Why we can not have more than 5GHz Processors

[mawyatt]

I see people regularly saying this about current games, yet the main games consoles (PS4 and XONE) moved to having a considerable number ( of low performance cores several years ago. Someone is being unrealistic, but I've never investigated who.

Doesn't the use of many cores in consoles date all the way back to the PS3?

Back in 2003 we were briefed by IBM on the Cell Processor intended for use in the PS, it was a central Power PC core with 8 numeric processors I recall. Ran on 1 volt and dissipated 90 watts. This was a very powerful processor at the time and IBM was paranoid about the USG imposing export control on the Cell. So much so that they changed the design so the Cell wouldn't work outside the PS (encrypted) because a company was buying the PS and removing the Cell and installing in another mother board running Linux for use in CAD applications. Very powerful processor at the time indeed!!

Best,

[coppice]

I see people regularly saying this about current games, yet the main games consoles (PS4 and XONE) moved to having a considerable number ( of low performance cores several years ago. Someone is being unrealistic, but I've never investigated who.

Doesn't the use of many cores in consoles date all the way back to the PS3?

Yes, but in a different way. The Cell chips had several fast specialised number crunching processors, and one more general purpose control processor, and the Cell chip handled the whole game. The 8 cores in a PS4 or XONE are supported by a powerful GPU. The 8 medium speed cores vs 1 or 2 very fast ones for the general purpose processing work that the PS4 and XONE designers made didn't really enter into the PS3 design choices.

[coppice]

Back in 2003 we were briefed by IBM on the Cell Processor intended for use in the PS, it was a central Power PC core with 8 numeric processors I recall. Ran on 1 volt and dissipated 90 watts. This was a very powerful processor at the time and IBM was paranoid about the USG imposing export control on the Cell. So much so that they changed the design so the Cell wouldn't work outside the PS (encrypted) because a company was buying the PS and removing the Cell and installing in another mother board running Linux for use in CAD applications. Very powerful processor at the time indeed!!

What happened to the Cell processor is rather sad. A lot of resources were put into getting it off the ground, and then very little was put into sustaining and developing it. I still have a Leadtek PCI-E board with a Toshiba SpursEngine chip, which is cell based. I bought it with the intention of developing media acceleration for it. However, it quickly became obvious that there wouldn't be good ongoing support from Toshiba or the other companies developing Cell. I didn't get much beyond finding that the chip could perform very well for my purposes before I realised it was a lost cause.

[mawyatt]

Back in 2003 we were briefed by IBM on the Cell Processor intended for use in the PS, it was a central Power PC core with 8 numeric processors I recall. Ran on 1 volt and dissipated 90 watts. This was a very powerful processor at the time and IBM was paranoid about the USG imposing export control on the Cell. So much so that they changed the design so the Cell wouldn't work outside the PS (encrypted) because a company was buying the PS and removing the Cell and installing in another mother board running Linux for use in CAD applications. Very powerful processor at the time indeed!!

What happened to the Cell processor is rather sad. A lot of resources were put into getting it off the ground, and then very little was put into sustaining and developing it. I still have a Leadtek PCI-E board with a Toshiba SpursEngine chip, which is cell based. I bought it with the intention of developing media acceleration for it. However, it quickly became obvious that there wouldn't be good ongoing support from Toshiba or the other companies developing Cell. I didn't get much beyond finding that the chip could perform very well for my purposes before I realised it was a lost cause.

It's sad to see what IBM is today. They were the ones paving the way in semiconductor technology with the SiGe Bipolar, Ge induced strained CMOS, SOI, Cu interconnects, FinFET and so on. They always didn't have a smooth ride tho, with the SiLK gate dielectric failures and thermal limited SOI FinFET, but as a leading technology company weren't afraid to push the envelope in the semiconductor field. They were the most influential semiconductor company over the past 4 decades in pushing the limits of this technology and keeping Moore's Law alive IMO.

Best,

[gnuarm]

Back in 2003 we were briefed by IBM on the Cell Processor intended for use in the PS, it was a central Power PC core with 8 numeric processors I recall. Ran on 1 volt and dissipated 90 watts. This was a very powerful processor at the time and IBM was paranoid about the USG imposing export control on the Cell. So much so that they changed the design so the Cell wouldn't work outside the PS (encrypted) because a company was buying the PS and removing the Cell and installing in another mother board running Linux for use in CAD applications. Very powerful processor at the time indeed!!

What happened to the Cell processor is rather sad. A lot of resources were put into getting it off the ground, and then very little was put into sustaining and developing it. I still have a Leadtek PCI-E board with a Toshiba SpursEngine chip, which is cell based. I bought it with the intention of developing media acceleration for it. However, it quickly became obvious that there wouldn't be good ongoing support from Toshiba or the other companies developing Cell. I didn't get much beyond finding that the chip could perform very well for my purposes before I realised it was a lost cause.

Sound engineering decisions are not always sound business decisions.  That's why most successful engineering companies are not run by engineers. 

[coppice]

Sound engineering decisions are not always sound business decisions.  That's why most successful engineering companies are not run by engineers.

Most business decisions are far from sound. Why put the heavy investment into something new, if you aren't prepared from the outset to see it through to a clear point of success or failure? That's just investment down the drain. I've seen more things fail due to this kind of half assed behaviour, than fail because the technology or the timing didn't work out. I've also seen a lot of business people cling on to something way past the point of clear failure.

[mawyatt]

Think the biggest failure mode in larger enterprises is the arrogance and stubbornness that reduces ability to be adaptable & flexible, while adding the next quarter vision limit mentality that Wall St. imposes on meeting "their" expectations.

Best,

[wraper]

A lot of talk about RISC, but not a single mention that for More than 2 decades Intel x86 processors are actually RISC in disguise.

[gnuarm]

Sound engineering decisions are not always sound business decisions.  That's why most successful engineering companies are not run by engineers.

Most business decisions are far from sound. Why put the heavy investment into something new, if you aren't prepared from the outset to see it through to a clear point of success or failure? That's just investment down the drain. I've seen more things fail due to this kind of half assed behaviour, than fail because the technology or the timing didn't work out. I've also seen a lot of business people cling on to something way past the point of clear failure.

You have that backwards.  Why continue to put money into something that turns out to be a poor business decision?  The "sunk costs" fallacy.  Engineers also make mistakes of being overly optimistic when estimating costs and schedules.  To be good at business you don't have to forget engineering, but you have to put it in it's appropriate place. 

The reality is a good business manager can be advised about the engineering by good engineers.  The reverse is not true.  The person in charge has to understand the business decisions in order to make them while delegating the engineering decisions to the engineers.

Of course many engineers won't agree with this.

[coppice]

Sound engineering decisions are not always sound business decisions.  That's why most successful engineering companies are not run by engineers.

Most business decisions are far from sound. Why put the heavy investment into something new, if you aren't prepared from the outset to see it through to a clear point of success or failure? That's just investment down the drain. I've seen more things fail due to this kind of half assed behaviour, than fail because the technology or the timing didn't work out. I've also seen a lot of business people cling on to something way past the point of clear failure.

You have that backwards.  Why continue to put money into something that turns out to be a poor business decision?  The "sunk costs" fallacy.  Engineers also make mistakes of being overly optimistic when estimating costs and schedules.  To be good at business you don't have to forget engineering, but you have to put it in it's appropriate place. 

The reality is a good business manager can be advised about the engineering by good engineers.  The reverse is not true.  The person in charge has to understand the business decisions in order to make them while delegating the engineering decisions to the engineers.

Of course many engineers won't agree with this.

You seem to have read the opposite of what I wrote.

[janoc]

I see people regularly saying this about current games, yet the main games consoles (PS4 and XONE) moved to having a considerable number ( of low performance cores several years ago. Someone is being unrealistic, but I've never investigated who.

Given that I regularly work with game engines like Unity and Unreal Engine, I can tell you that you are wrong about the threads.

A typical game using one of these will use one main thread running most of the logic and submitting frames to be drawn, one rendering thread where the submitted frames actually get sent to the GPU, maybe one thread for networking and paging in assets from disk. Perhaps one more for sound. And that is all. Many games have only maybe 2-3 threads max. There simply is no need for more, given what a typical game does.

Videogames are not heavily threaded because there is no benefit in it. The bottleneck is almost always the GPU, loading data from disk and such - and throwing multiple threads at this won't help you any because the hardware isn't "re-entrant", so multiple threads trying to do e.g. rendering or reading from disk will only fight over locks but won't make the rendering or disk spinning any faster. Also games rarely need to do any significant calculations on the CPU these days.

8 is not a "considerable number" of cores by any measure and that the consoles have them is only because they are based on commodity hardware (AMD X86-64 Jaguar for PS4, another AMD X86-64 chip for XBone). Not because they need or even use all of them all the time when playing games. Also the consoles typically dedicate one core to the OS functions. 

[wraper]

A typical game using one of these will use one main thread running most of the logic and submitting frames to be drawn, one rendering thread where the submitted frames actually get sent to the GPU, maybe one thread for networking and paging in assets from disk. Perhaps one more for sound. And that is all. Many games have only maybe 2-3 threads max. There simply is no need for more, given what a typical game does.

Were you living under the rock for more than a decade?

8 is not a "considerable number" of cores by any measure and that the consoles have them is only because they are based on commodity hardware (AMD X86-64 Jaguar for PS4, another AMD X86-64 chip for XBone). Not because they need or even use all of them all the time when playing games. Also the consoles typically dedicate one core to the OS functions.

Less than 6 core (modern, not Bulldozer crap) CPU severely limits gaming performance on PC. With 4 cores even if average FPS might be somewhat OK, there will be stuttering in many games. Saying that PS4 does not use all of the cores is complete BS, nobody will put more CPU cores than needed in console, nor architecture requires many cores. PS4 CPU is too weak, if anything.

[hans]

I don't disagree with multicore CPU's but I think the average person does not get how complex it is to the point ebay sellers sell PCs quoting the speed as the CPU speed times the threads (virtual cores) never mind the real number. 4 I think is the sweet spot unless you are actually going to use those cores like I do. more cores will not run a program faster, only make the machine more responsive to a point and you have to run slower anyway. This is why in a side by side comparison with solid edge more cores is less performance if the base speed is higher as it is a single thread program so my PC won over the work one as it was nearly 1 GHz faster.

2 cores was certainly a breeze to single core and if it's just office applications more is a waste.

Oh for sure, but so does Apple not like to quote which specific i5 or i7 your system is getting. Salespitches (e.g. on Ebay) only care about preventing false advertising because that's what drives refunds and problems. Wen you advertise a PC: does the CPU have 4 cores, run 3GHz, and is it an Intel i7? Great, that stuff sells because people know the i7 brand and 3GHz sounds good. We really don't care it's in fact an i7 950 from 10 years ago.

But that's not much different than buying a car from a generic car shop that doesn't know the product he/she sells, because he bought the car as trade stock (for lack of a better translation) from another dealer. Does this car use a belt or chain? How many speeds does the automatic transmission have? What is the fuel consumption like? "Oh I'd have to look that up sir.."

I think especially a question like the latter, which is user experience related, depends on what you need & expect. I know some people think that anything <9L/100km is economical. I start to complain when my car uses more than 6L/100km. Likewise, how much horsepower do you want a car to have? Even a Citroen C1 will get you in the fast lane on the highway, although you may need to push it a bit harder. The one advantage of cars is that we generally do not raise the speeds and performance bar, so there's that.

In contrast, computers will become 'slower' over time. Operating systems, program versions and webbrowers get updates. You install them because of security patches but bringing in new features also (eventually) slows things down. Likewise, a dual core machine will do office tasks just fine, but I wouldn't be certain about buying such a spec new today. Single-core clock speeds are indeed king for general day use, but IME most modern CPU's (especially from AMD) all clock pretty high, with decent IPCs to get that single-thread performance. Only Intel seems to think that selling 3 different i5's plus charging extra for unlocked clocks is an attractive product skew. But I digress, I have noticed that things like webbrowsers, non-hardware accelerated video decoding, etc. all chew up a considerable amount of CPU horse power, to the point where even an i5 3570K can start to feel slow when in use as a HTPC (with probably poor iGPU driver support, admittedly).

[janoc]

Were you living under the rock for more than a decade?

And what about a factual argument instead of gratuitous ad-hominem?

E.g. I have started Deep Rock Galactic right now - pretty recent game, using Unreal Engine. As recent and "AAA" as it gets.

This starts a dozen or so of threads - but almost all of them are blocked waiting for various things (i.e. not really using CPU). I have a 4 core i7-6700 CPU - and it sits at 30% utilization even with the game running.

But please do tell someone who has actually worked on the Unreal Engine rendering code stories about videogames being heavy threading users ...

If you have said databases or some sort of server where you often see a thread per connection, I would believe you. But not videogames.

Less than 6 core (modern, not Bulldozer crap) CPU severely limits gaming performance on PC. With 4 cores even if average FPS might be somewhat OK, there will be stuttering in many games.

And how do you know that this stuttering isn't simply due to bottlenecks like missing vsync cutoffs which causes an instant frame rate jump, insufficient GPU performance (again causes the same symptom), the game loading a ton of stuff from disk (again the same symptom)? Or million of other possible issues.

Did you actually profile a game? I did. It is part of my day job. And I can tell you that it is extremely rare for the CPU to be the bottleneck - in modern games the CPU is there pretty much to only shove buffers into the GPU as fast as it can - and that's all.

Saying that PS4 does not use all of the cores is complete BS, nobody will put more CPU cores than needed in console, nor architecture requires many cores. PS4 CPU is too weak, if anything.

And you have completely missed my point. That you have 8 or how many cores available does not mean that all are actually in use at any given moment by the application (i.e. game and OS). You can have even 32 or 64 cores available - and your game won't run any faster if the software doesn't have anything to actually utilize them for.

[hans]

I agree with @wraper. I ran Cyberpunk 2077 on my 3900X, and saw CPU usage typically at 45 to 50%. That's a full 6 core, 12 threads in use by a single game! Most people would have bought a 3600X and be done with it - those CPUs are brought down to their knees.

I have exactly the same experience with racing simulators, but also building games like Satisfactory. I have built a reasonable size factory. Upgraded GPU from GTX970 to RTX3080 (over 5x performance): still can only get 40fps.

Even MSFS , which is notorious for leaning very much on the mainthread, uses more >4 threads on my machine. Admittedly, I like to see it use as many threads as it can because the game still runs like absolute garbage in busy areas (as in 15fps garbage, with huge lag spikes in VR), but nonetheless: 4 threads for a modern game is not a lot anymore. In addition, high FPS gaming has become a thing in recent years. I run 1440p 144Hz displays, and if the game motion benefits, I do want to hit at least 2/3s of that framerate (even with adaptive sync tech).

There are plenty of games that are CPU bound by e.g. a Core i5 6600k from only a few years back. Just take a look at charts like these:
https://youtu.be/LCV9yyD8X6M?t=751
https://youtu.be/LCV9yyD8X6M?t=948

The reason why we get a PC for gaming is because we want to run these games at crazy resolutions, aspect ratio's, VR, framerates or interesting game concepts or niche simulators.  I for sure would be pretty annoyed if an i5 "gaming CPU" from only a few years ago only pushes half or 2/3s of what a new 6 core Intel CPU can do. Yeah okay, they use a 2080Ti to push 1080p medium/high details to highlight bottlenecks. But with the recent new GPU generations, even people buying a brand new 3070 (perhaps targetting 1080p/1440p 144Hz+) will have a similar level of GPU performance; so it's really necessary for game engines to step it up to as many threads as they can benefit from.

[wraper]

And what about a factual argument instead of gratuitous ad-hominem?

It's not ad hominem, and not attack on person.

E.g. I have started Deep Rock Galactic right now - pretty recent game, using Unreal Engine. As recent and "AAA" as it gets.

LOL "AAA"  , it's indie game with extremely simple graphics and physics, and thus low system requirements. It could be AAA 15 years ago.

And how do you know that this stuttering isn't simply due to bottlenecks like missing vsync cutoffs which causes an instant frame rate jump, insufficient GPU performance (again causes the same symptom), the game loading a ton of stuff from disk (again the same symptom)? Or million of other possible issues.

Because on the same hardware but with 6 core CPU there is not stuttering, it's especially bad on 4 core without HT.

Game from 5 years ago on mediocre GTX 1080 GPU by todays standards, disabling Hyperthreading on 4 core CPU significantly worsens performance: