Why do more cores mean slower performance?

[OnionRings]

I just watched this 

and it was way worse than I expected, why does this happen? Is it because the cores are all super slow cores? I assume this is also why people say ryzen for streaming but intel for gaming. 

[Crunchy Dragon]

Higher core count at slower speeds is better for rendering and other similar tasks.

Higher clock speed is better for gaming, but clock speed isn't everything.

 

People say Intel for gaming because their CPUs currently have the highest single-threaded performance, which usually dictates how well games will run. Ryzen is recommended for streaming because it's just better value; you get more cores for the price, but still good enough performance for gaming.

 

Although Intel isn't always better for gaming, as shown by this: https://www.techspot.com/review/1655-core-i7-8700k-vs-ryzen-7-2700x/

The performance difference isn't nearly as high as most people like to make it out to be.

[minibois]

If we just think linearly, twice the cores use twice the amount of power, thus twice amount of heat. So to counteract that, you slow down your cores, so the CPU doesn't get way too hot.

By that logic, the more cores you have; the slower you have to make them to have a CPU that makes sense. But games like single core performance most (they really don't use more than say.. 8 cores nowadays), so a CPU with like 1/8th the cores, which are way faster is more logical for a game and thus performs better.

 

It's just about what the software uses; which games don't use that many cores. What the Xeon Phi is meant for, are workloads that can use all cores. In those cases the single core performance is still a disadvantage, but the amount of cores you have makes up for that.

 

TL;DR: Because they have to slow down per core performance, which is what games and many other programs use most.

[JohnDongus]

They are judt very slow cores as you said, if you have a 4core cpu that runs at 5ghz and a 16core cpu that runs at 5ghz then the one with more corees will be a lot faster. 

 

But sometimes applications like single threads with a lot of speed rather then may with slower speeds. 

[porina]

Is that the one he used a Xeon Phi? The number of cores isn't the problem, it is the type of core. It is highly optimised for floating point performance, and for everything else, it is basically a low clock atom. It's like taking a drag racer to a rally and wondering why it isn't fast.

[AngryBeaver]

It comes down to the workload and how it is coded. If a game or application can take advantage of the high core count then you will have great performance, but in gaming especially dx11 games things will be singled threaded like the DX11 draw calls... this means that higher single thread performance is king so a quad core cpu would out perform a 64c with lower per thread performance.

 

Now some of the reason these cores are slower has to do with heat, some of it energy use and efficiency, but mostly at what the target use is. These high core cpu's are meant for heavy workload environments where virtualization will also be popular. So while they might not be fast on their own when broken up in to 128 threads they can tackle a lot of actions simultaneously that wouldn't be possible on the higher performance lower core count machines.

 

That is a very simple explanation of it, but in the end it comes down to what tasks that cpu is aimed at.

[Simon771]

14 hours ago, Crunchy Dragon said:

Higher core count at slower speeds is better for rendering and other similar tasks.

Higher clock speed is better for gaming, but clock speed isn't everything.

 

People say Intel for gaming because their CPUs currently have the highest single-threaded performance, which usually dictates how well games will run. Ryzen is recommended for streaming because it's just better value; you get more cores for the price, but still good enough performance for gaming.

 

Although Intel isn't always better for gaming, as shown by this: https://www.techspot.com/review/1655-core-i7-8700k-vs-ryzen-7-2700x/

The performance difference isn't nearly as high as most people like to make it out to be.

I do agree with you, but there are some games that really prefer Intel strong cores ... In my case that game is (well it was) Black Desert online. With Ryzen 7 1700 event at 3,85GHz on all cores my FPS would drop to 35-50 in cities, but as soon as I went to i5 8600k at 5GHz, my FPS never dips below 60, and it most of the time at 75.

Difference between 35 and 60 fps is a big deal for me, so that was a nice improvement.

But it's also worth mentioning, that this scenarios was about 1 year ago. Since then, BDO made new improvements and patches, that fixed issue with Ryzen not performing that well, and now I see users with Ryzen 2700, with 70+ fps in cities and 100fps in other parts of the game.

 

I'm still waiting for new Zen version of CPUs that will be able to clock to at least 4,7GHz on all cores. That's when I'll be buying AMD again.

[pstarlord]

To add a bit of personal experience here, not going off of any benchmarks or anything other than my personal experiences here, so take from it what you will. 

 

I've got two systems right now, one is for gaming, the other is for music production, but I have gamed on both of them just for fun, here's what I noticed.

 

System 1:  (music production PC)

Intel i7 7700k - 4 core, 8 thread. - I run this at a light overclock up to 4.5Ghz

32 Gigs of RAM - 2400MHz

 

System 2: (gaming PC)

Ryzen R7 1700x - 8 core, 16 thread - Overclocked to 3.9Ghz

16 Gigs of RAM - 3000MHz

 

I swapped my Vega 56 between the two and did a little comparing, and in gaming, there was little to no difference.  Witcher III, Skyrim, Shadow of the Tomb Raider, Evil Within II, Final Fantasy XIII, Nier: Automata...none of them showed a noticeable difference in performance between the two systems.  I can't speak to technical reasons why, this was just my experience. 

[JacobFW]

To really understand this you need to understand a little bit of the physics behind this.

 

One of the primary factors limiting performance in any kind of chip is the heat output.  If too much heat is being generated without being removed quickly enough, the chip will overheat and die.  The heat output of a processor is related to the amount of power consumed and the efficiency with which it uses that power.

 

The equation that roughly describes the amount of Power consumed by a CPU is  P=C*F*V^2, where P is power, C is capacitance (i.e. more or less how many electrons the cpu can contain), F is frequency, and V is voltage.  

 

The important thing to notice here is Voltage, which is squared in that equation, meaning that a small change in voltage leads to a much larger change in power consumption, and therefore heat generated.  

 

This is important, because as the frequency of a CPU increases, the voltage almost always has to increase as well.  You can observe this for yourself if you download a program like CoreTemp.  When the computer is idle, the CPU will go into a low power mode, where the frequency and voltage will drop, then jump up as soon as you start doing something.  So if you go back to the equation, a higher clocked cpu not only has a higher F, but also a higher V, and again because V is squared, this means that a slight bump in frequency will lead to a much larger increase in power consumption, and therefore generated heat.

 

This is why traditionally the higher end Xeons have much lower clock speeds.  They sacrifice per core performance because it also means much much lower heat generated per core, meaning they can cram more cores into a single cpu die, without the risk of the entire think melting the moment it booted up.