> the moment games or desktop applications use more than 8 cores

>>59537900
>I can't handle the fact that practice trumps theory the post

>muh games

>>59537974
> source: arse and manboobs
inb4: gentoo compilation

>>59537982
>source
Any competent game that runs on more than 1 core.

>>59537980
> I can't read
interactivity

>>59537988
Yes you stupid fuck. Those idiots aren't talking about a couple of cores - we know the sweet spot is around 4core/8thread right now, and many believe it's still at 4core/4thread. There are pathetic idiots that believe 8core/16thread or 16core/32thread can be easily become a sweet spot for interactive applications, but they are delusional manchildren.

>>59538010
You seem autistic. A game will use more than 8 cores eventually.

I would love to have a 32 core machine, it would speed up compile times a lot. You may be right in that many cores aren't required for typical desktop applications, but when it comes to workstations the story is different. Compiling, rendering, stuff like that are workloads that benefit greatly from many cores. I don't know why you are focusing on games, that's a not that important niche.

File: Ep2_Announcer.jpg (19KB, 444x329px) Image search: [Google]

19KB, 444x329px

And welcome back to "Ridiculous Threads on /g/" , ladies and gentlemen.
Today our topic is "What Is The Lock Free Programing?".
Meet our first guest OP.
Would you like to say something, mister OP?

OP: "Ehhh.... I'm stupid faggot, please rape my face!"

So the sweet spot is 128 core

Is this the 7700k buyers remorse thread?

lets talk about how heavily multithreaded tasks doesnt scale well with n cores because as we know something like running multiple tasks at once doenst exist in computing lol.

did you know that you can run stuff in background while playing games on your pc?

>>59537900
>x can't be done
>y is impossible
You will never get anywhere in life with that attitude.

>>59537900
>being this retarded

File: 1489122339460.png (495KB, 1070x601px) Image search: [Google]

495KB, 1070x601px

>>59538010

>>59538026
You are a stupid fuck that doesn't know how interactive applications work.

File: file.png (230KB, 701x479px) Image search: [Google]

230KB, 701x479px

>>59538081
You are a stupid idiot that doesn't get a "Workstation" does not mean "I render 24/7".

>>59538139
>I'm a mentally challenged manchild that can't read
I said "UNIVERSALLY" on the OP, stupid child.

>>59538483
No, I have a 4790K and it's still stupid to be replaced for 99% of people.

>>59538581
You are a mentally deficient manchild. Most of those "background" processes are at approximately 0% CPU usage. They don't even use the RAM exclusively if you were not technologically illiterate.

>>59538656
You are a stupid manchild. Learn to read the OP. It says "UNIVERSALLY".

>>59538748
This. Workstations are not render farms.

File: cure for brainlets.png (75KB, 976x685px) Image search: [Google]

75KB, 976x685px

>>59537900

>>59538822
Do you even know what that graph means manchild?

>>59538859
I bet he thought it's a refutal. It's almost identical to Amdahl's law for most intends.

Why is everyone in this thread so angry?

>>59538859
Its a fix for brainlets who think Amdal's law is the end all be all to multi-core optimizations.

Mainly you.

>>59538871
Because single core > multicore.

AMD Ryzen is FAIL. INTEL IS RAIN DOMINANT.

>>59538872
Please kys. Gustafson's law does not refute the OP. It's almost identical to Amdahl's law for most intends and purposes.

Learn to read your own copy pastas manchild.

>>59538882
Are you okay?

>>59538886
The problem is you don't understand the problem of Amdal''s law and OP's assumption. Gustafsons law fixes that.

Video games won't always be running be running on 640 kB worth of data. And its never enough.

For future programs/games, data size will gradually become larger. Thus larger core count will almost always benefit over single core.

Amdahl's law sounds nice in a time vacuum, but it doesn't work in reality.

>>59538934
I repeat, kys, and learn to read your own copy pastas. Gustafson's law is based on the wishful thinking that enormous problems will not be able to run at all on a smaller machine. Since we are on a thread about gaymes and workstation applications, you are ridiculing yourself, since by definition we are talking about machines that can run the application anyway.

It's possible to solve by breaking everything up into asynchronus components. It's not something that has been common practice in video games though. If 100 core cpus were common they would quickly find ways to use them.

>>59538966
You can't run modern 50 GB games on 1955 super computers.

Rethink your argument.

>>59537900
>you stupid fucks.
>you stupid idiots.
>you fucking assholes.

Who are you addressing?
Also, why have you provided a picture for ants?

>>59538980
No. That's stupid wishful thinking. By definition, an interactive application - such as a gayme or any desktop/workstation app during operation - must adhere to the rules of a global loop.

That MUST ALWAYS have slowdowns due to the necessity to protect variables consistency by locking them up.

Sure, you can IMPROVE it, but the notion "all" gaymes can use it is utter stupidity due to different needs.

>>59537900
But vidya do use multiple cores.
Graphic cards have loads of cores being used.
Just because it says processors doesn't mean CPU

>>59538992
You are still off topic. Nobody is talking about render farms and global simulations of the human brain, and climate change simulations here.

>>59539002
you fucking asshole

>>59538771
so if i run several vms at once only one of them will consume performance?

please teach me.

>>59539019
The scaling of cores means future games/programs will be much more complex.

What we see as super demanding tasks like human brain simulation now will be trivial for future computing.

Thus the problem of Amdhal's law is not relevant and has not been relevant since Gustafson's law fixes the issue.

>>59539010
Of course, the point is it has severe limits due to interactivity. GPUs are by definition parallel machines, you can just stick more cores into them and automatically get more performance (assuming you are graphics/etc. capped).

PS. It's why NVIDIA just sells "SLI chips"(so to speak). They just make a gigantic chip for the 1080 Ti, it costs way more but it's automagically faster because moar cores.

Sooo... Should i just buy a 7700K Kaby Lake or wait till September for a Skylake-X and spend a gazillion more $$$?

>>59539042
You are literally stupid. I wonder if you are trolling or just have a severe care of buyer's remorse.

EVERY FUCKING DAY we see that the OP pic is fucking relevant even by the best coders.

>>59539028
You are off topic. The thread is about interactive applications such as gaymes and workstation software during operation. You are talking about servers and server farms.

Nobody is that stupid here to claim servers and server farms can't easily run multiple cores.

>>59539059
People are taught Newtonian model of physics at grade school. Its not wrong per se, but its not really relevant in modern physics.

Amdal's law is fine if you're new to computer science but its not the be-all-end-all of computer science. Its definitely not the limits of core scaling.

>>59539052
The X line is generally stupid. It's a luxury product. You get 10% performance for 90% more.

The sweet spot is still a good 4core/8thread for desktop users.

Some claim 4c/4t is still better but I'd say maybe barely.

>>59538934
>data size will gradually become larger
feels like only a year or so ago that every fucking new release became a 40+gb download

>>59539088
Obviously, that "law" is tongue is cheek, nobody should take it that seriously (just as moore's law is stupid at least since the first dual core was released).

But it shows very clearly that one can not just assume everything is 100% multithreaded.

There are specific reasons that stop it on most interactive applications, if not all.

File: 170310TerryADavis.jpg (234KB, 1440x1440px) Image search: [Google]

234KB, 1440x1440px

>>59537900
>hurr durr don't buy dual core stick with pentium
>hurr durr don't buy quad core stick with dual core
I mean just wait around 4 years and almost anything will work optymized on 8 cores.
it's not hard look at the templeos
And LLVM or other abstraction layers will help

>>59538744
I understand a few things,
50% of the gaming market uses 4 thread cpus, and the reason we have games that gear themselves to 4 threads is because of this, why make something that will use more cpu cores if people dont have more powerful?

The reason we have games that scale to 8 threads is because of the consoles, people think those are 8 cores, but they aren't, the function more like threads then cores. in fact only using one core per module results in far better performance, so we won't see games push 4/8 by default till the next gen, but if amd gets some marketshare, it will incentivize developers to do a bit more than bare minimum.

We will likely see it in unreal/cryengine/frostbite first where every game can use 6-8 cores because the engines just do it, and you will have unicorns like mafia 3 that do it on their own engine for some reason.

but we will see the biggest benefit from dx12/vulcan where they FAR easier support 4+ cores whereas with dx11 you need to specifically code for it.

we will see with bethesda in the coming months if 6+ cores for gaming is worth a damn or not, personally I would go 8 core just for load balancing purposes, something proven time and time again is a benefit of 6 and 8 core machines.

>>59539112
Assets are irrelevant to CPU needs. e.g. Doom 2016 has like 60GB of assets. Most of that is multiple levels and 4K textures that are GPU/VRAM assets and 100% irrelevant to what a CPU does.

>>59539069
why would multiple vms not be a workstation task?

pretty common for shit like web/backend development (for complex shit where you're running a local instance of each moving part) or desktop software testing (eg: run several windows versions at once)

that shit would be painful with less than 8 cores but would be comfy on more

>>59539069
It was just an example.

Not every background process takes like 0% of your performance. The more cores you have, the more stuff your pc is capable of doing at once.
Of course, >8 Threads won't give you more performance in a game, but additional tasks like recording or so won't impact your gaming performance.

>>59539122
> that fucking meme again.
No. Ask any programmer that has actually delved into multithreaded programming. Most gaymes or even desktop/workstation applications DO NOT NEED multithreading simply because they could become slower with it, they do most of that on the GPU anyway, or they don't have a need for the specific algorithm that can actually use it.

About 99.9% of the gaymes that you see that can use 8 cores or more have a very specific need that can be easily identified. e.g. they have a lot of bots that use complex AI. Or they run other game logic that needs very complex branching.

There are a lot of more games and desktop applications that simply can not do that because they simply do not need that. They do not run bots or other complex methods of that sort.

>>59539154
>Of course, >8 Threads won't give you more performance in a game
Depends on how the games are designed.

If a game is highly multi-threaded, it will run better (within the GPU limits).

>>59539150
>why would multiple vms not be a workstation task?
It's a server task you stupid fuck. If you run vms for a workstation in most cases the vm clients are near 0 to 1% cpu usage in the background.

Besides, it's still off topic, the OP specifically talks about being stupid to "UNIVERSALLY" expect 100% multithreading on CPUs.

>>59539100
4c/4t is pointless unless you can't afford the 8t version, i5s are only preached here because there's so many poorfags

>>59539177
i guess, oversimplifying for the arguments sake is a bad idea on g.

>>59537900
heck I'm barely doing anything and my PC is at over 1800 threads and 113 processes right now. PC's do a lot of stuff (updates, time, virus scan, network, go check out your services and see)

>>59539223
>PC is at over 1800 threads
that are generally idle you stupid fuck

>>59539223
x fucking d

>>59537900
https://www.youtube.com/watch?v=QJOIvACRY6g

Haskell users don't have this problem.

>global std::vector<Event> queue
>first thread to read from the queue executes the event
>everyone can push new events to the queue
why not?

Butthurted tech-illiterated has read wikipedia on Amdahl's law, episode #82, you stupid fucking manchilds.

>>59539047
Sure, but it does show that OP is an idiot.
The graph never talks about CPU's, it just says processors.
It also doesn't say if it's processors for desktop, server or super computing purposes

>>59539162
>>59539173
there were some haswell benches on 4 cores 6 cores and 8 cores, even when games can't fully utilize the cpu, they still benefit due to

I mean for fucks sake I know there are many games that come out every year that have use for more than 4 cores, I have looked at the benchmarks, they aren't the most common, but then again hex core and octa core systems are rare as fuck to begin with even when amd put out an actual hex core back in the day. But every time I try to look up a concise list of games, every single time, people like you argue that we only need 1 core for games.

The fuck is it with people who play games on the computer. I get not using windows 10, or upgrading immediately to a standard whose use is in question, but why the fuck do they deny facts? you can see single to
dual core, dual core came out on top
dual core to quad core, even with the autistic screeching of pentiums are the best, and i3's are all you need, most games run better on i5 or i7

we have 2 prior examples of just straight core count upgrades being great, and we have cases where threads =/= cores

what basis besides your ass is there for 8 cores won't be the same? even, and lets use are imagination, if they aren't the same, the load balancing we can see in the people who did the haswell benches in real world, not bench test scenarios is undeniable.

>>59538081
gaming is a fairly large profit market, possibly niche, but its what drives people to buy new hardware outside of hardware failure.

>>59539223

Aaaand most of them are idle.

>>59539291

Because picrelated is not about perfectly independent tasks. As the computational problem scales parallely you will suddenly find out that time needed for merging results is now scaling down well.

See video gayms for example: if you scale computation blocks but do not scale antialiasing block you will soon reach the limit of antialiasing performance.

Every interactive problem has a single threaded part. If it does not, it is a useless or data collecting computation

File: screenshot_3070.png (32KB, 681x795px) Image search: [Google]

32KB, 681x795px

>>59538771
say hello to my idle

>>59539445
Now show your processes by ordered by CPU usage

>>59537900
You can never get perfect parallelization, no. But you can decouple AI, some parts of physics, sound processing, rendering, disk access, network access, even animation from your main loop. It doesn't have to be perfect to be better.

>>59539445
wtf anon are you stupid this is fake, background stuff doenst take performance!!

>>59539455
Because it surely wouldn't truncate the values in task manager...

Bought into intlel witk sky/kabylake and feeling remorseful m8?

>>59539481
b-b-but... muh manchild...muh 'UNIVERSALLY'...

>>59538483
My i5 4xxx (can't remember the specific) is still going strong. Won't need to replace it for a while.

Just because Ryzen is great, doesn't mean I'll throw out a perfectly acceptable CPU and buy a new one. What sort of economically illiterate retard would do that sort of thing?

>>59539491
That's quite a tangent you're going on about.

Now how about you show those screenshots of those supposed "background" programs.

>>59539506
even if you had a second gen i5, shit wouldnt pay off i guess. intel has plenty of performance

>>59539445
>>59539513

Different anons mate.

I just realize that if the task manager has only 2 digits available to display cpu usage, it probably casts the values to int and displays them

Do I need to link you a fancy graph so you can understand how retarded your request is?

>thread title: the moment games or desktop applications use more than 8 cores
>talks only about games

at least be honest about what you're going to be talking about

back to /v/

>>59539569
But anon my shitty optimised AAA games tested in a vacuum with no background process are the ideal benchmark for >4 core CPU

AMD BTFO, INTEL IZ KAAAANGZ.
pls no look closer and see how the 7700k is getting higher frames than the 6900k in some titles, that's not relevant, I need the rupees to feed family, else we're off again to scavenge on the shitting streets.

>>59537900
>for instance the common "GTA/GTA clone" that has a cazillion of AI bots running about, which is the main reason you see most of those games using a ton of cores, you stupid idiots.

I've been pointing this out for years already.

Developers SHOULD be focusing more on increasing AI count and complexity to create more realistic and dynamic games. There is NOTHING wrong with this. We already have enough graphics and processing power to produce a nearly-lifelike game set in Tokyo City. What we don't have is the power to fill Tokyo city with 13 million NPCs which all have independent and relatively complex AI.

This is a development approach that can actually produce something new and increase gameplay value, complexity, player choice, player impact and agency. A game developed with those goals in the era of 16-thread processing should already be incredible and would present a new standard for interactive worlds and experiences. This is one of the reasons I was very excited when Oblivion was first announced with Radiant AI back in 2005 - too bad the AI was not ambitious enough and the NPC count fairly low - I'd like to see a competent developer take another crack at this approach today.

What we get instead is the same game with new graphics which hasn't changed despite the last 20 years of massively increasing processing power. If you ignore graphics, there is fundamentally little difference between Quake 1 and the latest 2017 shooter.

>>59537900
Kill yourself, shilltel retard. Even Intel is moving to MOAR CORES with Coffeelake.

>>59539611
>But anon my shitty optimised AAA games tested in a vacuum with no background process are the ideal benchmark for >4 core CPU

holy shit couldn't agree more on that. i always asked myself why no one makes like everyday life workload benchmarks with lots of stuff running in the background. :-(

Those would matter so much more.

>>59539232
Could you be less hurtful

Is there a way to see which are idle and which aren't?

>>59537900
|
|>
|
|3
|

you just have outdated notions of parallel programming

the main barrier to parallelization is OOP

https://www.infoq.com/presentations/Are-We-There-Yet-Rich-Hickey

>>59537900
Fucking retard, there's tons of work in games that can be offloaded to separate cores.

That's sweet dear but Ryzen has good single thread performance too.
So you're basically getting a fuckload more.

>>59539642
You forget that making a game is a business. You can easily double or even quadruple the cost of a project by adding details nobody will notice. So a good dev/project manager will just cut the programming of the 13 million AIs (and drastically reduce the costs) while only slightly reducing quality, but increasing performance, stability, debugability, predictability etc.

>>59539003
>By definition, an interactive application - such as a gayme or any desktop/workstation app during operation - must adhere to the rules of a global loop.
What kind of fucking definition is that?

>>59539501
>>59539481
>>59539445
Are you literally mentally challenged? The OP specifically talks about against generalization. If you respond by mocking "but I talk about generalization" just makes you literally retarded because that is precisely the topic you retarded piece of shit.

>>59539569
>>59539611
The OP talks against generalization you retarded piece of shit. The entire point of the argument is that interactive applications can't just use 100% multithreading just like that. A lot of them have no use of it, in some cases they may become even slower or it's generally best to be done in GPU.

So, stop being a mentally deficient manchild and learn to speak on topic.

>>59541330
every program is parallelizable

>>59539642
Don't be stupid. Some games have NO use of AI, or they certainly don't have use of multiple bots. The entire point of the OP is that it's an absolutely retarded idea to generalize and pretend with the delusion that ALL interactive applications -including gaymes and most workstations software- can just use 100% multithreading just like thatl.

>>59539661
> I have no argument, let's just shitpost

>>59539445
Anon... This is two processes hopping between different cores. The 51% usage should have been a clue

>>59539776
You are literally mentally challenged. In 95% of the cases, the background processes idle at ~0% CPU usage. That includes not just gaymes, but the overwhelming majority of Workstation users.

>>59540746
>>you are outdated
>posts some shit from 2009
>about fucking OOP

>>59541364
OP's graph shows that even at 75%, 32 cores are worth it, and at 50%, you get a nicer increase for 8

>>59540776
not always you retarded piece of shit.

>>59541425
just shows that the solution has been known for a time, and proving that OOP has been holding us back since it's still dominant

>>59541229
The definition of anyone that has actually any REAL experience with interactive applications.

You can't run a game like it's render farm.

Sorry, only stupid children think that.

>>59541346
To some degree you stupid asshole. Especially in gaymes - and some workstation desktop software - you can even use the GPU for most of the parallelism. Hell, in several cases - and especially when you don't need special branching stuff like complex AI(bots) - you can even make it SLOWER by trying to make it more multithreaded on the CPU level since the GPU is MUCH better in certain cases for THAT job.

Most manchildren haven't really understand how CPUs work, they are by definition serial machines, GPUs are the ones that can do parallelism by definition (albeit for the simpler stuff, though lately not too simple).

File: kant.jpg (118KB, 294x371px) Image search: [Google]

118KB, 294x371px

I use virtual machines a lot in my daily life because I'm a sysadmin and 8 cores is barely enough.

Besides only braindead /v/ermins shitting up a tech board care about fucking video games.

File: file.png (113KB, 500x375px) Image search: [Google]

113KB, 500x375px

>>59541427
You are literally mentally challenged. Do you even understand how hard it is to get 50% on a gayme or even a regular Workstation application during normal desktop interactivity. If you get 20% you are lucky unless you do special shit in a rarity of games like a ton of bots on GTA and GTA-clone games.

That graph is not even realistic. It should start at 10% which is what most gaymes can achieve.

>>59541450
You literally don't know what "by definition" means, and now you're trying to cover it up with some no true Scotsman fallacy.

Remember that /g/ is 18+

>>59541501
You are a minority. The OP talks specifically against generalization. You are off topic.

Besides, even you have most of them on ~1% CPU usage on the background.

Especially if you're not a retarded piece of shit and you trim the vm's services.

>>59541577
OP should take his video gaymen shit back to the board it actually belongs to.

>>59541545
Fuck of child. What you learned in school is nothing. Go do some real multithreading and come back to tell us how interactive applications of the Desktop (or gaymes) don't create a HELL of reduction in multithreaded performance due to variables protection tied to a global loop.

>>59541528
>get 50% on a gayme
3D Games are the most easily parallelized applications, using thousands of cores. Adding more still increases performance, especially at higher resolutions.

Obviously we use GPU cores, but if you had to use CPU cores the result would be the same.

File: _______.jpg (48KB, 919x721px) Image search: [Google]

48KB, 919x721px

>>59541617

HSA when?

>>59541603
Applications that are 99% idle like Excel? Yes, they're really hard to multithread but it doesn't matter.

Applications like Photoshop that spend seconds applying effects? C'mon m8

>>59541489
the only difference between a stream processor and a core is the instruction set, as far as the program is concerned

games are eminently parallelizable, the main obstruction is the render API being single threaded. DX11 already has some multithreaded functions and DX12 opens the floodgates for multithreading renderers

it's a paradigm problem, not a hardware or programming problem

>>59541656
wew. you can almost see where his balls used to be.

>>59541656

What ever happened to Bristol Ridge? I want a cosy engineering sample placeholder for my AM4 board until Raven Ridge :^(

>>59542119

Pretty much mobile only. Was supposed to be AM4 but Idno. You can find the A12-9X00 in some r a r e laptops

>>59537900
>just because there are shitty and non-existant parallel-processing software means there is no use for multi-core processors


lol you're a fucking retard, i bet you haven't even heard of things like intel phi

>>59541330
>>59541308
>>59541397
>>59541450
>>59541489
>>59541577
>>59541603
>>59537900

A few things:
1.
Stop samefagging. Your way of speech is so autistic it's pretty obvious
2.
>>>>>>/v/
3.
So you're telling me if I have shit compiling/rendering/simulating in the background while I gaym it automagically completes without taking up any CPU clocks? Same for fucking anything in the back end like services/daemons/etc?! Stop the fucking presses you need a fucking Nobel!
Also if you tell me people who have services running are a fucking minority you really should have prefaced the OP with the fact that you just got lobotomy.

>>Inb4 I was talking universally
Yes you lil bitch, almost every program takes some amount of CPU, even when in background so the "universally" excuse doesn't work.

>> GPU and CPU are different by the definition I pulled out of my ass
Gee wiz, if you weren't so busy huffing glue at the shit school you went to, maybe you'd know that a CPU and GPU are the "same", gpu has a lot more weaker cores (mainly)

>> Also things mean what I want them to, not what they do
Shouldn't you be on medication?

PS: you need to be 18 to post here.

>muh interactivity
OP is an idiot as usual. If your program does that much interaction with user it doesn't need any processor resources worth talking about because it spends most of its time waiting for user input making any discussion on serial performance pointless. Most other things can be multithreaded.
>muh gayemz
Modern games do insane amount of shit in the background mostly regardless of user input. They don't utilize multiple cores properly right now only because no one made them to. It will change soon.

>>59537900

>photoshop isn't multithreaded when it doesn't need to be

you don't say :)

>>59537900
What games fully utilize different threads to optimize fps all the time not just during loading screens?

>>59544108
look at battlefield 1 core utilization and performance

Why do people assume game developers will want to push hardware.

There has been no evidence of this since the clockwars when your top-spec machine would fall below minimum spec in three years.

I'm sure they are as capable as anyone else of reading the Steam data and this generation of consumers is, frankly, out of the habit. If it gets too heavy they'll jump to console at this point anyway.

When was a "minimum spec" an actual news story last?

I'm sure a graphics card won't last forever but a required architecture change? I can't see them having the balls or the consumer being willing.

File: Gustafson.png (93KB, 976x685px) Image search: [Google]

93KB, 976x685px

>>59537900
Literally wrong.

Moar coars, if used effectively, will yield almost linear progress as more shit is rendered.

>games
Who cares?

>>59547227
>you
Who cares?