Why does it seem that games use 5950x less than they do 12900K ?

[papajo]

So in the above video 12900K uses 8c 16t (Hyper Threading) 
and 5950x 16c/16t

And in all games the CPU usage for the ryzen cpu seems noticeably lower. 

I mean e.g if we were to open task manager on both systems (set up as they are in the video) we would see 16 boxes for each system 

And if for example 8 of them boxes (on each system respectively) were at 100% then the total CPU usage would show up as 50% on both. 


On top of that since Intel is clocked higher it should show an even lower percentage use (that's not related to the task manager example I gave, to put it simply  e.g if a game needs 1000 things/second and one CPU can do 2000things/second when @2.5GHz then it will show up as 50% usage if it can do 4000 things/second @ 5GHz then the usage will show as 25%) 


Is Intel having some extra optimization or relationship with the devs of the depicted games (e.g as nvidia does with Nvidia Gameworks SDK for devs) ? 
 

[An0maly_76]

I believe this is a simple difference in how AMD's processor cores are operated and controlled versus Intel. Intel's Hyperthreading technology is a bit different than AMD's PBO.

[LIGISTX]

22 minutes ago, papajo said:

So in the above video 12900K uses 8c 16t (Hyper Threading) 
and 5950x 16c/16t

And in all games the CPU usage for the ryzen cpu seems noticeably lower. 

I mean e.g if we were to open task manager on both systems (set up as they are in the video) we would see 16 boxes for each system 

And if for example 8 of them boxes (on each system respectively) were at 100% then the total CPU usage would show up as 50% on both. 


On top of that since Intel is clocked higher it should show an even lower percentage use (that's not related to the task manager example I gave, to put it simply  e.g if a game needs 1000 things/second and one CPU can do 2000things/second when @2.5GHz then it will show up as 50% usage if it can do 4000 things/second @ 5GHz then the usage will show as 25%) 


Is Intel having some extra optimization or relationship with the devs of the depicted games (e.g as nvidia does with Nvidia Gameworks SDK for devs) ? 
 

There is so much more to it then ghz in your ghz portion of this. The way in which the CPU’s process data, prefetch info, pass data around internally… this is why you can’t compare Intel ghz to AMD ghz; it’s just not apples to apples. 
 

But more than anything for your specific question, the 12900k only has 8 cores, the other 8 are are “just” hyper threading. The AMD CPU has 16 actual cores. A core > virtual thread. Again, it’s complicated, but full on cores will out perform virtual threads. 

[papajo]

6 minutes ago, An0maly_76 said:

I believe this is a simple difference in how AMD's processor cores are operated and controlled versus Intel. Intel's Hyperthreading technology is a bit different than AMD's PBO.

  

5 minutes ago, An0maly_76 said:

I believe this is a simple difference in how AMD's processor cores are operated and controlled versus Intel. Intel's Hyperthreading technology is a bit different than AMD's PBO.

PBO is an OC setting if you mean SMT (τηε AMD's counterpart for "HT" )  then it is disabled in this video. 

[papajo]

3 minutes ago, LIGISTX said:

But more than anything for your specific question, the 12900k only has 8 cores, the other 8 are are “just” hyper threading. The AMD CPU has 16 actual cores. A core > virtual thread. Again, it’s complicated, but full on cores will out perform virtual threads. 

Yes but the CPU usage is measured by logical core count doesnt matter if one core is faster than an other core (so physical core vs ht logical core) that's why I gave the task manager example to explain that. 

An other example that would prove that is if you use all the cores (e cores included) so e.g if 10 cores are in use then the CPU usage would show as 50% not less (because the 2 e cores are slower than P cores) 

[An0maly_76]

2 minutes ago, papajo said:

  

PBO is an OC setting if you mean SMT (τηε AMD's counterpart for "HT" )  then it is disabled in this video. 

My point is that the 11900K's core / thread structure is different than the AMD's, and so are the methods used to control them.

[papajo]

1 minute ago, An0maly_76 said:

My point is that the 11900K's core / thread structure is different than the AMD's, and so are the methods used to control them.

The load should be the same though(unless the specific game engines in the video have an intel optimization) , so in any case (since amd is clocked lower) it should have a higher percentage use. 

[LIGISTX]

6 minutes ago, papajo said:

Yes but the CPU usage is measured by logical core count doesnt matter if one core is faster than an other core (so physical core vs ht logical core) that's why I gave the task manager example to explain that. 

An other example that would prove that is if you use all the cores (e cores included) so e.g if 10 cores are in use then the CPU usage would show as 50% not less (because the 2 e cores are slower than P cores) 

Except that a full core will do more work in less time then a virtual thread… so utilization will be lower on a system with all real cores vs one with virtual threads and true cores. But, again, you can’t compare Intel to AMD apples to apples like this. They are simply not comparable. 

[LIGISTX]

4 minutes ago, papajo said:

The load should be the same though(unless the specific game engines in the video have an intel optimization) , so in any case (since amd is clocked lower) it should have a higher percentage use. 

That is precisely incorrect. The CPU’s don’t process data the same. There is so much more advanced stuff going on under the hood of a CPU then simple “I need to process 1 billion things, and my cpu is running at 5 billion cycles a second, so it will take 1/5th a second to compute”. That is so far beyond an oversimplification it’s barely even useful in todays CPU’s. 

[papajo]

15 minutes ago, LIGISTX said:

That is precisely incorrect. The CPU’s don’t process data the same. There is so much more advanced stuff going on under the hood of a CPU then simple “I need to process 1 billion things, and my cpu is running at 5 billion cycles a second, so it will take 1/5th a second to compute”. That is so far beyond an oversimplification it’s barely even useful in todays CPU’s. 

The game engine load is the same though (unless the game engines are optimized for intel that is)  which means that the only way for AMD to show less percentage use especially while clocked lower is by accepting two criteria.

1) the AMD CPU is way faster, WAY faster (given the percentage difference and that it runs at lower clocks) in terms of IPC

2) The game load is finite and already saturated by the current system at hand (so e.g even if intel enabled e-cores and OCed higher the rest of the CPU and extra clocks wouldnt be used by the game) 

But I am quite sure that AMD has lower IPC or best case scenario comparable. 

[papajo]

22 minutes ago, LIGISTX said:

Except that a full core will do more work in less time then a virtual thread

Which affects the frame time (and in turn FPS value) not the percentage value of CPU usage. 

E.g if we run a cinebench bench instead of a game and used the single core option, the intel would do work faster but the usage would still be the same (6% if intel was set 8c 16 t and amd 16c16t) 

[LIGISTX]

2 minutes ago, papajo said:

Which affects the frame time (and in turn FPS value) not the percentage value of CPU usage. 

This assumes infinite GPU power and no loss post CPU….

 

Your trying to make a comparison that just doesn’t really exist. The CPU’s literally process data differently. CPU’s make assumptions and pre fetch data, store data on cache (AMD has a lot more cache), access RAM different, pass data between itself very differently, etc etc. 
 

Looking at a simple utilization graph doesn’t actually tell you very much about what’s going on under the hood.

 

I am not intelligent enough on this to explain to you the exact ways in which this comparison doesn’t make sense because it’s extremely complicated, but I do understand CPU architecture enough to know it’s just not that simple, and honestly doesn’t matter. What task manager reports is not all that useful. You can have a load that shows as 100% utilization but only draws a fraction of the wattage of what the CPU is capable of pulling… so obviously the CPU isn’t actually putting its “full force” behind completing that task. Again, there is just so much more to this. 

[papajo]

9 minutes ago, LIGISTX said:

This assumes infinite GPU power and no loss post CPU….

 

Your trying to make a comparison that just doesn’t really exist. The CPU’s literally process data differently. CPU’s make assumptions and pre fetch data, store data on cache (AMD has a lot more cache), access RAM different, pass data between itself very differently, etc etc. 
 

Looking at a simple utilization graph doesn’t actually tell you very much about what’s going on under the hood.

 

I am not intelligent enough on this to explain to you the exact ways in which this comparison doesn’t make sense because it’s extremely complicated, but I do understand CPU architecture enough to know it’s just not that simple, and honestly doesn’t matter. What task manager reports is not all that useful. You can have a load that shows as 100% utilization but only draws a fraction of the wattage of what the CPU is capable of pulling… so obviously the CPU isn’t actually putting its “full force” behind completing that task. Again, there is just so much more to this. 

I am not refuting that they have different architectures. 

But you are (not out of malice I am not accusing you for something) just making things more complicated than they are. 

Yes how they work is complicated. But the CPU percentage usage as a value isnt really complicated. 

Again (You can confirm this yoursefl if you have two PCs with different CPU architectures )

If you have CPU 1 and CPU 2 and run a cinebench test choosing the single core option. 

Yes both CPUs  (if they have a different architecture) will handle the load differently
Yes one CPU will be faster at handling the same load than the other and finish first and get a higher score. 


But BOTH CPUs will display the SAME CPU % usage value  (given that they have the same number of logical cores in e.g task manager) 

[LIGISTX]

1 minute ago, papajo said:

I am not refuting that they have different architectures. 

But you are (not out of malice I am not accusing you for something) just makes things more complicated than they are. 

Yes how they work is complicated. But the CPU percentage usage as a value isnt really complicated. 

Again (You can confirm this yoursefl if you have two PCs with different CPU architectures )

If you have CPU 1 and CPU 2 and run a cinebench test choosing the single core option. 

Yes both CPUs  (if they have a different architecture) will handle the load differently
Yes one CPU will be faster at handling the same load than the other and finish first and get a higher score. 


But BOTH CPUs will display the SAME CPU % usage value  (given that they have the same number of logical cores in e.g task manager) 

Yes, because that is a CPU only test, it doesn't rely on anything outside of the CPU.... like, oh I don't know, something complicated like a GPU. The speed at which the GPU actually requests draw calls of the CPU is going to be dependent on how fast it can spit out FPS. Cinebench is going to hit a CPU core for everything it has, it WILL use everything the CPU has to crunch that data as fast as possible, and there is no lack of data to crunch, as soon as the CPU finishes what it was asked to compute, another thing is being shoved in right behind it. So, yes, the CPU is fully utilized, both via what task manager will report to you, and it can be measured by the wattage used by the chip as it does the computation. But just because task manager says its 100% utilized doesn't actually mean the CPU is working "100% hard". This is not the same as a game though, again, the GPU will ask for data as it needs it, cinebench will shove more rendering at the CPU until the entire scene is fully rendered, but a GPU will not continually push out draw calls because it can only work so fast and will only need data provided to it so quickly. Also, you can see your SSD show as 100% utilized in task manager as its only reading and writing a few megs a second... It is being asked to do many things at once, and it doesn't have the ability to go and try and do more things, so it shows as being 100% utilized...... enough tho its only working at a fraction of its capacity.

 

Another example is a crap coded game like battlefield V. It almost always results in task manager showing 100% utilization.... but my CPU temps are usually in the 60's. When I run a real stress test, temps are in the 80's. This is because battlefield isn't actually taxing the CPU all that hard.

 

But, again, you are trying to compare apples to oranges. The way the CPU's interact with the drivers, chipset, GPU, PCIe bus, etc etc etc is all so different just looking at a windows graph of % utilization is not at all indicative of anything. Your cinebench example is not a real world representation because of the reasons I described, but if you don't believe that, I would suggest you try and find a chip engineer or a software engineer who really understands whats going on under the hood and ask them to explain it better.

[LIGISTX]

15 minutes ago, papajo said:

If you have CPU 1 and CPU 2 and run a cinebench test choosing the single core option. 

Best way I can put this... is a CPU with 16 cores was asked to crunch a set of data as fast as possible on a single core, and a CPU with 16 threads (8 cores 8 virtual threads) was asked to do that same like in cinebench, they would report the same % while under load, but one of them would finish first. Where your logic fails is that your forgetting the time portion of this... That is a bound on this because if you want to take this analogy and apply it to a game, there is a bottleneck, its the GPU.

 

So as the CPU is crunching data, as I said above, its only going to crunch as much GPU as the engine needs it to crunch for the GPU to produce a frame. So if the GPU is more or less imposing a time bound on the system (the bottleneck), the CPU that can do more work per given time (the one with all cores vs virtual cores) will get that little sliver of work done faster, and then it will wait until it needs to provide more data to the GPU. 

 

I hope this makes sense, because I don't know how else to describe this.

[papajo]

You wrote all that stuff that is irrelevant and/or inaccurate. 

CPU usage is based on "busy time" and "idle time" 


Time here is not exactly time as in what you see when looking at your watch, its measured in clock ticks so if a CPU can handle 1000 ticks per second this is its time. 

If now every second on 500 of those ticks it has to process something (being GPU communication, calculating Pi, doing database stuff, mp3 encoding is irrelevant whatever)  thats 500 ticks of busy time out of the total time that is 1000 ticks so the CPU usage will show as 50% 

There is nothing more to it. 

So the only way for this percentage difference to occur in the video is if

A) AMD has way WAY WAY better IPC than intel, or in other words AMD can do "more stuff per tick" so it needs less ticks to do the same stuff<-- which is not true so we exclude that
.

B) the busy ticks (load) caused by the game are finite and saturated already by both systems. 

A AND B must be true in this case. 

The only other explanation is that for some reason (e.g optimization bias) these games keep more busy ticks when played on intel (e.g they give them extra data to compute) 

For which I can see some signs (e.g the intel system uses more Vram and Ram for the same game. which could point to that "giving extra stuff to compute") 

[jaslion]

5 hours ago, papajo said:

The load should be the same though(unless the specific game engines in the video have an intel optimization) , so in any case (since amd is clocked lower) it should have a higher percentage use. 

Not true

 

Say a 9900k will have a higher load than a 12900k simply because its cores have a lowet ipc so itbhas to work harder.

 

Same story with amd also that amd cpubwill have 32 squares in task manager

[papajo]

29 minutes ago, jaslion said:

Say a 9900k will have a higher load than a 12900k simply because its cores have a lowet ipc so itbhas to work harder.

 

Exactly so in the video the AMD CPU should have more usage since it has lower clocks and same or slightly slower IPC. 

 

29 minutes ago, jaslion said:

Same story with amd also that amd cpubwill have 32 squares in task manager

Nope since SMT (amd's HT) is disabled it would show 16 squares. 

[jaslion]

31 minutes ago, papajo said:

Exactly so in the video the AMD CPU should have more usage since it has lower clocks and same or slightly slower IPC. 

 

Nope since SMT (amd's HT) is disabled it would show 16 squares. 

This does bounce back to architectural difference too and how the cpu handles workloads. Hence why you can't really do a apples to apple comparison

[LIGISTX]

4 hours ago, jaslion said:

This does bounce back to architectural difference too and how the cpu handles workloads. Hence why you can't really do a apples to apple comparison

I don’t think OP actually wants to understand… I think they are trying to force some weird point that doesn’t even make much sense. We are somehow comparing totally different CPU architectures, AND comparing 16 cores to 8 cores and wondering why things seem different. I give up on this one ¯\_(ツ)_/¯. You can keep trying tho   lol. 

[LIGISTX]

10 hours ago, papajo said:

AMD can do "more stuff per tick" so it needs less ticks to do the same stuff<-- which is not true so we exclude that

Yes. Again, the AMD CPU can do more stuff per time which is what I said many posts ago. It literally has twice the amount or cores crunching away at the game. It has 16 cores vs intel’s 8. It’s going to do more in less time regardless of what other tricks there are under the hood happening. 
 

If you can’t understand this, and you think a virtual thread can do the same work as a real core and thus “task managers 16 squares of AMD = 16 squares of Intel”, there isn’t much explaining we can do to help you understand and better. The AMD CPU in this situation has a lot more raw power being allowed to tackle the problem, it’s going to do the calculations quicker. 

[papajo]

2 hours ago, LIGISTX said:

It literally has twice the amount or cores crunching away at the game

The sole reason that HT/SMT exists is exactly because a thread doesn use 100% of the core's capacity to crunch, How you think for example an 8 core CPU with HT manages to have better results than a 8 core without HT? because HT basically takes up that free core time that would not have been used in each particular tick otherwise. 



A logical core is not slower than a physical core it is in a sense a physical core (it just uses what's "left" of it )  usually "what's left of it" is less than 50% of the physical core's capacity, that's the sole reason why HT doesnt double the performance in most cases. 

In cases where a thread uses 50% or less of a physical core HT will double the CPU's performance on that particular task(assuming that extra threads can be generated by the program or more simply the program is optimized for multithread usage without a fixed limit of threads or with a fixed limit greater than the sum of the cores -logical- of a CPU ) 

[LIGISTX]

5 minutes ago, papajo said:

The sole reason that HT/SMT exists is exactly because a thread doesn use 100% of the core's capacity to crunch, How you think for example an 8 core CPU with HT manages to have better results than a 8 core without HT? because HT basically takes up that free core time that would not have been used in each particular tick otherwise. 



A logical core is not slower than a physical core it is in a sense a physical core (it just uses what's "left" of it )  usually "what's left of it" is less than 50% of the physical core's capacity, that's the sole reason why HT doesnt double the performance in most cases. 

In cases where a thread uses 50% or less of a physical core HT will double its perfomance (assuming that extra threads can be generated by the program or more suimply the program is optimized for multithread usage without a fixed limit of threads or with a fixed limit greater than the sum of the cores -logical- of a CPU ) 

No… a physical core is a discrete, individual core. A logical core is shared (as you said) and sharing is inefficient. 
 

HT will not double your performance, ever, at anything. Hell, a second physical core will almost never double your performance. In a lot of games for many years you would gain performance by turning HT off in BIOS just to reduce the inefficiency of even assigning work to a virtual thread. And none of this even takes into account the OS scheduler difference between the architectures either. Your turning what is inherently an extremely difficult problem into “just look at a 16 square graph in windows”.
 

But since you are clearly the expert here, I am not sure why you are even asking the question in the first place. Nothing anyone tries to explain you will accept as plausible, you don’t even take what is being suggested into account. So, since you already fully understand how CPU’s work and are not interested in hearing any information counter to what you already believe, I don’t think there is much more to be said here. 
 

Good luck. 

[jaslion]

15 hours ago, LIGISTX said:

Yes. Again, the AMD CPU can do more stuff per time which is what I said many posts ago. It literally has twice the amount or cores crunching away at the game. It has 16 cores vs intel’s 8. It’s going to do more in less time regardless of what other tricks there are under the hood happening. 
 

If you can’t understand this, and you think a virtual thread can do the same work as a real core and thus “task managers 16 squares of AMD = 16 squares of Intel”, there isn’t much explaining we can do to help you understand and better. The AMD CPU in this situation has a lot more raw power being allowed to tackle the problem, it’s going to do the calculations quicker. 

True the 5950x and 12900k single core performance in all core load profile isnt far off so its only logical that the 5950x has to do less work. Sure the 12900k can reach higher fps and such but it having a worse all core performance will also show up in hoe much load it gets in games. Again logical, virtual threads never have the same performance as physical ones.

 

I think op can be helped to understand better possibly by looking at the e cores. Intel included these to increase performance because they KNOW that an e core is flat out better than a ht so instead of 1 core + ht you can get more performance from 2 e cores. Looking at the physical size of those ecores I dont doubt that they were also just a cheaper option. 2 e cores look to take up less space than a core+ht.

[papajo]

16 hours ago, jaslion said:

 

I think op can be helped to understand better possibly by looking at the e cores. Intel included these to increase performance because they KNOW that an e core is flat out better than a ht so instead of 1 core + ht you can get more performance from 2 e cores. Looking at the physical size of those ecores I dont doubt that they were also just a cheaper option. 2 e cores look to take up less space than a core+ht.

HT doesnt take physical space 

HT means ( I will use the spare "space" of the physical core to compute thread B that is not taken up by  thread A)  HT IS the physical core it self working more efficiently. 

e- cores (which were disabled in the video btw) probably are more performant under most scenarios than HT (so e.g if we just used 4 P.cores with HT vs 4 P cores + 4 e-cores with HT disabled) but I doubt they are "flat out better"  because as far as I know they are somewhere like 40% as "good" as the p-core. 

HT has not a specific percentage in terms of performance it is heavily depended on the code/threads it can vary from near to 0% to near to 100% but usually is about 30%. 

e-cores ( I assume) have been developed by intel mostly because a) they take less space than P-cores b) they consume less power so they are ideal for laptops when you just browse the internet etc.