>zen http://wccftech.com/intel-lga-3647- socket-skylake-

>>56941040
This is sick

I can't wait for a dual socket motherboard with this thick ass bitch

>only $2000

>>56941040
>MOAR CORES

>>56941040
> Knights Landing will be built using up to 72 Airmont (Atom) cores with four threads per core, supporting for up to 384 GB of "far" DDR4 RAM and 8–16 GB of stacked "near" 3D MCDRAM, a version of High Bandwidth Memory. Each core will have two 512-bit vector units and will support AVX-512 SIMD instructions, specifically the Intel AVX-512 Foundational Instructions (AVX-512F) with Intel AVX-512 Conflict Detection Instructions (AVX-512CD), Intel AVX-512 Exponential and Reciprocal Instructions (AVX-512ER), and Intel AVX-512 Prefetch Instructions (AVX-512PF).

holy fuck this is more insane than bulldozer.
someone more knowledgeable please explain who would want this and not just a GPU.

>>56942639
>72 atom cores
hahahahahaha.
add moar coars.

>>56942639
>someone more knowledgeable please explain who would want this and not just a GPU
because a gpu is not a cpu you ninny

>>56942639
Ray tracing and video encoding. Though even then a xeon-D 1587 would be a much better choice.

File: Intel-Skylake-EX-Processor.jpg (60KB, 800x450px) Image search: [Google]

60KB, 800x450px

>>56941040
THICC

File: 1475064776011.jpg (124KB, 790x1080px) Image search: [Google]

124KB, 790x1080px

>>56942639
>Zen is gonna make Intel lower prices, guys! Competition is good!

instead, Intel is forced to make retarded products like these to compete with even more retarded AMD curryshit.

Anyone else still think AMD isnt a fucking tumor on the tech industry?

>>56941040
>http://wccftech.com/intel-lga-3647-socket-skylake-ex-knights-landing/


ohh dear the days of the enthusiast Core i7 sharing the same socket as the server processors is over it seems. looks like they're going to LGA2066.

>>56941159
>I can't wait for a dual socket motherboard with this thick ass bitch
There won't be one.

>>56942934
>LGA2066
what's even the point of adding that few pins?
that's like barely enough for another 4x PCIe link, if even that.

>>56942833
Moar cores is better!

>>56943015
These aren't amd's """cores"""

>>56943037
Exactly, they're literally worse.

>>56942639
While GPUs handle massively parallel small number workloads, the Xeon Phi handles massively parallel larger number workloads. Its only purpose for existing is basically AVX, larger vector ops.
They're pretty handy for some simulations which is why they're becoming more and more present in HPC.
The number 1 super compute cluster in the world now is based on a custom MIPS design and intel Xeon Phi coprocessors.

>>56942833
You're utterly retarded.

>>56943062
How?

Are you retarded?

>>56943090
You clearly are. Those are airmont atom cores on that piece of shit which are actually just silvermont atom cores shrunken down to 14nm lithography.

They literally have worse IPC than a fucking AMD steamroller core.

That's not even worst part which is because so many of these tiny turds are used, it results in a ~300W TDP which is a joke compared to a Xeon-D 1587 (65W TDP).

>>56942666

/thread

>>56943156
It is Silvermont mod. Core is similar, but they added4-way SMT and two AVX-512 units, which give it all the brute power (normal Silvermont has weak in-order FPU without even AVX support).

>>56942666
>I don't get what these chips are for but want to say something cool

>>56943365
These things are a massive meme. Please tell me what they have (beside shit IPC) compared to a $1,700 Xeon-D 1587.

>>56943584
>says "meme"
>What is HPC

>>56943627
That didn't answer my question. The Xeon-D 1587 is mainly targeted toward low power consumption HPCs.

>>56942639
>someone more knowledgeable please explain who would want this and not just a GPU.

GPUs have limited memory pool capacity, currently maxing at 24-32 GB. Xeon Phi in socket form can pack 6 DDR4 DIMMS (up to 384 GB) *in addition* to GPU-like high bandwidth memory with 16GB capacity.
GPUs are pain to program
GPUs don't have integrated 100Gb/s links for cluster communication
GPUs need extra CPU to host them
GPUs need complex driver that all your code has to go through. This causes problems with underterminisms/incompatibilites

>>56943696
No it isn't, HPC is high performance computing, supercomputers. FP64 calculations on massive scale with clusters.

Xeon D is for 1socket servers that are not heavy on compute.

>>56942966
Read the fucking link OP posted
>The LGA 3647 socket was first pictured back at Computex. The socket was featured on a server aimed, Gigabyte dual socket motherboard.

>>56944235
Skylake-server will defo have 4S/2S boards and it will use the same socket.

and still only one m2 PCIE lane

File: amd_magny-cours_cpu[1].jpg (36KB, 320x369px) Image search: [Google]

36KB, 320x369px

I remember when /g/ laughed at amd for meme wide cpus ;(

>>56941040
>W E W
>E
>W
Yet another socket.

Im an intel fag, and even I wish amd would have won.

>>56942833
If you legitimately think that competition does not force innovation and/or lower prices you are seriously retarded. I don't even think Zen is going to be worth shit, but you actually have to be retarded if you think a closed market benefits anyone but Intel.

>>56944492
You have to have new socket if you want to add memory channels, anon. For the same reason, server Zens will have new socket too.

>>56944725
hmm, maybe, but maybe they should have included it in skylake in the first place.
Why should they have skylake on two different sockets? inb4 its not skylake, its skylake EEeeeEEEEeeeeeee
This is fucking stupid.

>>56945255
What do you mean, include in Skylake? Skylake for servers hasn't come out yet - when it comes, it will use this monster socket for 2S+, and for 1S and workstation it will use somethign similar to LGA 2011, only with 4 channels. All the info is available from leaks.

Thi sis going to be server platform, mainstram desktop/notebook does not need 6channel, nor would it be economically viable for it.

When did you see mainstream consumer platform (like LGA 115x) use the same infrastructure as server highend?

>>56945394
I dont know that much about this stuff desu. All I know is intel changes sockets like crazy, and I just bought a skylake and at some point want to upgrade to an even better cpu like the i7 extreme editions. Im making the assumption that that the E is the extreme editions for this generation.
I've been living in 2007 and am getting up to date on alot of stuff at once, sorry if my assumptions are wrong.

File: 1472835856740.png (583KB, 536x598px) Image search: [Google]

583KB, 536x598px

>>56942833
>intel makes a product for which I don't understand the purpose
>blame AMD

Quite possibly the single stupidest post I've read in all of my years on /g/ if not a troll.

>>56945563
>I dont know that much about this stuff
Not the nice guy you were talking to but holy shit man just read up on the differences between intel's consumer vs server technologies. It's not a matter of "including" features or not. Consumers don't need and shouldn't pay for features that are required in server environments that will just be unused.

If you're complaining that Skylake motherboards aren't compatible with Haswell chips then that's just straight retardation. intel never wants their new consumer chips running on "old" mobos that may or may not be able to be optimized for the new CPUs features and power needs.

It looks like the modern day version of a Q6600
>Okay guys, let's just glue two of these together and call it a day.

>>56943069
>>56943090
>>56943156
>>56945794
>ur a retard lmao xDD
>no u :^)
>fk u...u r even moar retarded x^D
>yes im more retarded get gut fag rofl :)))
now kiss. this board is truly /b/ 2.0.

>>56945794
I write stuff for a tech news page.
Whenever I cover some enterprise/HPC/server chip/GPU/technology that very obviously is not meant for consumer space and you point it out in the article to be safe, people are still going to compare it with their OC 2500K, 850 Evo or GTX 970 (and you can see that they only count with PC gaming usage as the criterion) and announce that they don't see how would anybody want it.

>article about 64core ARM for chinese supercomputers
>man, this is stupid, 2 cores with better performance like apple has rule any day

In short:
>Not everybody reads about things and not everybody of those understand what they have read
>But everybody wants to post about it

>>56941040
Literally not for consumers

>>56945852
>not the nice guy
Hi dick, Yea I've been reading up on alot. I do some processor intensive stuff, so something like an i3 wouldn't cut it for me, and Id like a ton of cores too so thats why Im kind of annoyed at the lack of options on the lga 1151 socket.
At the time, it seemed like the best option and so far I dont regret it, but in the future when I get back into doing real heavy vm work Id like the option to upgrade to something thats better suited for this kind of stuff.

From the sounds of it, they are about to make 1151 obsolete already. Im under the impression they are releasing kaby lake which is nothing more than a skylake with new video codecs and nothing else, then they want to move on and resocket again.

>>56946126
For more cores, you always needed something from the server-derive LGA2011 line.

Maybe they will be sixcores based on Coffee Lake though, for LGA 1151. In 2018 though.

>>56944235

Yeah whats the power consumption the thing will be a gluttonous pig; the flip side is it will probably have 48 cores.

File: JqHHCHl.png (897KB, 841x1007px) Image search: [Google]

897KB, 841x1007px

>>56943713
>Xeon D is for 1socket servers that are not heavy on compute.
>has the multi-core performance of a 6-core skylake desktop i7 processor
>not heavy on compute
Are you literally retarded?

http://www.cpubenchmark.net/compare.php?cmp%5B%5D=2835&cmp%5B%5D=2785

>>56946188
Dude, what the fuck is that supposed to show? That you know jack shit?
Learn what HPC is. Look what sorta chips go into supercomputer nodes. Look what Intel actually advertises as HPC product and what purposes it proposes for Xeon D. Then come act big here.

>>56946204
Power consumption is very important in HPC applications you dumb fuck. Why would you use a 300W TDP cluster of shitty atom cores when a 65W TDP solution exists?

Also these things aren't used alone, FP64 stuff is done on GPUs which have better performance per watt in FP64 stuff.

Just admit it, these shitty knight landing co processors have no real world use.

>>56946239
>Why would you use a 300W TDP cluster of shitty atom cores when a 65W TDP solution exists?
Workloads that benefit more from parallel processing power than raw performance.

>>56946188
>>56946204
>>56946239
>>56946250
Xeon-D is not for HPC.
Whichever one of you is the asshat saying that is wrong.
So wrong. Holy fuck.
>Why would you use a 300W TDP cluster of shitty atom cores
Why you would even ask that? Have you ever had access to run jobs on a supercomputer or are you just talking out of your ass?

>>56946250
>Workloads that benefit more from parallel processing power than raw performance.
Why the fuck would your run such things through a co-processor? Why not just run them through a GPU that has better FP64 performance like a normal human being?

The Rx 480 has over 300 GFLOPS of FP64 performance.

https://en.wikipedia.org/wiki/AMD_Radeon_400_series

>>56946298
It's used in conjunction with GPUs in HPC. Why do you think it supports 32 PCI-express lanes?

File: puhelin.jpg (254KB, 1280x1038px) Image search: [Google]

254KB, 1280x1038px

>>56943627
High Powah Computah

>>56946239
Literaly cretin.

>cluster of shitty atom cores
Okay... you missed >>56943353 (and >>56943698 which also matter but not directly related). The AVX-512 units are like the single most important thing about Knights Landing architecture. Why TF do you dare speak about it when you literaly know nothing, kiddy?

Now explain to me like you are going to generate 3+ TFLOPS of FP64 compute throughput on Xeon-D? Not even the highest-corecount big Xeon-E5/E7s go near that level of performance.
(and that is good old CPU code, no convoluted CUDA/openCL that usually gives big numbers in linpack for purpose of TOP500 ranking but performance in real tasks then plummets)

>>56946319
>Why not just run them through a GPU that has better FP64 performance like a normal human being?
Okay, you're sounding like a normal human being there.
>The Rx 480 has over 300 GFLOPS of FP64 performance.
>The Rx 480
Now you are full retard. I am so sorry to hear of your plight, but where the fuck do you work that they only supply you with fucking RX480's to work with?

>>56946325
*explain to me how
(getting really late/morning here, sorry)

>>56946319
>HPC
>gaymur card
anon

>>56946325
>72 slightly modified airmont atom cores running at ~1Ghz can do 3+ TFLOPS of FP64
lol

>>56946362
http://www.nextplatform.com/2016/06/20/intel-knights-landing-yields-big-bang-buck-jump/

The details for people who don't want to make an ass out of themselves.

Good night and stay classy, /g/

>>56946411
>more than 3 teraflops of double precision floating point performance.
Talk is cheap, where are the benchmarks proving this?

File: 1253675389.jpg (94KB, 641x960px) Image search: [Google]

94KB, 641x960px

>>56941040
>avx
>avx everywhere

>>56946424
>I'm gonna ask for benchmarks that are under NDA/not public to shut him up instead of admitting being wrong
>It worked in presidential debate

TFLOPS ratings are always theoretical "meassurement", they don't need to be backed up by numbers, because they are directly derived from number of exectuing units/their per-clock throughput and the frequency at which it runs. Nothing to dispute there.

Now you can then run some actual workload and measure "TFLOPS" in that. But in such case, you can be sure that a CPU like KNL is going to lose less of its theoretical perf. rating than a GPU will.

File: YO DAWG.png (196KB, 321x236px) Image search: [Google]

196KB, 321x236px

>>56946473
Also isn't all this like elementary knowledge. You guys have awful of reading to catch up on, for a "technology" board. Inverse proportion between insight and self-confidence ahoy!

>>56946494
autismspeaks.org

>>56946494
If you're frusterated by this bullshit there's plenty of online communities for better discussion. Smartfags don't stay on 4chan too long :)

>>56942639
>someone more knowledgeable please explain who would want this and not just a GPU.
can't virtualize on a GPU
seems like a serious SPARC killer with all those threads too, though the per-core performance will obviously be shit and it won't have the hardware crypto acceleration gimmick

File: Giant_European_Eagle_Owl_Face__by_Etinogard.jpg (675KB, 1440x900px) Image search: [Google]

675KB, 1440x900px

>>56946522
Yeah, usually when I check /g/, I would leave in under 5 minutes of reading some thread. Usually it's /pol/ and not outright ignorance though.

>>56946537
SPARC from Oracle is not targeted at scientific FP math/HPC workloads, they go for databases. I don't think there is overlap there. KNL is really just for HPC and not much else (perhaps neural network uses like Nvidia is chasing).

SPARC64 from Fujitsu yeah, those are HPC chips, but Fujitsu rolls with their own CPUs. Actually they will replace their bespoke SPARC64s with something built on ARMv8 ISA with SVE (I think). So no space for KNL there.

>>56946473
>>56946494
>TFLOPS ratings are always theoretical "meassurement"
They aren't, they're literally measured using Linpak most of the time
Of course there's a theoretical Rpeak in TFLOP's, but there's dozens of benchmarks to measure TFLOP's all with wildly different results, Linpak being the shitty norm

>>56941040
So, will the new E5-2xxx upgrades be absolute shit for gayman, or what?