I've just been wondering about something, /g/.
Basically, why don't Intel or AMD make CPUs with one or two HUGE cores to increase single threaded performance, and like 2-6 smaller cores for multithreaded performance?
From what I understand, adding more cores adds more performance per area, and is more power efficient, but doesn't scale well with linear workloads, which would benefit more from powerful cores.
What if they made a 6 core CPU, which half the die space dedicated to two primary cores, and the other half to 4 secondary cores?
why don't they just make a 3kg cpu processor
>>51603623
Interesting idea. I've noticed that some of the ARM core variants are starting to do this sort of thing.
>>51603692
I mean, it's a super simple idea. They probably already thought about it 10 or 15 years ago when they were developing the first multicore CPUs. I'd just like to know why it's not working/worth the effort.
The best idea I can come up with is that the cores they use are already the most powerful they can make at the time, and they just can't make any bigger cores than they already do.
>>51603623
Because Intel's cores are already the best in the business and they give you 4 of them for a song.
Secondary cores? See hyperthreading. Why waste the development time and die space on a shitty core that nobody is going to code for?
Because operating system schedulers aren't designed for it. Even as it is, they still suck at properly scheduling threads for hyperthreaded processors, i.e. staggering them so that you're using all 4 physical cores before using the "secondary" logical cores.
Also, what >>51603786 said. The PS3's Cell processor was just like you describe, but barely anybody actually used it to its potential.
>>51603786
All Intel worked on since Sandy Bridge is, quite literally, power efficiency. I mean, I'd take a quad core with 50% faster cores over an 8 core any time, even for the same price.
>>51603818
Don't the schedulers get designed for the hardware, not the other way round?
>>51603692
>are starting to do this thing
They've BEEN doing bigLITTLE architecture, and it's NOT for performance reasons. The powerful cores are turned off when they're not needed so that the weaker cores can sip power for increased battery life. This is the best compromise that can be done to simultaneous increase performance on small workloads while increase battery longevity.
>>51603623
Because every core is already a "HUGE" core, retard. And by "HUGE" I mean as physically tiny as possible while being as FAST as possible. It still takes a finite amount of time for electrons to pass through transistors and gates throughout the processor to get to their end destination. This stream of electrons dictates the result of a "calculation". This calculation is faster if the distance they have to travel is smaller.
Think of it like this. You need to go on a flight to Japan, but you're already traveling at the limit that aircrafts can travel without exploding due to air pressure. It takes 6 hours to travel there as it is. A die-shrink is like shrinking the planet. Instead of travelling thousands of miles to Japan, the world is shrunk to make the total distance to Japan smaller, making the end "calculation" faster.
Making a PHYSICALLY LARGE core would be like trying to get to Japan faster by increasing the size of the planet. You want process node shrinks, not expansions.
They've hit the limit to how much they can increase IPC per core per year, so they instead do clever tricks like multi-core and hyperthreading to allow software to make use of parallel processing, which can achieve a similar effect of increased IPC but only for processes which can be computer in parallel
>>51603774
Electronics don't work like engines, you complete fucking moron.
Bigger engines = moar power.
Bigger cores != moar speed.
>>51603848
>I'd take a quad core with 50% faster cores over an 8 core any time
Then make one.
>>51603623
>>51603658
Cores can only be made so big before you have to start lowering the clock rate to make them stable. Light only travels 75mm in 1/4000000000 of a second (4 Ghz clock rate).
>>51603818
>The PS3's Cell processor was just like you describe
It wasn't actually.
What OP described would be like a chip with two Haswell cores and four Cherry Trail cores, or two Excavator modules and four Jaguar cores, for example. In this case all units are fully-functional cores.
Cell is not that, it only has one fully-functional core and eight additional SIMD units, which are highly specilized processor and are not even remotely near the versatility of an actual core. Cell's SPEs are like GPGPU, good for highly parallel tasks but way more limited than a CPU in what it can do.
>>51603918
Thanks, that makes sense. But, as >>51603859 pointed out, light travels 75cm per clock cycle in a 4 GHz CPU. Google tells current moves about 30cm in the same time frame, so technically, we still could make cores like 3x3 cm in size, and they're still small enough for the current to move around.
Guys I need your help. So, currently I have an 8320 and my computer just straight up pooped on me twice over the years. 2 of my gpus(770s over the years) have died with the most recent being 2 days ago. I'm absolutely pissed and just want to get rid of my motherboard(which had the onboard sound die on me) and power supply(some seasonic 750x). So I'm going to sell all 3 components on eBay and use that money to buy all new cpu, mobo and PSU.
Another option is to just replace the motherboard and power supply I guess. I just don't see the need to spend about 100 on a board for 3 generation old processor. I wanna taste that Intel performance whole also being cost effective.
I was wondering, what Intel chip to go for? A 6600k or a 4690k? I don't do anything intensive at all really. The most I'll be doing is writing up papers while using multisim/matlab with maybe a bunch of tabs opened. Of course I don't expect any of these chips to even break a sweat doing these tasks. So what would you guys do granted you have limited funds to spend(hence why I'm selling the parts) Also, for long term use in mind, what's the deal with directx 12?
>>51603692>>51603692
And have failed miserably with the snapdragon 810. The 808 didn't throttle because it had 2 less big cores and a weaker gpu. Only Samsung have kind if done well with it in the exynos 7420. Don't even mention mediakek, they have a bloody 12 core big.LITTLE.medium.TINY set up which benchmarks along with a 2 year old snapdragon 600.
Probably blame Qualcomm for milking the krait cores for too long and then when tasked with making a 64 bit soc to contest the apple a8, they thought that they could throw a few a7 and a57 cortexes together and call it a day.They were idiots who were in control of an effective monopoly, thus oems had to rely on the 810 or the 808 as a current gen soc. The 808 as said wasn't good enough and the 810 was a house fire. They tried to "fix" the 810 by throttling it so hard that it actually did worse than last gen 801 child in numerous benchmarks.
The problem wasn't the chip tbqh, but that it had a high tdp to run effectively. In tablets like the z4 and the mipad, its fine and in fact the big.LITTLE arrangement dies show battery life improvements. But when you out this in a 5 inch phone, fermis are bound to happen.
>>51604063
You do not mention doing any tasks that require a lot of processing power. An AMD APU should be plenty for your needs.
DX12 will make more efficient use of multiple cores, and doesn't require as powerful cores as DX11 to achieve the same frame rates.
>>51604054
That's milimeters per second, not centimeters, lrn2metric senpai.
>>51604167
Still enough for a 3cm core. The speed at which current flows is, currently, not the limiting factor of clock speed, but rather heat and transistor flipping speeds. It's also not the limiting factor for core size.
>>51604054
Electromagnetic wave travel through different materials at different speeds regardless of the clock speed. In general, the pcb material (also referred to as the dielectric) can be made of various materials. For example, a common material is FR-4 (cheapest). The Er (Epsilon real) is varies greatly with different epoxy/fiberglass composition, but generally specified around 4.2. The factor that EM waves are slowed down in this materials is by a factor of sqrt(Er). So, for the Er=4.2 example, EM waves would travel at approximately 1/2 the speed of light in a vacuum.
>>51604243
>light travels 7.5 cm/clock cycle
>current travels at 3.0 cm/clock cycle
>making a core so large that the current won't necessarily be able to make it across in one clock cycle
>thinking this would result in a stable processor
baka
>>51603623
Bigger Cores != Faster Cores
Faster Transistors = Faster Cores
>ITT People who can't into electronics
>>51604130
So what do you recommend me doing? Just get a new board and psu right? I'm honestly just tired of the amd chip. The heat and the power is a pain to deal with. It also seems like my overclocks only last for a few months before it randomly crashes and I have to spend a night overclocking again. The clocks are good as I test with aida64 and prime95 for hours before deeming an OC stable. It's also an absolute necessity to overclock since the difference between stock and a good overclock is 20 fps using Valley benchmark.
Idk, maybe I'm just trying to find an excuse to go Intel this time but I feel like staying with this chip would be a waste. Anyone have any more insight to share?
>>51604427
Do you want to game on it? Do any other tasks that need a high performance CPU?
If you want to get a good PSU that'll last you a decade or so get a Snow Silent 750W or a RM1000x. I don't know anything about cheaper PSUs.
Get a AMD APU if all you're doing is office/media tasks.
Get a 6600K for gaming.
Get a CPU with hyperthreading for heavy CPU workloads. 5820K gives you a lot of bang for the buck in that regard.
>>51604489
For cheaper/lower wattage PSUs, Seasonic M12II Evo or bust.
>>51604560
Personally, I've started thinking that everyone should just drop 100-150$ on a good case and 150-200$ on a highest tier PSU once every decade, and then enjoy. Both will last through many builds.
>>51604489
I see, I forgot to mention I do a lot of gaming( or at least try to) hence the sli 770s lol. I also use dolphin emulator as well as pcsx2. I know for sure those emulators run way better on Intel chips over amd chips. I get a bit of stutter in dolphin sometimes and pcsx2 is kind of a no go for some more intensive games. Seems like I should go on and get the 6600k.
As for the PSU, I just get 750w gold minimum for dual card setups. I will probably get an evga psu since I hate corsair now.
Thanks for the input anon. I need some more opinions though, probably should've posted in a stupid questions thread as well.
>>51604655
Go with the Snow Silent 750W. It's worth the extra 50 or whatever bucks.
>>51604646
Yeah, everyone I've helped with their builds so far I've told to get a nice, solid case and PSU because those are reuseable components for the next one.
I might be dropping €140 on an Enthoo Luxe sometime in the near future myself, my current case is alright but it has some quality and design issues that come with a €90 price point.
>>51604560
I have a feeling it'd my seasonic psu that's causing the trouble and fried my second card. Now I have 1 gtx 770 left and the other 2 are just plain dead. I'm getting desperate and might bake them to see if anything happens.I did send the psu for rma last year though so I don't know what's the deal with my system. Every other pc I've built for people have worked flawlessly for years and mine can't last even 1 without something going terribly wrong, this shit sucks.
Bigger physical cores = bigger chance of physical failure on the hardware.
>>51604760
Weird, Seasonic is known for making some of the best PSUs out there.
>>51604812
Exactly. I'm just really stumped as to what the problem could be. I forgot to mention I even had to rma the motherboard as well during that time. If I can, maybe I'll try and rma the board and power supply again... Just leaning on the idea of selling them both though and starting clean. 170 bucks for the snow 750 is a bit too steep for me. Most I'd be willing to spend on a psu is 100-120.
I think I'll tough it out on the gpu side of things and wait for the new ones next year and sell the 770 that's still working(ironically, the ones that died were both msi gaming models).
>>51604990
Maybe try the Antec Edge 750W. It's a Seasonic OEM unit and the price is pretty resonable over here.
>>51605027
Thanks for the suggestion anon. It's sold out on newegg though. I'll wait for cyber Monday and see if I can find a good deal on a psu. As long as the psu is modular, has a fan that only spins up when the psu gets hot and is 80+ gold or higher. If I can get all of that for about 100 or so, I'll get it regardless of brand except corsair.
I'm wondering if seasonic will compensate me if I say their psu killed my onboard sound and graphics card. Gonna look up if they even do such a thing and give it a try. As you can tell, I'm really desperate at this point.
