Intel Delays Mass Production of 10 nm CPUs to 2019

Intel is unironically sponsored by the elite. Not even memeing. Wouldn't be surprised if they pull a 5nm node out of thin air by 2019.

>intel new uarch drops x86 support
Can't wait to watch the whole world burn.

>intel new uarch drops x86 support
That was never the intention. They're just dropping a few specific parts of it that noone has used for two decades, like x87 FP or BCD instructions (more like 3 decades).

That will just make validation faster, not make the CPU faster.

Also isn't x87 done in microcode nowadays, or is that AMD?

>Also isn't x87 done in microcode nowadays, or is that AMD?
FADD/FMUL are fully pipelined on both Skylake and Zen, so I doubt it. Either way, it still requires 80-bit registers, so it would be hard to push into microcode alone.

>That will just make validation faster, not make the CPU faster.
Of course; I was just saying that they're not dropping x86.

>Either way, it still requires 80-bit registers
Those registers are still needed for MMX

MMX would also most likely be among the dropped extensions. Also, MMX only uses 64 bits of those registers.

Keller isn't working on a new architecture.

If someone told me 2 years ago that AMD would have a superior node for their CPU than Intel I'd spit in his face.
Now we're actually there.

Some people don't wanna admit it but the x86 market is changing, and fast.

>mfw Coffe Lake barely released and new chipset for 1151 already leaked

Attached: 1521328846377.gif (413x243, 51K)

A chipset change a month

>new chipset
They have learnt nothing

is a decentralized microcontroller production economically feasible?

Does anyone have an incentive to go against the botnet?

EUV is some pretty cool shit
ASML is working on high NA upgrades already

intel get your shit together

x87 FP is actually used pretty frequently. Adding more bits of precision is usually the first option people reach for when they discover floating point error accumulation.

Friendly reminder that if you have anything newer than Ivy Bridge you were scammed.

Attached: 1516499433079.png (801x1500, 1.05M)

>no 10nm but how about a new chipset and socket
lel

No, he's doing the same as he did at AMD - organizing and managing a team who themselves will developing a new uarch.
A team in Plano, Texas.

>distant screeching from Haifa, Israel as the employees find out that *Lake is their Prescott/CedarMill/Presler and that *Rapids is their Tejas and Jayhawk.

>i5 2400
Fucken great but I might go for a poozen 2600 soon. Finally feels like an upgrade for reasonable monies.

Attached: test.jpg (4771x859, 659K)

Zeno's chip fab

How do they keep getting away with it?

It's definitely not used "pretty frequently". Even though it has more precision, it's so completely useless performance-wise that you're only using it unless you have no other choice. Which you do, in just using another float for holding the excess mantissa. Forgot what the technique is called, but there is a term for it.

In practice, the only reason x87 is used is because the 32-bit ABI mandates it, and only because that ABI precedes SSE.

I'm putting together a team

>decentralized microcontroller production
Please expand.

market penetration with jewish tricks.
Intel has more than a handful of lawsuits that they lost due to anti-competitive tactics, both in E.U. and the states.
Intel tries to keep the stock holders satisfied, they don't care about products.
Did you see anyone bat an eye about ME, meltdown or spectre? they left half of their cpu without a fix.
Did you see anyone bat an eye about the 3dxpoint failure?
Did you see that with the altera+cpu socket fails?
Did you see anyone care about sudden the RIP of xeon phi?
Nobody fucking cares at intel about anything else but the stockholder gains.
They are literally very happy because they bought some of the most popular reviewers, either via purch media, or by contacting each one individually, so they can keep their mob happy about the incompetent product, and they cut down from various fields in the company to save gibs to the idiots who bought shares and care only for the stock price.
That's what intel has become.
AMD otoh, 1 year of ZEN sales went into paying almost all of their debt. A year that quadrupled their stock price and they didn't give anything back to stockholders, they instead had a tiny amount of loss per share. Why? because they care about the survivability of the company and their R&D.

High performance low-memory requirements stuff should be using x32 ABI anyway.

Well, my rule is that when upgrading the CPU, you should apply Moore's law, so the Ryzen 2600 is a good upgrade actually for what it costs from an i5 2400.
Saved.

Yes, of course. I was mentioning it precisely because using the x86 ABI is for legacy only.

wtf, how can they drop mmx? what about all the software that uses mmx?

brand loyalty (which is, ironically, what kiketel shills accuse AMD buyers of)

There's no software that uses MMX that isn't 20 years old.

Is there still hope Intel bro's

It feels like there is no hope

Attached: 000000045.png (653x726, 84K)

all software uses SSE2+ now

Also they won't 'drop' it - it'll just be emulated going forward.
It'll take a performance hit, but as the other poster said - 20 year old software.

Of course there is, unless you're meaning hope that the Haifa guys are keeping their jobs past 2022.
Hope is in Plano now.

>new 8 core cpus are going to be literally 2 7700k glued together
>intel's year of progress
lmao

>Also they won't 'drop' it - it'll just be emulated going forward.
I doubt that. I'm pretty sure that part of the reason that they're dropping support for many of these useless instructions is just to get a bit of encoding space back, now that many vector instructions are like 8 bytes long and use like 4 prefixes and escape bytes.

>surprise it's CedarMill2
History repeats.

If I remember right the 2400 cost me £240 isn't after launch so the 2600 would already be cheaper. Fuggen ram prices tho.

Proper Tim or delid and that sounds pretty badass

something like Threadripper?

Attached: 000000000009.jpg (1280x878, 293K)

so like a CCX

Attached: 1524750536840.png (882x758, 316K)

well, at least each die is an 8-core

Well, now I feel a lot better about my Luddism

Attached: Capture.png (502x54, 2K)

Well yes, the point perhaps being that TR is a variation on Ryzen, whereas gluing to 7700Ks together being Intel's only progress in a year is a bit underwhelming.

Don't hold onto it too hard, though.

Attached: intel-7700k-blender-benchmark.png (806x416, 24K)

presler was two Cedarmill

Basically the same as Threadripper, only Intel will need two dies just to deliver an octocore instead of a potential 32 core or actual 32 core with EPYC.

Attached: 0000006.jpg (1024x684, 231K)

Haswell was actually a bigger improvement than it's usually given credit for. It saw some pretty cool technical features too, like reg-reg move elimination, and increasing the execution ports to eight.

>Basically the same as Threadripper
Haha yes, except that little detail knows as the FSB.

Does anyone know what the limit of the bingbus seems to be?
When does the performance start to drop off and when does the mesh interconnect overtake it?
You know numbers of cores and whatnot

Attached: 1282749837022.gif (200x150, 1.55M)

History repeating itself. I believe AMD was the first to deliver an actual dual core instead of using the 2 die trick.

Attached: 0000000000004.jpg (439x969, 174K)

>more than half as good as a $400 processor, more than 7 years later
moore's law was a lie

Yes that part sucks, what you save on the AMD CPU you spend on faster RAM.

Many high end processors from 7 years ago are still very competitive on the low to mid end of the spectrum.

Even the FX-8370 overclocked to 4.8/5Ghz or FX-9590 stock are about the same speed as a Ryzen 5 1500 in multi threaded applications and they double as a space heater for colder climates.

>Every AMD cpu and apu are currently implemented as MCMs
like pottery

A reminder that 94C under load is fine & that AMD usually gets hotter than this when under any sort of strain.

Attached: 1524186862809.png (1000x500, 415K)

>tfw 2600k in xeon better silicon form

Attached: Xeon.png (761x561, 54K)

>mfw AMD won't add more cores to their CCXs
>mfw they will just glue two shrunken Zeppelin dies as a 3700x

Attached: the legs are too bright.jpg (743x836, 447K)

kikes

what?
Even with an OC ryzen runs 70c tops
???

Here is my path my moores law cant into
>Q8200 4c stock
>i7 990x 6c12t the only real 7.9 windows score on g
>i7 3960x blew up
>i5 6400 4c OCed 4.4 yes locked OC
>xeon e3 4c8t stock nonintrude

Attached: 990X.png (410x393, 226K)

A chipset change each year keeps the goyim in fear.

>mfw AMD won't add more cores to their CCXs
I don't think they're going to add more cores to the CCXs at all. I strongly suspect that the reason for the CCX to exist and the reason there are four cores per CCX is because they're using some local interconnect that intrinsically doesn't scale, like a crossbar (that doesn't scale in cost) or a shared bus (that doesn't scale in performance), and using the IF only to interconnect beyond its limits.

If they're adding more cores, I strongly suspect they'll be adding more CCXs instead.

>If they're adding more cores, I strongly suspect they'll be adding more CCXs instead.
At which point it might not even matter if they're adding those CCXs on the same or on different dice.

Intel has a lot of respect from old boomers with money. Amd is 10 dollars a share shit and investors dont like fucking with it because it could go back to 3 dollars a share any time

To be fair they still have the server market locked up.

Why isn't there something like a Threadripper with ECC sold for server use?

I think we will see a 6 core CCX. Maybe on 7nm+ but maybe on first 7nm.
I dont think 7nm is going to be a die shrink like 12nm was.
This would allow 12C 24T consumer level chips
Not sure how the memory configuration would work or if dual channel would provide enough bandwidth but I dont think we will see it but it might take a new socket and on DDR5 in 2020~

>If they're adding more cores, I strongly suspect they'll be adding more CCXs instead.
That makes sense and checks over the rumored 48 core EPYC. I thought about 6 core CCXs, but adding another 4 core CCX would be the same thing.

There are a ton of chips just like TR on EPYC
TR is just a EPYC chip with two dies not connected
They have 4 16 core chips starting at $650
They performance very well compared to intels current xeon line. Not just being 50% cheaper but overall performance is better or on par at the same core counts.

en.wikipedia.org/wiki/Epyc

>I think we will see a 6 core CCX.
As I said, though, I think the reason we aren't already seeing those is because the local interconnect on the CCX intrinsically doesn't scale beyond 4 cores, not for electrical but for architectural reasons. The reason I think this is because 4 cores is a very common architectural grouping across much of the industry, from 4-core ARM chips, to Intel's Silvermont cores, the Jaguar/Puma cores, &c&c. Some of these do come in more than 4-core variants, but when they do, it's almost invariably again in some sort of CCX-like configuration, with a more complex interconnect between the 4-core groups. I think that 4-core architectures are so common precisely because there is some simpler form of local interconnect that can only be used for up to 4 cores.

>This would allow 12C 24T consumer level chips
You could just as well do this with 3 CCXs on the chip instead.

And the laptop market

>BCD instructions
>mfw I have actually seen them being used

>64 threads
We need to set our dev sever up on one of these at work

Fix your screencap. Include the part of the thread in which the tripfag said not using thermal paste is the way to go.

Its actually 128 threads
Those are 2P chips so two CPUs per board
1P chips are 1 processor per board
Check out how many PCI-E lanes it has

I'm curious if this was on literal 40-year-old programs, or if anyone used them in newer programs.

There's absolutely no reason to do that.
The 5.000 transistors that do those things won't help at all on a billion transistor thing.

10 /12 cores.

Does it really matter if they delay a new processor a few months when shit from 5 years ago is still beating current Ryzen?

At this point I just laugh at your existence.

Yes, because this same shit from 5 years ago still beat most of the intel CPUs as well.
And if AMD don't beat intel and hard, 10 years from now this same CPU will still beat everything.

You're so hung up on fanboy memes you don't want faster/better CPU's?

Is your only goal for CPU improvements to beat a competitor? Super sad, user. I want faster CPUs regardless.

Do you honestly believe that AMD competing is the sole driving force for CPU innovation?

That's a fucking joke, AMD are irrelevant in the mobile sector and the only thing pushing intel to improve is ARM companies like qualcomm pushing into the laptop market.

Incredibly sad that you imagine i said anything about AMD or competition in general.

I said
>You're so hung up on fanboy memes you don't want faster/better CPU's?
Implying that better Intel CPUs now is better than better Intel CPUs in 5 months.

Stay brainlet user.

I didn't imagine anything, I assumed it.

Attached: autism-asshole-disease.jpg (600x750, 45K)

>Sponsored by a bunch of autistic Goldeneye speedrunners
Sure thing buddy

>first to prioritize IPC over pushing clockspeed
AMD

>first to build a real x86 dual core CPU
AMD

>building x64 extensions for x86
AMD

>pioneering the replacement of the FSB with high speed busses like Hyper Transport and DMI
AMD

Historically speaking, AMD's competition has been the impetus for most of the good stuff that Intel has done. Intel may do things better, but they wouldn't do it in the first place if AMD wasn't around and we have nearly three decades of history that shows that. I'd tell you to quit being a retard, but history also indicates that you may have irreparable brain damage.

Not really.
Qualcomm is Intel's biggest threat, as admitted by their own CEO.

The only reason AMD is trying so hard at the moment with their fuckhuge dies is because if they shit the bed there won't be an antitrust case to bail them out.

Attached: tripshit.jpg (1293x1730, 1.16M)