Clock Speeds, Gaming, and how the CPUs ideal workload is determined

[Shiruxriu]

I think I've never found a result for this and every time I search my google foo frustrates me to no end. The basic question that I want to set up a conversation for is put simply, when does clock speed matter? To explain, I keep hearing from almost every build question on the forum, usually by the same forum users, that the Intel i7-8700k is the best gaming cpu, but as far as I understood this was due to it's single thread performance. So why? Does this tie into clock speeds? I see that among even the most poorly binned chips, it can hit 4-5 GHz without LN2. Now in saying all this a CPUs performance is not just Clock speed but also I wonder, why the 8700k is considered the best Gaming CPU, and how much Clock Speed plays into that. But also, when does Clock speed for a processor cease to be important for gaming purposes, including in today's AAA titles. Anyone have some info?

[Jurrunio]

Clock speed and instruction per clock (in multiplication) determines single core/thread performance. Coffee Lake ticks both boxes, hence does the best in this category. AMD's bulldozer architecture and relatives tick the frequency box, but is garbage in the IPC department, hence single core performance suffers (heavily). Zen architecture gets good IPC (matches if not exceeds Coffee Lake if your RAM is good), but it doesnt clock nearly as high as Coffee Lake so it does worse in single core performance.

 

Back in the past (about 10 years ago), games use two cores at most (quite often, just 1 core). That's why single core performance means everything and when the highest single core performance = best gaming CPU concept comes out.

 

Over time, games start using more and more cores. However, load on cores isnt spread evenly and there's always 1-2 cores that handle more stuff than others. That's why single core performance nowadays still matter, but is no longer the only considered factor.

 

23 minutes ago, Shiruxriu said:

when does Clock speed for a processor cease to be important for gaming purposes, including in today's AAA titles

by clock speed, you mean single core performance?

 

It will cease to be important when your GPU doesnt keep up.

[Shiruxriu]

1 minute ago, Jurrunio said:

Clock speed and instruction per clock (in multiplication) determines single core/thread performance. Coffee Lake ticks both boxes, hence does the best in this category. AMD's bulldozer architecture and relatives tick the frequency box, but is garbage in the IPC department, hence single core performance suffers (heavily). Zen architecture gets good IPC (matches if not exceeds Coffee Lake if your RAM is good), but it doesnt clock nearly as high as Coffee Lake so it does worse in single core performance.

 

Back in the past (about 10 years ago), games use two cores at most (quite often, just 1 core). That's why single core performance means everything and when the highest single core performance = best gaming CPU concept comes out.

 

Over time, games start using more and more cores. However, load on cores isnt spread evenly and there's always 1-2 cores that handle more stuff than others. That's why single core performance nowadays still matter, but is no longer the only considered factor.

 

by clock speed, you mean single core performance?

 

It will cease to be important when your GPU doesnt keep up.

Okay, Then in keeping with that single core performance is considered top dollar for gaming, say the GPU is as an example a 1080 Ti, when specifically the CPUs Clock speed cease to matter in it's relation to gaming performance? For example you mentioned CPU IPC, but that it's only part of the equation and so I guess, to be clearer, when should high clocks stop being the focus for whether a CPU will do well in gaming in cases of say MOBA's, RPGs, Scenic Games, And High refresh rate Gaming?

[Jurrunio]

17 minutes ago, Shiruxriu said:

when should high clocks stop being the focus for whether a CPU will do well in gaming in cases of say MOBA's, RPGs, Scenic Games, And High refresh rate Gaming?

Cant tell because it swings a lot from game to game. If there's unlimited GPU power, then high clocks will always be better and you will never reach the point of stopping (until you've run out of money)

[Shiruxriu]

1 minute ago, Jurrunio said:

Cant tell because it swings a lot from game to game. If there's unlimited GPU power, then high clocks will always be better and you will never reach the point of 'you should stop because things are getting crazy'

That's a bit frustrating, makes the idea of picking a CPU for multiple uses way harder. Then I guess, what would you recommend to someone as a minimum clock speed and core count for a gaming CPU? Cause I always see recommendations going to "pick X because then you can overclock it and get the best performance available." But I always worry that in my experience, it wont do what I want or even would need it to do. 

[Jurrunio]

2 minutes ago, Shiruxriu said:

That's a bit frustrating, makes the idea of picking a CPU for multiple uses way harder. Then I guess, what would you recommend to someone as a minimum clock speed and core count for a gaming CPU? Cause I always see recommendations going to "pick X because then you can overclock it and get the best performance available." But I always worry that in my experience, it wont do what I want or even would need it to do. 

if practicality is given top priority (I didnt with my previous responses), 6 core CPUs are currently the lowest dollar per day it stays relevant option, out of which will be the cheapest Ryzen 6 core. Applying the same concept to GPUs, a gtx 1070ti/1080 will be the best option as they are fast enough to let you skip the next gen cards for cards 2 generations away (not that I recommend actually buying one when next gen is this close).

 

Ultimately, you still buy things depending on use case and BUDGET. 

[Shiruxriu]

Just now, Jurrunio said:

if practicality is given top priority (I didnt with my previous responses), 6 core CPUs are currently the lowest dollar per day it stays relevant option, out of which will be the cheapest Ryzen 6 core. Applying the same concept to GPUs, a gtx 1070ti/1080 will be the best option as they are fast enough to let you skip the next gen cards for cards 2 generations away (not that I recommend actually buying one when next gen is this close).

 

Ultimately, you still buy things depending on use case and BUDGET. 

Alright, but in that regard, how much should I care about clock speed, obviously I shouldn't get something like 1.6 Octo-core processor and pair it with say the 2080 that is about to come out right?

[Jurrunio]

1 minute ago, Shiruxriu said:

Alright, but in that regard, how much should I care about clock speed, obviously I shouldn't get something like 1.6 Octo-core processor and pair it with say the 2080 that is about to come out right?

If you can afford the best GPU available, then get the CPU with the greatest clock speed, then the highest core count, last the highest thread count. Hence, 8700k atm (or 9th gen, if you can wait since that's supposed to bring 8 core CPUs that clock higher than 8700k)

[Shiruxriu]

Just now, Jurrunio said:

If you can afford the best GPU available, then get the CPU with the greatest clock speed, then the highest core count, last the highest thread count. Hence, 8700k atm (or 9th gen, if you can wait since that's supposed to bring 8 core CPUs that clock higher than 8700k)

Fair enough, I'll be honest, I was looking at the 8700k or 2700x as a starting point, since my system is 5th or 6th gen non k i3. (Really dumb young me.) And then at the high end (Price wise mostly) I'm looking at either the 2950x or an Intel Equivalent. Thus I'm curious how much does clock speed matter to performance for gaming, It's been a long consideration, 1 year with many changes.

[Jurrunio]

7 minutes ago, Shiruxriu said:

Fair enough, I'll be honest, I was looking at the 8700k or 2700x as a starting point, since my system is 5th or 6th gen non k i3. (Really dumb young me.) And then at the high end (Price wise mostly) I'm looking at either the 2950x or an Intel Equivalent. Thus I'm curious how much does clock speed matter to performance for gaming, It's been a long consideration, 1 year with many changes.

usefulness of cores start dropping at 6 cores (no streaming), 8 cores (1080p 30fps streaming) and 10 cores (1080p 60fps), so more cores isnt always better.

[Shiruxriu]

1 minute ago, Jurrunio said:

usefulness of cores start dropping at 6 cores (no streaming), 8 cores (1080p 30fps streaming) and 10 cores (1080p 60fps), so more cores isnt always better.

Alright, my core breakdown intent at the high end would be 8 cores for a server, 4 cores for gaming, and 4 cores for gaming, in that regard, and assuming all core affinity's are set up properly per program (Probably a nightmare), would that have the potential to make the seemingly lower boost of say the 2590x at 4.4GHz worth more for the 3 purposes than letting windows auto assign tasks for the above usage on say a 5.2 GHz 8700k or 4.xGHz of a 2700x?

 

[Jurrunio]

1 hour ago, Shiruxriu said:

Alright, my core breakdown intent at the high end would be 8 cores for a server, 4 cores for gaming, and 4 cores for gaming, in that regard, and assuming all core affinity's are set up properly per program (Probably a nightmare), would that have the potential to make the seemingly lower boost of say the 2590x at 4.4GHz worth more for the 3 purposes than letting windows auto assign tasks for the above usage on say a 5.2 GHz 8700k or 4.xGHz of a 2700x?

 

if you have multiple tasks running then of course 2950x is better than the other two. Make sure you use all 4 memory channels since it's still Ryzen

[Shiruxriu]

1 minute ago, Jurrunio said:

if you have multiple tasks running then of course 2950x is better than the other two. Make sure you use all 4 memory channels since it's still Ryzen

Admittedly if I go that route I won't be able to afford the memory immediately, at best 4 sticks at a time. Now so in terms of that use case, clock speed is less of worry, but what if that is only say part of the time? Like otherwise it's gaming while web browsing and maybe running a game server? Still the same, where despite it's performance the 8700k would be lacking despite it's beastly single thread performance?

 

[Jurrunio]

2 minutes ago, Shiruxriu said:

Like otherwise it's gaming while web browsing and maybe running a game server?

'and', or 'or'? Big difference there

[Shiruxriu]

16 minutes ago, Jurrunio said:

'and', or 'or'? Big difference there

For the sake of discussion let's assume two different answers, one where it is just gaming, game server and web browsing, and one where it's just web browsing and gaming with nothing else really in the background.

 

[Jurrunio]

6 minutes ago, Shiruxriu said:

For the sake of discussion let's assume two different answers, one where it is just gaming, game server and web browsing, and one where it's just web browsing and gaming with nothing else really in the background.

 

All tasks at once: Threadripper or i9, not getting away from HEDT stuff with more than 10 cores.

 

Seperate: 8700k will do

[Shiruxriu]

Just now, Jurrunio said:

All tasks at once: Threadripper or i9, not getting away from HEDT stuff with more than 10 cores.

 

Seperate: 8700k will do

Alright then, so in that regard, the HEDT chips would be better for the intended use case, and therefore, despite it being the best at gaming, for the use case I originally presented, the 2950x or an Intel equivalent would be better. Would it lose performance when not doing the other tasks?

 

[Jurrunio]

1 minute ago, Shiruxriu said:

Would it lose performance when not doing the other tasks?

2950x slower than 8700k in games without other tasks hogging CPU time? Yes. 8700k's ring bus architecture lowers latency compared to mesh architecture used by Skylake-X and Zen, also making it faster in games aside from frequency. You wont be able to clock Threadripper higher just because you're using less cores anyway, Ryzen (16 cores or less anyway) is limited in clock speed by GlobalFoundaries' process node first, not heat

[Shiruxriu]

1 minute ago, Jurrunio said:

2950x slower than 8700k in games without other tasks hogging CPU time? Yes. 8700k's ring bus architecture lowers latency compared to mesh architecture used by Skylake-X and Zen, also making it faster in games aside from frequency. You wont be able to clock Threadripper higher just because you're using less cores anyway, Ryzen (16 cores or less anyway) is limited in clock speed by GlobalFoundaries' process node first, not heat

Oh wow, I was not aware that Threadripper's limits were a manufacturing limit and not a heat one, I was aware of the difference in coffee lake's topology though. So for multicore ships, it is actually not a heat gate that prevents the cores from clocking higher? 

[Jurrunio]

5 minutes ago, Shiruxriu said:

Oh wow, I was not aware that Threadripper's limits were a manufacturing limit and not a heat one, I was aware of the difference in coffee lake's topology though. So for multicore ships, it is actually not a heat gate that prevents the cores from clocking higher? 

24 and 32 core TR2 are limited by heat. For Zen+ (Ryzen 2000 and TR2 are based on) process node wall comes in at about 4.3-4.4GHz. If a Zen+ will hit those clocks only after thermals getting crazy, it will be considered hindered by heat.

 

fyi the intels with more cores than 10 (delidded), 6 (no delid) are limited by heat.

[Shiruxriu]

33 minutes ago, Jurrunio said:

24 and 32 core TR2 are limited by heat. For Zen+ (Ryzen 2000 and TR2 are based on) process node wall comes in at about 4.3-4.4GHz. If a Zen+ will hit those clocks only after thermals getting crazy, it will be considered hindered by heat.

 

fyi the intels with more cores than 10 (delidded), 6 (no delid) are limited by heat.

But are both the intel and amd chips that exceed 10 cores limited by heat or is that more an intel specific limitation? Also is it a design flaw in general or just a design flaw due to lack of capable cooling? This is kind of the meat and potatoes I was curious about, how do the parts play realistically due to their design and technology.

[Jurrunio]

3 minutes ago, Shiruxriu said:

But are both the intel and amd chips that exceed 10 cores limited by heat or is that more an intel specific limitation? Also is it a design flaw in general or just a design flaw due to lack of capable cooling? This is kind of the meat and potatoes I was curious about, how do the parts play realistically due to their design and technology.

no Intel is intel, AMD is AMD. Intel just get heat wall earlier because their process node wall comes in much later in the frequency range and they arent soldered, making heat dissipation more difficult and the heat wall come in earlier.

 

On AMD's side: Not exactly a 'flaw', it just bugs out earlier. Think of it as someone who panics easily, they freak out with less stress.

On Intel's side: Design flaw in using paste under the lid, so cooling the CPU die becomes more difficult.

[Shiruxriu]

1 minute ago, Jurrunio said:

no Intel is intel, AMD is AMD. Intel just get heat wall earlier because their process node wall comes in much later in the frequency range and they arent soldered, making heat dissipation more difficult and the heat wall come in earlier.

 

On AMD's side: Not exactly a 'flaw', it just bugs out earlier. Think of it as someone who panics easily, they freak out with less stress.

On Intel's side: Design flaw in using paste under the lid, so cooling the CPU die becomes more difficult.

So basically on AMDs side of things, the CPU albeit "better" for multiple tasks, still has a working limit due to it's design and manufacturing? Where as with Intel, it's just very improperly designed thermally but it's work ethic is top notch right?

[Jurrunio]

8 minutes ago, Shiruxriu said:

So basically on AMDs side of things, the CPU albeit "better" for multiple tasks, still has a working limit due to it's design and manufacturing? Where as with Intel, it's just very improperly designed thermally but it's work ethic is top notch right?

you can put it like that.

[Shiruxriu]

1 minute ago, Jurrunio said:

you can put it like that.

Hmm, so then when it comes to the whole GHz race, that doesn't matter as much in comparing the two sides but at the same time shouldn't be thrown aside as a consideration right? Does it come down more to just a node design race in the end then?