Chinese kaby lake review, 1% IPC increase

Chinese kaby lake review, 1% IPC increase.
diy.pconline.com.cn/851/8515185_all.html

Other urls found in this thread:

nickcraver.com/blog/2016/02/17/stack-overflow-the-architecture-2016-edition/
twitter.com/NSFWRedditImage

lmao

where's the 40% ipc increase u cucks

So slightly higher speed with a lower power usage?

AMD is fucking finished. Zen can't even reach 4 GHz and has lower IPC too.

its the same power usage you massive dongle
skylake is probably worse because its older and more worn

worse by 1%

intel isnt trying anymore

Factory overclocked skylake on same tdp

thank you based intel

Also core voltage is higher than skylake, not likely to OC better

its over intel is finished zen will be the hot new shit.

>AMD is shit so intel is not forced to innovate
oh wow gee who would've guessed

So this Chinese website is trying to tell me that Intel is releasing a chip that consumes only 4% less power and performs the same? Sounds like bullshit, even for Intel.

I have a top binned 2600K

I'm gonna be sad when my motherboard inevitably dies and I have to spend £450 for a 13% average speed increase (assuming my 6700/7700 will hit 4.5GHz comfortably).

thats over at the AMD ZEN department

buy used

decent used z77 mobos are like £150+ now

hopefully zen will be out by the time it dies

looks like it's useful only for notebooks and servers

just get a 4790k and a z97 mobo, the combo goes for around 200€

delidded my 4790k and am getting comfy 4.8GHz @ 1.25V max temp is 72 during prime

where the fuck are you buying your 4790Ks from? Even on eBay they're >£200 and retail they're the same price as a 6700K.

local used hardware marketplace forum... got my gtx 1080 strix from there for 605€

You realise that they already did this with Devil's Canyon, right? The 4790K was quite literally just a 4770K with a more aggressive factory overclock, despite all Intel's hype and bluster. The promised higher overclocking was complete bullshit.

These "refresh" releases aren't even going to be anything more than that. Kaby Lake literally only exists because Cannonlake, i.e. the one with the potential to bring some improvements, isn't ready yet.

>literally margin of error

>Cannonlake
Eh, they are likely to refresh it again with Coffee Lake

>buying k series CPUs
You are all so retarded

>he doesn't want FREE performance

>this retarded "theory"

The last great architectural improvement from Intel was the u-op cache, which Zen (finally) catches up with.

There's no real obvious "next big thing" in improving x86 integer pipelines, hence Intel focusing on adding new shit to the ISA like more accelerated functions, fatter SIMD, TSX, and iGPU shit.

>"""""""""""""theory""""""""""""""

AMD being shit gave Intel fatter profit margins, but it isn't like Intel hasn't been trying (and largely failing) to do better.

I'm still looking forward to a non-x86 future. Pipelines for data through efficient architectures like ARM and POWER offer better optimizations translated into assembly code.

there are always tradeoffs.
microcode translation and a stronger memory model take more area and power, but you do get denser machine code and the many consequential benefits from that.

They did OC better but much less than what was promised. Intel marketed the Devil's Canyon K series CPUs being capable of "5 GHz on air" which ended up being true for a tiny fraction of the chips. They still clocked higher than Haslel but suffered from the same poor heat transfer from the chip to IHS. Overall lackluster.

did they perform significantly better when de-lidded?

> not knowing what kaby lake is intended for
> being this retarded

kaby lake is literally just a die shrink of skylake with a few added items not related to the performance of the CPU(native USB 3.1 is one)

> Kaby Lake is a die shrink of anything.

Stopping you right there m8.

>kaby lake is literally just a die shrink

2016 is the tipping point where desktop hardware stagnated so bad we literally saw no improvements except in gpu Tech.

Cpu x86 is dead in the water even Moar corez didn't save it

Arm based apu gpgpu cpu when?

KL is an overclocked respin of SL with a new chipset.
Next year's Coffee Lake will just be a respin with 6 cores.
You aren't getting your shrink until ca. 2019, and that's just if Intel doesn't fuck up 10nm any further.

Intel might be getting close to the limit for how fast x86 can actually go. If thats true this is good news for the market because amd willl eventually catch up in like 10 years and then they can't price gouge us.

>limit for how fast x86 can actually go
Never in my entire life I've read something as stupid as this.

It would help if they made a 8 core i7 or i5 that was actually fucking affordable instead of rehashing the i7 960 for 9 fucking years 4 cores are absolutely useless these days my i5 4690 is 100% just loading most games or installing shit
Lol you think they won't hit a wall? They hit it 8 fucking years ago.

That and and royally fucked up if Zen doesn't make Intel panick and release another i7 920 were fucked

>Lol you think they won't hit a wall? They hit it 8 fucking years ago.
The x86 architecture itself? No. Assuming you'll have better chips using this architecture it should scale nicely. Intel not wanting to come up with something better is the lack of competition, but that doesn't mean the architecture would not scale.

You have no idea how cpus work if you think an instruction set can scale indefinitely.

Nice strawman argument.

It's not as dumb as you might think.
We've clearly long since passed the point of diminishing returns (see: Pollack's Rule), and hard bounds on pipeline stages are imposed by seemingly esoteric factors like the distance between the integer execution/retirement logic and the far edge of the L1D array.

user would probably have been more accurate in saying we're approaching the end of traditional Harvard/von Neumann architectures on silicon lithography though.
We might be able to get one or two more shrinks, but there's not a lot of obvious room for improvements in architectures anymore.

Yes if you use good paste. Nearly everything is better than whatever insulator Intel used. I have a 4790K which can hit 5 GHz on 1.4Vcore. Delidding and changing the paste to Liquid Pro lowered temps in Prime95 from near 100c to 82-85c. In other heavy loads like video encoding it never goes above 65c and with the 1.35Vcore I keep it at it's stable 4.9 GHz and below 60c. I also had 4770k which I delidded and the results were similar but that chip wasn't stable at 4.4 GHz or above.

Those are limitations of superscalar architectures rather than x86, ARM will have the same problem. This x86 sucks ARM is the future rhetoric doesn't make sense, especially after the instruction of x86 microops and ARM introducing complex extensions, they look the same superscalar shit inside

>1.4v

Jesus fuck user. Do you wear a skimpy leather outfit while you dominate your vrms like that?

Well done, Intel.

Was the highest I tried when I pushed it. I don't use 1.4Vcore in any normal usage as the tradeoff isn't worth it for me. Huge temps and barely any added OC.

Also I killed my Z87 board with that already.

>but it isn't like Intel hasn't been trying (and largely failing) to do better.

No, that's exactly what it is like. Why try to do better when they have no competition? Just repackage the same shit for 4 years. Bump up the clocks by 5% to get IPC increase each time.

Plus they are also fucking over overclocking, first by limiting it to the K cpus (which sell at a premium), then by fucking up the thermal interface on those cpus so they cannot fucking overclock anyway due to thermal limits, then they sent C&D letters to motherboard manufacturers who re-enabled BCLK overclocking.

Factors like OOO and superscaler do complicate things, but even a single-issue in-order chip would not clock much higher nor have a much shorter pipeline than what we see now with x86, Power, ARM, or whatever.

Even in-order VLIW like Itanium with all its static scheduling and branch predictor coddling tricks didn't go *that* much further in integer when you look at it on a larger timeline.

Intel's R&D budget is staggering.
If you have sauce on them spending it all on coke and hookers, pls share.

my paste is long overdue. what do you recommend? is that liquid pro difficult to apply? i'm running [email protected]

Oh VLIW, it was the future that was promised, until they saw the compiler and optimization hell. All that Itanium shilling killed Alpha

not c&h, rather on IoT™™™

AMD's automatic overclocking software uses 1.5v by default.

WE DID IT!

you cant compare the two arches like that.

AMD's transistors are almost 5x the size of intel's on those chips. Ofc they need more power.

People say it is but a toothpick did the job easily. The surface tension it has makes applying a bit more tricky than normal pastes. The manufacturer has a video in YouTube on how to install. Liquid Ultra is nearly as good and supposedly easier to spread.

and mobile users surpassed desktop in internet usage.

On tumblr/reddit-tier websites, yes. Not overall.

Time for AMD to shine.

but if it's the same thing why release it

Is it more power efficient?

They had no competition in the self build market for a long time. But lot of cheap laptops are AMD and better than Intel of the same price so the win the laptops.

Suckers. My 5820K will hold for years with this trend going.

AMD needs to get Intel in competitive mode again because this is absolutely pathetic.

>If you have sauce on them spending it all on coke and hookers, pls share.

Gentlemen do not tell, user.

Probably has a new IGP with more features.

>muh phones are the future

If ARM is so good why do they not fucking release a server board with 8 or 16 A72 cores? Smartphones with these cores were available a fucking year ago. It's not like it's impossible. Even the Raspberry PI 3 is a joke compared to phones. Even the companies that did release something with their outdated shitware are trying to hide their products as much as possible. On top all of this is the shitty locked down nature of most SoCs. They give you an outdated kernel with their own custom patches. You need a community based around a specific SoC just to run software on it.

ARM is shit.

largely, cpus are fast enough, there are very few applications where you are required to have a faster single chip, opposed to more chips.

lets put it this way,

Intel makes a 60 watt chip, and next year lets say they make a 30 watt chip.

same power for number crunching, just less power.

for average people, this is fuck all, but for a server farm, supercomputer, cluster this is cutting the cost of operations almost in half and when you are talking megawattts, this is fucking sgnificant.

I have no doubt that if intel didn't reign in the thermals, we would see legitimate arm servers by now because power use > power when their ability to get shit done is the same.

that said, intel does have VERY shitty practices, from the way they handle heat spreaders, the overclocking, and if the c&d is true.

Next cycle is going to smash all kinds of records with a 0.009% performance gain.

And everyone must buy it or be left behind.

(as new software will only work on new hardware, and don't look to Linux for answers as new gimmick features tethered to proprietary divers are made just to mess with power users and new SoC models for true pain)

(lowest competition as companies refuse to admit this is really become a very costly commodity market, and any economist can tell you why thats a future horror story for us all)

>largely, cpus are fast enough, there are very few applications where you are required to have a faster single chip, opposed to more chips.
Most RDBMS scale vertically. Stackoverflow runs with four database servers.

nickcraver.com/blog/2016/02/17/stack-overflow-the-architecture-2016-edition/

>(((FREE)))

K E K


>not buying i5 6400 and OC to 4.6 Ghz
WTF I hate Sup Forums now.

what the fuck kind of goal post moving bullshit is this?

>relying on what the People's democratic dictatorship thinks of an American CPU producer and supplier

Ps. OP sucks dick for skittles