture. Don't write them off because of it. 

[jiminy]

Just now, Megah3rtz said:

First off most AMD Ryzen CPUs have lower clockspeed due to their architecture. Don't write them off because of it. 

Erm ... I didn't.

[Drak3]

On x86, there simply is no need for big.LITTLE schemes for stock clocks. There would also be little point to handicapping random cores on higher core count chips with some schedulers, like Windows', moving threads between logical processors, meaning that a single core perf loving program can randomly start performing worse than it should. Also, most current x86 chips, at stock settings, will dynamically change the clockspeeds based on the needs of software, thermal headroom, and power draw.

[Damascus]

This actually already exists, we have high core low clock xeons and ultra fast low core i3's (stupid i3's but still) 

 

The current mainstream flagships ARE (7700k/1700) the happy medium.

[uncfan07]

I believe the reason is why have underclocked cores? Phones have the big-little architecture because of power consumption. The lower clocked cores do the remedial tasks for less power. On a desktop that has basically unlimited power has no need for that. Why not run all cores at maximum

[PianoPlayer88Key]

Also I've seen different cores on the same CPU boost to different speeds simultaneously.  For example on my 4790K I might see Core #0 at 799 MHz, Core #1 at 4397 MHz, Core #2 at 1399 MHz, and Core #3 at 3197 MHz, all at the same time.  (Of course they're quickly switching around speeds & cores on-the-fly.)

[Drak3]

7 minutes ago, Damascus said:

This actually already exists, we have high core low clock xeons and ultra fast low core i3's (stupid i3's but still) 

 

The current mainstream flagships ARE (7700k/1700) the happy medium.

OP's talking about big.LITTLE x86, and having just a big and just a LITTLE variant as well as the big.LITTLE.

1 minute ago, uncfan07 said:

I believe the reason is why have underclocked cores? Phones have the big-little architecture because of power consumption. The lower clocked cores do the remedial tasks for less power. On a desktop that has basically unlimited power has no need for that. Why not run all cores at maximum

x86 chips can adjust their dynamic clocks to meet any and every power constraint they need, as well as to match load and thermals. ARM's big.LITTLE cores are static clocks that can't change, so having low performance cores that consume little power, and high performance cores that suck it down, and switching between them is necessary.

Also, note than on an octocore Snapdragon, only 4 cores can be used at any given time, same with hexacore SD chips, only 4 can actually be used at once. In contrast, every core that can be used on an x86 chip, will be used.

3 minutes ago, PianoPlayer88Key said:

Also I've seen different cores on the same CPU boost to different speeds simultaneously.  For example on my 4790K I might see Core #0 at 799 MHz, Core #1 at 4397 MHz, Core #2 at 1399 MHz, and Core #3 at 3197 MHz, all at the same time.  (Of course they're quickly switching around speeds & cores on-the-fly.)

So much this. Dynamic clocking, the exact reason big.LITTLE is worthless on x86.

[xitywampas]

10 minutes ago, jiminy said:

Is it possible to make two cores with a given speed take up less space than a single core which is twice as fast?

 

Strictly speaking of Intel(AMD is probably the same). When they make a CPU it is done on mass. It would be harder and more time consuming to make different cores on the same die. Also most lines(kabbylake, skylake) of processors are the same processor cores. The clock difference comes from quality control. A really good processing core will go in a high clocked k. A lower quality core will go in a lower clock non k processor. https://www.quora.com/How-does-Intel-design-and-produce-so-many-models-of-CPUs

So the different cores are physically all the same size in i3, i5 or i7 in a given line. Besides is a modern CPU really that big?

[jiminy]

1 minute ago, PianoPlayer88Key said:

Also I've seen different cores on the same CPU boost to different speeds simultaneously.  For example on my 4790K I might see Core #0 at 799 MHz, Core #1 at 4397 MHz, Core #2 at 1399 MHz, and Core #3 at 3197 MHz, all at the same time.  (Of course they're quickly switching around speeds & cores on-the-fly.)

This can apply to all three CPU examples that I listed. The difference would be apparent at high load.

 

 

4 minutes ago, uncfan07 said:

I believe the reason is why have underclocked cores? Phones have the big-little architecture because of power consumption. The lower clocked cores do the remedial tasks for less power. On a desktop that has basically unlimited power has no need for that. Why not run all cores at maximum

The cores wouldn't be 'underclocked'. You could have more cores (potentially w/o taking up as much space as another 'big' core).

 

 

12 minutes ago, Damascus said:

This actually already exists, we have high core low clock xeons and ultra fast low core i3's (stupid i3's but still) 

 

The current mainstream flagships ARE (7700k/1700) the happy medium.

Would CPU2 not be equally good at single thread performance, but better at multi thread performance?

 

 

 

[Drak3]

20 minutes ago, jiminy said:

Would CPU2 not be equally good at single thread performance, but better at multi thread performance?

Nope. It'll only have the same single core perf as CPU 1 on 2 cores, and performance doesn't scale linearly as you add cores. If it did, it'd still be the same multicore performance as both CPU 1 and 3, as the total theoretical perfomance would equate to a 16GHz single core on all of them. But it isn't, the more cores you add, the higher the latency you incur at multicore, and scaling takes a minor hit with each core, and scheduling those cores would be a nightmare.

[jiminy]

1 minute ago, Drak3 said:

Nope. It'll only have the same single core perf as CPU 1 on 2 cores, and performance doesn't scale linearly as you add cores. If it did, it'd still be the same multicore performance as both CPU 1 and 3, as the total theoretical perfomance would equate to a 16GHz single core on all of them. But it isn't, the more cores you add, the higher the latency you incur at multicore, and scaling takes a minor hit with each core, and scheduling those cores would be a nightmare.

I meant as a comparison to the stated existing happy mediums: 7700k / 1700

[jiminy]

18 minutes ago, Drak3 said:

Nope. It'll only have the same single core perf as CPU 1 on 2 cores, and performance doesn't scale linearly as you add cores. If it did, it'd still be the same multicore performance as both CPU 1 and 3, as the total theoretical perfomance would equate to a 16GHz single core on all of them. But it isn't, the more cores you add, the higher the latency you incur at multicore, and scaling takes a minor hit with each core, and scheduling those cores would be a nightmare.

 

14 minutes ago, jiminy said:

I meant as a comparison to the stated existing happy mediums: 7700k / 1700

 

And I misread 7700k as 7600k.

 

 

[Megah3rtz]

1 hour ago, jiminy said:

Erm ... I didn't.

I know just a general statement.