How do I get ready for a 256 core CPU?

Let's suppose in a years time, there is some kind of technological breakthrough, and it is now possible to sell a 256 core CPU for the price of today's 4-core CPU.

But... wait... there's no games that can take advantage of this new breakthrough.

Except... someone has thought ahead and designed a game that needs 256 CPU cores to run... and he (or she..) becomes a millionaire with their astounding game.

What would this game look like do you imagine?

 

Utilizing all the CPU cores is nice but it's not a measurement of success in the game industry. Frankly, no one really cares. It's a possibility to use all the cores you want but you won't be a millionaire, just because you made a game which need a very niche CPU. On the contrary.

Make a game which is fun to play, don't screw over the players with bad monetization and bad DRM and be able to run on many devices as possible. And then you have a chance to have something gold in your hands. CPU utilization is a good to have but not needed.

 

This. Out of every indie game that became wildly successful I cannot think of a single one that fully utilizes the hardware we have available today. Keep in mind four cores may be common on the desktop, but consoles have had eight cores for several years now and very few games truly try to take advantage of them.

 

Anyone else want to reply to this light-hearted imagination based thread?

 

Think of us when you're swimming in your gold, Scrooge McDuck.

 

Even if there was a massive breakthrough allowing 256 core CPU for the price of a 4 core CPU, it would take a while to penetrate the market. It would likely take a year or more to gear up production, and then several more years to get widespread adaption throughout the marketplace. It would most likely require a new motherboard, not just a new CPU. And it would likely also require some operating system level updates and some updates to the drivers for popular graphics cards. Anyway, if there were a massive breakthrough in a lab somewhere today, it might be 3-5 years before that breakthrough leads to a marketplace with millions of gamers using that technology.

It probably would not make sense to spend a lot of time figuring out how to fully utilize 256 cores when most gamers would not have that tech for at least several years after an announcement about the hypothetical breakthrough tech.

Remember, gamers don't really care how games are made or how much tech they support. Gamers care about the experience the game gives them. Great games deliver awesome experiences by faking a lot of things. The correct question to ask is what is the most amazing experience you can give gamers regardless of the current and future tech.

One of the best examples is to look at the experience that Doom (1993) delivered relative to anything else at the time, and then look at the hardware requirements for Doom (1993). Using computer hardware that most people thought was not good enough to even run a game, id Software delivered a reasonably immersive 3D FPS in 1993. Doom (1993) revolutionized the PC gaming industry. If id Software had chosen to focus on building a super realistic game that ran well on 256 cores, they would have completely missed the market.

 

I think that you guys are interpreting his post all wrong. I don't think he is suggesting that someone is going to get rich with a 256 core game. I think that he is just asking if there is something that you could do with 256 cores that you could not do now, that would allow you to make a really cool game.

I think that VR is a good example. If you think that just adding VR to your game is going to make it great, you are totally missing the point. It's not the tech, it is what the tech would allow you to do. I am doing some really cool things with VR which I just would not be able to do without it.

Back to the original question. The only thing that I can think of is that for things like racing sims it may allow you to run AI and car physics for each car on different cores which may make it more realistic. Just a thought.

 

GPU has already many many cores.
If each of CPU cores would be 3+ GHz, that would be a deal I suppose, rather than 200+ x100Mhz, or even x1GHz
However, Many cores, doesn't mean necessary great advantage, in software, or gaming.
We know many games are single cores. So actually stronger fewer cores, that would be something. Like 10x stronger.
But devs always will find, how to make CPU a bottleneck anyway

 

Well, he was talking about being a millionaire just because "someone" (asking for a friend... LOL) utilized all the 256 cores on a hypothetical CPU.

BTW the short answer is this: jobify everything you can and leave the decision to Unity to utilize as many cores as it can. This is why they're building this entire infrastructure (mainly Job system this time). And of course, make a fun game, whatever it means.

 

Something with detailed simulation for a large number of separate entities going on, like Dwarf Fortress, might benefit the most.

 

If Dwarf Fortress used more than one core we might be able to see the entire world simulating rather than just a very tiny subset of the world with everything else getting updated semi-yearly.

 

@Arowx alt account?

 

Check out Amdhal's Law https://en.wikipedia.org/wiki/Amdahl's_law


You get diminishing returns with parallelism so to get a real benefit from 256 cores you would need a game with logic that can be 95% parallel and then you would get about an 19x speed up compared to single core.

The question is what types of game mechanics can be so highly parallel. You would need a game that has very atomic game mechanics* where everything in the game has very limited range impacts on other elements.

* Atomic game mechanics that would have to fit within the L1 cache of the CPU cores to ensure you maintain the speed benefits of DOTS.

 

Yes and no. Current processors only truly run a couple cores at the highest speed allowed by their model with the rest of them running very close to the base speed. With a high core count you could offload more of the non-gameplay related code allowing the highest performing cores to only deal with gameplay.

Keep in mind too our current ways of determining what will run in parallel and what won't is currently limited to design time due to limited performance available. With large core counts you could theoretically scan objects in the scene every few frames checking to see if anything gameplay-related will affect them and if not run them in parallel until the next check.

My biggest problem with charts and theories of this nature is that they're working off of our current knowledge with the assumption that we can't find a good solution for it. An assumption that I find ridiculous when you consider that even Unity, a game engine that has had an only-single-threaded API for years, is gaining ways to assist with parallel programming.

 

I was unaware that Unity had this level of thread core speed affinity for it's jobs system, can you assign jobs or tasks to the highest speed cores?

 

We don't have it now but there's no guarantee we won't have it in a few years. Modern operating systems already have a way of detecting and acting on which cores have the highest clock speeds. It wouldn't be that crazy to imagine programmers asking for the ability to do it themselves if it doesn't already exist in OS APIs.

 

buy nitrogen, tons of it.

 

It is possible there are some types of games that could benefit. But it really requires a ton of processing be performed on the cpu side that is quite complex ie not well suited to run on the GPU side. If there were something like that and everyone else was running it on 4 cores or even 1, and you suddenly are taking advantage of many more, you could see huge increases in framerate for example which could be an advantage. But it's gotta be obvious that all the time spent on the cpu pays off in something that likely LOOKS visually extraordinary.

 

With all this in mind, don't forget ECS won't help in this case either.
You're probably going to be GPU bound in the end anyways with 6-12 threads.

256 seems to be over-overkill

 

Wouldn't that CPU have to follow the GPU and have high bandwidth memory next to it (on a board maybe) just to maintain the processing throughput or would there be as much L3/L4 cache as processing cores e.g. AMD Zen 2 Chiplets IO Chip?

 

I have nagged about gfx job for ages. Someday

 

Not if you OC, my 2700x runs at 4.2 ghz all 8 cores. And Ryzen 3xxx will go beyond that

 

Very few people overclock, and the consoles are not overclockable at all.

 

Most gamers OC, not PC owners in general. Consoles are for peasants
But a few hundred more ghz per core is pretty much irrelevant, what we need is good support for rendering in multi core

 

Complete rubbish.

 

But like I said it doesn't matter. Core count matter, and that engines need to support it better. 8 threads are not the norm anymore 12 and 16 are. And it will keep on going up.

 

Please provide solid prove of your claims, because we know, how you tend to generalize your inadequate and often misleading information.

 

Show some actual quantitative evidence please and link the source, otherwise sorry but this sounds like rubbish.

 

I cross ran the CPU brand names to the clock frequences, And it does not add up. People run them at faster than stock clock. Not perfect, but I dont think there is actual statists on it. Also if yuo hang around PC gamer forums you know evertbody OC

edit: Steam hardware survey

 

I don't know what all the fuss is about. https://store.steampowered.com/hwsurvey/cpus/

By the time it becomes relevant we are at Unity 2024.1 (or later) with Full ECS+Jobs etc and it will take care of it all.
Not to mention that if in the future people stream their games from remove servers their will be many other pressing cons urns for developers.

 

Mostly we know that. But that does not fit to @AndersMalmgren claim. So I am waiting for some statistical prove otherwise.

 

I personally enjoy building high end PCs and then overclocking them. But I also admit that I am more of an exception than the rule. Most gamers in general do not build ultra high spec PCs and overclock. Statistically speaking, most gamers have relatively inexpensive computers.

According to Steam's stats, most users (55%) have a quad core CPU, and the next biggest group (25%) have a dual core CPU. Then six core CPUs for the third group (14%). The 8 core CPU group is still under 3%.

 

I hang out around taxi drivers. So every car owner carry people around in their vehicle.

 

Another data point worth mentioning is that the percentage of steam hardware users above 3.7Ghz is less than 6%, with more than 50% less than 3.29Ghz. I doubt the majority of them are overclocking.

 

And to question, how many gamers are actually tech savvy?

 

In my experience dealing with user support and bug reports in games: five.

Not five percent.

Just five.

To be serious for a moment though, the claim that most gamers overclock is on-its-face ridiculous. Most gamers don't even change the settings on their games from default. I mean any settings. I've seen stats from games that don't autodetect resolution where 84% of users with 1080p displays were still running at the default 720p and that was the most changed setting. The rest of them hovered in the high 90s.

 

I wonder who are the other four

 

Ever considered the possibility that only a tiny fraction of gamers even uses those forums?

Don't many multicore cpu's dynamically boost singlecore clock speed if only one core needs a lot of computing resources?

 

Lower and boost. My system is entirely stock, but my CPU can bounce between a lot of values depending on load.

 

Depends on the make and model. My Ryzen for example in stock clocks 1 core

 

If someone designed a game that "needs" that many cores to run, I would probably laugh at them. I'd hope they are being paid well by the company that made that processor. Because gamers/customers won't be giving them money.

Now, if we mean "can use" that many cores... I guess I would tell the game itself, "good for you, here's ~8 cores."

 

No. This isn't even remotely accurate. Most gamers on the PC do not overclock their hardware.

 

I can attest to this likely being true. I am a programmer and a pretty big gamer. I buy some pretty hefty hardware. Have spent 2.75k, then 3.5k (and next month I am spending 4.2k) on gaming PC's over the year. One of the machines had factory overclocking. I have never once tried to overclock something myself. I've never needed to to get max settings, and I just haven't cared enough in situations where it might make sense.

That might make me "not a hardcore gamer", but oh well. I suspect I am a fairly average gamer in how I consume my games.

I choose generalized settings in most games. I choose the "Ultra High Settings", and let games auto adjust everything. I only manually adjust in very specific scenarios. I can image when I get the GTX 2080 ti next month that I will turn ray tracing off in some scenarios, for performance reasons. But generally, I don't care about the individual options.

Just me though, but I get a feeling from most people that game as much as me that they are in the same boat.

 

Don't know what is max limit of thread run by ECS / DOTs, but by design games utilizing this paradigm, can run already many-many threads. Hence, weather is 2 or 100, is less relevant. AI learning at runtime would benefit massively for example. Path planning and collisions procssing, that another. And that only one of many possible potential usages.
Wonder, who is going laugh now

 

I didn't say there isn't a use. I am not laughing at the concept of that many cores. I am laughing at this hypothetical scenario where someone makes a game that REQUIRES 256 cores. They would make no money.

Apple could make an iPad with 256 cores, and sell it for 20,000 dollars. You are right. Someone somewhere could buy it and laugh hysterically that their machine learning is fast. Meanwhile, Apple sold 2 units and didn't make up the cost of actually developing it.

I think the majority of laughter would be from the people on the outside looking in.

 

There are games, which are scaleable. So any larger number of cores would already potentially benefit.
For example factorio, which uses multi threading. But is pretty dam well optimized by far.
Maybe there isn't many such games, but if number of cores would show significant of increase over time, then people surely would take advantage of that. Many games been designed, to look ahead in time. When released, full spec was only run on few machines. Few years later, full spec run on many machines, while such games still look fresh.

But then, there are as well range of software, besides just gaming, which would take much of advantage of that.
Raytracing could be as well then run on CPU, assuming hypothetically GPU stagnated.

Designing approach of mechanics also changes. More CPU power, less memory is required. Or want save on CPU, use more memory instead (Modern browsers as example).

Servers are another thing. Nice when complex physics and simulations is handled on servers, for many players, at the same time. Large online games, could be looking more at multi core solutions. Not that it is something new however. Larger Minecraft servers with multiple mods I.e. (anyone? )

So yeah, scaleability design is a thing.

 

I certainly don't...even if I was interested, it would have little or no effect on 99% of games, so what's the point? You make games go faster by getting a new GPU. (Which actually makes some sense to overclock, though I haven't done that either.)

--Eric

 

Unless you're being limited by your CPU, but then if your CPU is holding back the entire system the only solution is to buy a new CPU. Overclocking won't let you go from poor performance to good performance. It's only good for a very minor boost and even then only if you're running low resolutions as otherwise the GPU becomes the bottleneck.

 

Also, getting CPU limited is a more and more rare affair in games as they start to really get into exploiting the fantastic parallelism GPUs provide. You have to be running pretty low power or old to get severely CPU limited these days.

 

Have any of you guys played anything Paradox recently? CPU limited games are still very much a thing.

It’s very easy to go beyond current CPU capacities with an even mildly complex simulation.