>Intel introduced a third line, and announced they're going to cease their tick tock development, with Kaby Lake coming before Cannonlake's shrink to 10mm
>AMD can't even put a fight in the enthusiast CPU market, Zen is a desperate effort at competing with mid tier CPUs
>developing smaller nodes is becoming more expensive, harder and with the barebones competition there's less of a reason to push hard for it anyway
>silicon is about to be tapped out with those infinitesimal shrinks
>there's far more money in the mobile market anyway, where they're way far behind Intel's x86 chips, not to mention the entire market pushes for ARM in detriment of x86
Is the desktop CPU market done? Or is just the x86 architecture done? It seems destined to stagnate a slow death. What is ahead of the CPU innovation front? Graphene CPUs?
>>52291372
>Zen
>mid tier
Pick one.
>>52291398
I'll pick everything because it's an architecture not a fucking CPU of any tier.
Zen is going to be a success
>>52291750
I took a screen shot.
Let's see
>developing smaller nodes is becoming more expensive, harder and with the barebones competition there's less of a reason to push hard for it anyway
Everyone on /g/ to get all stuck up on smaller processes and novel new materials.
Apart from switching materials away from silicon (i.e. a major fucking cost as you need to retool fabs -> billions of dollars in capital required), you could also look at improving gates (less leakage etc). FinFETs were the first step.
>there's far more money in the mobile market anyway, where they're way far behind Intel's x86 chips, not to mention the entire market pushes for ARM in detriment of x86
Far more money? Mobile has become incredibly commoditized now, it's a race to the bottom with Chinese SOCs.
Intel is going to focus on servers/datacenter/HPC.
They are slowly gaining a huge lead over everyone else with their Xeons. The acquisition of Altera is just one of their plays in this area.
So far it seems that graphene might be too much of a bitch to get working right.
There are tons of other possible materials that are competing against it.
It's just that graphene was the first to get the spotlight and is now being heralded everywhere as the next go to material.
I think that they're going to try and prolong the life of Silicon by mixing stuff with it, like they just did with the IBM's 7nm chip. It's a mix of germanium and silicon.
Of course that's just a temporary solution.
That tech is fucking dead but like anon above me said, it's a huge cost to move onto other materials, so no one wants to be the first to do it.
The scientist have been looking into photonics too for quite some time.
They just got a first photonic chip working and it shows 10-50 times the speed, which isn't really that impressive if you think about it, but apparently photonic stuff has very low power draw.
This is probably something that we're going to see in the mobile world.
>>52291832
why do you have tri-gate after omega-gate?
doesn't having more gate around the drain make for a better transistor?
Isn't it supposed to go FinFet->PiGate->TriGate->Omega-Gate->GAA?
>>52292325
I just got the image off wikipedia.
We need to move to parallel programming. GPU-style.
Multiple cores > higher clock speed.
More cores is exponentially better for heat dissipation, power consumption and instructions per second.
Programming in parallel is hard to implement, debug etc so stagnation is what we're experiencing.
A new company called Parallella are doing it. 64+ cores
https://www.parallella.org/
>>52292047
>photonic
Star trek coming true!
>>52292505
>moar cores
Coming soon from AMD:
6400 cores at 1Hz.
Debug this, motherfucker.
In cinemas Winter 2036
>>52292047
>photonic chip
This... Might be interesting.
Laptops and tablets are the new focus. That's what most people buy these days.
And if you want to still keep CPU innovation, go back 10 years and make sure that Apple sticked with PowerPC.
Desktops are becoming speciality products.
>>52293377
>Coming soon from AMD:
Nah, more likely from Intel or IBM.
>>52293611
what exactly is this?
>>52293647
A metric fuckload of atom cores for highly parallelized tasks.
More cores, seriously.
IPC improvement is getting more insignificant every year. The only way to get more performance is to toss in more cores and hope devs will stop being lazy nigers.
>>52293647
>what exactly is this?
Project Larrabee. Lots of advanced Pentium M cores, put in a cluster, for doing GPGPU tasks.
In the end they didn't manage to turn it into a functional GPU (possibly because Intel just does not have a gpu driver team), so they turned it into a sole accelerator card.
>>52293754
So is it like a GPU?
>>52292047
>photonic
will there be aftermarket RGB LED kits available?
Or at least blue ones for low temps.
>>52292047
>They just got a first photonic chip working
some sauce on this? Who, where?
>>52292047
>photonic
It's smaller, less power, less heat, more speed
>>52293802
kinda. Higher bandwidth, cpu features, (iirc) hyperthreading, and using standard ddr instead of gddr memory.
>>52292505
Do you know what GPU-style parallel programming is?
All the cores of the GPU execute the same instruction, just on different sets of data. This is not as practical in non-3D applications.
Secondarily, making multiple cores isn't the challenge. Making our programs fully utilize them is. We're having a hard enough time taking advantage of a mere 4 cores.
>>52292047
Bring on the photons.
>>52292505
>A new company called Parallella are doing it. 64+ cores
The site you linked says 16 cores.
>>52292047
I seriously hope people here didn't read "speed" as "performance". My personal bet is on spintronics though, spin memory coming first. And before that, molybdenum disulfide CPUs, or some better 2d material.
>>52292505
There's no "move" to parallel algorithms. There's only a small number of parallelizable algorithms. An example of always linear algorithm - "do something with the result of the previous function". Also check out Amdahl's law.
>>52293611
This one is old.
>>52294628
>four threads per core
That's new I think. Intel has avoided >2 threads for regular CPUs due to the awful scaling. I guess these co-processors must be designed around tasks that have a lot of pauses to make the additional threads worth it.
>>52294735
Its partially that, partially the fact that the actual CPU cores themselves dont have as much of the fancy execution hardware that normal desktop CPUs do, so they have a larger number of execution bubbles.
>>52292505
Most problems aren't parallelizable though...
(Is that a word? Parallelizable?)
>>52292505
>>52291750
Like Bulldozer was, I fucking hope AMD dies. They can't make anything competitive to save their lives
>>52291372
x86 has nothing to do with the material the transistors are constructed from...
>>52294480
Amdahl's law only applies to a fixed set of data
>>52295806
I didn't say that. I said x86 seems to be in a dead end, where the improvements over the last chips are slowing to a crawl. The desktop market isn't growing, and the big money is in mobiles which are dominated by ARM.
AMD slacking big time in x86 CPUs is also hurting the overall picture, without competition and with silicon being on its last legs, who will invest in x86 chips? This could change with maybe graphene, other material or photonics as other anon said.
GO AMD GO
>>52291372
X86 is more than 40 fucking years old. It's just a bad architecture with no room for further improvement. We desperately need something new for high power CPUs. For the low end ARM is perfect
>>52296119
I like you.
There is lots of shit to do, change materials, trenary transistors, photonics...
>>52296280
Fuckin this!
>>52296280
I'm not well versed in tech related stuff.
But if a new architecture came out, all software made for x86 would have to be ported/could not be utilized, right?
Also there won't be a new architecture cause there's no money to be made with one.
Desktops are becoming niche, if they aren't already.
>>52296447
Yeah, you'd need an x86 emulator
>>52291799
Careful guys, this guy printscreened it
>>52296479
Gross. Like putting a geo engine in a f150 body.
>>52291750
That's what they said about bulldozer.
>>52296280
It'd mean dumping most software pretty much, and you can forget about emulating something as big as x86 to any reasonable speed for the forseeable future
>>52296280
Lower end: ARM
Higher end: Power8
Pseudo Quantum cpus
>>52296447
>>52296627
Yep, software is the problem that we are still stuck on x86. Intel has top notch engineers, but at this point it's like they're putting a formula 1 engine into a fucking Bobby car. X86 needs to die
>>52296764
is the age of an architecture that much of a limitation? Aren't the architectures IBM uses also old as fuck?
>>52296786
The problem is that back then, when x86 was made, they didn't consider all aspects of a chip like stuff related to computer science. Thus the design is flawed. Just look at ARM it also 20-30 years since it's design, but they considered everything and now we see apple already beating Intel in the sub 15watt market, because ARM had a specific goal in mind:Low power.
Some people could probably explain it better than me though
CPU transistors are approaching to the size of an atom, it is the physical limit and therefore this industry is fucked
>>52291750
For intel.
>>52296764
perhaps one that retains the x86 microcode to ease the transition, as anything that can't run x86 is severely handicapped. Part of the reason the RPi and similar projects have not seen greater success is their ARM architecture cuts out 80% of software.
>>52296786
Its RISC vs CISC. x86 has many features available at a very low level, allowing complex operations to be handled within microcode. RISC architectures offer a more limited set, but is much more efficient at what it instructions it does have. This allows things like arm to operated at comparable speeds and lithography, yet with significantly lower TDP. However, you're not likely to see something like VT-D and other such complex mechanisms to be handled at the opcode level.
>>52297025
>their ARM architecture cuts out 80% of software.
no it doesnt
the Debian ARM repos have somehing like 99% of the software thats in the amd64 repos
youre just a faggot who uses proprietary software
>>52297116
Raspbian has a notably smaller amount, due to the earlier boards using ARMv6. For as long as they want to support the first 4-5 boards, their going to suffer this handicap. Not to mention the dearth of 64 bit. There's like a dozen RPi-alikes, and few are of a recent architecture revision.
So no, the early players in this low power renaissance dun fucked up, and are going to support their fuckups for quite some time, and kneecap the biggest x86 alternative. This will inevitably further fracture the x86-alternative front, and cement Intel's strangle hold (and AMD's death grasp) on the non-ultramobile market sectors.
And yes, proprietary software. There's a lot of it.
So fuck off, or apply yourself.
>>52297025
>However, you're not likely to see something like VT-D and other such complex mechanisms to be handled at the opcode level.
IBM's mainframes supported nested VMs in hardware before the 8086 even existed.
>>52296280
>The age of the architecture determines how good it is
Wow, ARM must be only marginally less shitposting since it's 31 years old now!
Seriously though, all current ARM chips are made for power efficiency at low enough TDPs to be passively cooled, there is no ARM chip that will meet even the most basic definition of high performance computing. That's OK though, they're not being designed for it. The only benchmark dragged around is geekbench, which is terrible to say the least. I highly advise people to look at linpac instead.
The x86_64 architecture is doing very well, with the best single threaded performance available and that's before you start going into overclocking consumer chips. POWER8 is very, very good at a large number of very good threads but they're still not quite up to what Intel can provide in most single threaded use cases. Generally speaking, this is irrelevant in the server world though so they're still the better choice if you can afford them.
/g/ has a lot of hatred for x86 and not much of it is rational. I agree that Intel trying to shoehorn it into tablets and phones is not exactly working out but ARM will never take over the desktop or go beyond microservers for that matter.
>>52293868
HP The Machine.
Make an google anon
>>52291372
>enthusiast market
Video games?
>>52297827
>The x86_64 architecture is doing very well, with the best single threaded performance available
>What is IBM Power8 @ 5 GHz?
>>52297827
>>52299559
>>52299579
Intel plebs BTFO
>>52299590
>Intel
>floating points performance
Pick one.
>>52297827
>/g/ has a lot of hatred for x86
It comes from whoever is teaching them in school, most courses try to avoid x86 (due to its relative complexity) and they believe this means it is deprecated and all our desktops will be powered by RPi CPUs.
>>52299676
2016 will be the year of the ARM laptop
>>52296102
>The desktop market isn't growing
Probably because you still can use your 10 year old Core2Duo for web browsing and office work without any problems.
>>52292047
IBM has invested (don't know how much) into graphene. It's probably going to be the next paradigm. However, after that's been taken to its limit, we're probably going to use photonic chips. It's definitely sci fi, and definitely a possibility.
Keep in mind: silicon and solid state transistors were a mind blowing invention at their inception. A photonic circuit is completely possible.
>>52297025
>Its RISC vs CISC. x86 has many features available at a very low level, allowing complex operations to be handled within microcode. RISC architectures offer a more limited set, but is much more efficient at what it instructions it does have
>not knowing that x86 just decodes CISC to RISC
There's no difference these days, it's all down to how well it executes instructions.
>>52295470
>I fucking hope AMD dies
thats a spirit you have there to show off what a narrow minded faggot you are
>>52291372
>Zen is a desperate effort at competing with mid tier CPUs
>>52299579
>IBM marketing slides
>Still shows Intel performing better per thread
Yeah, you're not disproving my point.
>>52301608
>This whole thread is filled with idiots complaining about things and demanding new things they don't understand
>One anon isn't a complete putz.
Thanks for not being a complete putz.
>>52299700
>The performance of a smartphone in a laptop
Yeah, no. The only reason the core modules series is vaguely acceptable is because of the extremely aggressive turbo boost letting it massively exceed its TDP for as long as temperatures allow.
>>52301608
This guy gets it.
>>52303420
How is it exactly wrong?
Excuse people for not expecting AMD to go toe to toe with Skylake, but at least match Sandy Lake IPC, after Bulldozer and other failures.
>>52292047
SiGe isn't new, its already extensively used. What IBM did was create a SiGe channel as apposed to using it around the source and drain wells like we see today in bulk chips.
>>52292325
>gate
>around the drain
Jesus. Stop.
>>52301608
That's what my drunk brain was trying to explain. They handle microcode differently. x86 pulled ahead because it had more low level features early on. Nowadays the difference is marginal.
What's stopping them from creating larger CPUs (like increased width/length or the height) to pack more transistors?
>>52307555
Stop posting this copypasta bullshit in every thread.
>>52298510
I meant something that exists not nice sounding ideas.
We'll see when Zen actually hits market, if it's good, then great, we got some competition, if not, there will be threads all over /g/ for at least weeks after about it, and much butthurt to flow.
I'll probably pick up a FX-6330 or FX-8320E to upgrade from my X2 250.
We've come to the point where most of the normal masses don't need anything more than an i5/FX. The next exciting thing on the forefront where everyone's pouring their money into is VR/AR and the surrounding ecosystem.
With DX12/Vulcan faster CPUs will become even less of importance.
Personally, I don't think ARM will ever be of importance outside of low power consumer devices like the phone/tablet and hobby projects like the Pi, it's not going to replace x86_64, heck with all those Intel android devices popping up you'd think that the ARM CPU market is actually slowing down.
Mediatek does lots of cores but nothing really supports it.
Snapdragon has a monopoly on fire breathing hot CPUs with not much innovation.
Allwinner, Rockchip and other crappy ARM CPUs suck compatibility wise and are slow.
Samsung is supposedly going to sell their Exynos SoC to other brands to eat into Snapdragon's marketshare, we'll see...
>>52299590
Is this a joke? How's this trash still alive?
We need an event-oriented architecture. It's possible to make a clockless one. And instructions with higher level of abstraction - the CPU/compiler will have more information about the programmer's intent. This will allow to eliminate whole parts of operating system, the very concept of cores and many other benefits. Also on-chip RAM instead of caches - binning will be pretty bad but the idea is jewish enough(imagine, a new CPU to have more RAM) to be pulled off, and the performance benefits are evident.
