18 core intel CPUs incoming

videocardz.com/69900/exclusive-intel-to-launch-18-core-core-i9-7980xe-cpu
HAHAHA
YOU CAN'T MAKE THIS SHIT UP
>analysts ask intel if ryzen will have any impact
>n-no we're n-not worried
>AMD announces 16 core HEDT CPUs
>worrying intensifies
>MOAR COARS

Other urls found in this thread:

segmentnext.com/2017/05/22/amd-threadripper-benchmarks-score/
pcgameshardware.de/Skylake-Codename-259478/News/Bis-zu-18-Kerne-bestaetigt-und-nicht-verloetet-1228994/
twitter.com/AnonBabble

Oh yes, Intel went nuts with housefires.

>18 cores
>Not a power of 2
This bothers me greatly.

ah yeh, I'm sure it'll be extremely desirable, what with their 16 cores already pushing on 200w TDP

It's a multiple of 2. Be glad it's not a prime number.

and it will prob cost 4 times as much as the threadripper

You do realize they have tons of chips which arent powers of 2? I have dual 10 cores.

Intel sure got caught with their pants down. It's Pentium-D all over again.

SHUT THE FUCK UP

lol exactly what I was thinking

BIBELINE
BINGBUS

GPUs must trigger you immensely then.

WHAT IS THE END OF THIS MOAR COREZZ RACE? 64 CORE DESKTOP CPUS?

You know what happened when intel decided to step their game. Finally amd has pushed them enough for us to expect a new core 2.

>what is the end of this memory race? 2GB hard drives?

Kek, why do you think AMD uses the same socket for threadripper as Epyc? They can shit out a Epyc anytime, either a 16 core with 8 channel memory, 24 core or 32 core.

It's 2021 at best and it sounds like another Itanic.

>next year AMD releases 48 core CPUs
>intel releases 64 cores
>AMD releases 72 cores
>intel releases 96 cores
>AMD releases 128 cores
>intel releases 144 cores
>etc...
the ride never ends

Yes, but it's modified socket without the ability to use xGMI links, thus no 2S mobos.
Intel can't keep us with the COARZ race due to lack of IF.

Skylake-X isn't multisocket either so no problem.
If worse comes to worse, AMD should have no problem releasing SP3 to consumers.

BUT WHY NOT JUST GET A SERVER?

And Intel won't be able to compete with consumer-priced EPYC because it's 4 Zeppelin dies lmao.

At what? 2800MHz? Literally uncompetitive against 3.6/3.5 threadripper.
And AMD still has the option of going even higher at 205W

fire departments going on strike while basements everywhere burn

>At what? 2800MHz?
No, higher, but prepare those LN2 pots.

>we're not worried about poozen and their core count, goys!
>releases 7 CPU's with moar coars

This is hilarious.

Considering their 12 core is at 3.2GHz base, I'm not expecting higher at all.

And they still won't be able to compete against Threadrippers in price because they have no IF to MCM dies lmao.

This is exponentially more easy for amd since their cpu's are basically just smaller cpu's glued together, the scallability is huge

So it will come at $2500 and perform really close to the 16c/32t threadripper?

In theory amd can release a 32core with only 4 memory channels on the consumer boards without much hassle. Do not want to see the costs for something like that though. Just like I don't want to see intel's pricing.

>and perform really close to the 16c/32t threadripper
No, clocks would be much lower.
>In theory amd can release a 32core with only 4 memory channels
How and why?

This is a losing race for Intel, both performance and price, AMD would have no problem releasing a 32 core assuming they're not all backordered by companies.

well, they'd have 2 more cores to compensate, and they're really not caring about tdp this time

>they're not all backordered by companies
They probably are, someone from 7+1 (probably microdicks) are ordering fuckton of EPYCs since January.

adding to that, some x299 boards photos leaked, and they have 2 cpu power connectors, how much wattage can you drop through that?

God fucking dammit ebin bintel housefires :-------DDDDDD
are fuking back. Best timeline.

>how and why?
It's the same socket so naples is already the right size and shape for it. My guess is the sp3r2 consumer socket would be close to the sp3 server socket with only 64io and 4memory channels per socket. This is useless tier halo product theory crafting though,.

It would be still 4 Zeppelin dies in a single package, thus they have 8 memory channels.
I really want to see how madmen from ASRock would make an SP3r2 ITX mobo.

but naples has some differences with threadripper, the most notable one is that it doesn't use a chipset

I don't think they'd be easily compatible

>It would be still 4 Zeppelin dies in a single package
I get that, but the consumer x399 boards are not going to have access to all the memory or IO, but the dies can probably get power and the gmi interconnect and work fine.

Did amd actually do it ? Did they secure the timeline ?

Yes, they have no competition till 2021 and i doubt they'll have it after because for current AMD management.

>AMD released EPYC for consumers at 205W, the socket is already there
>32 cores at 3.0GHz

Literally another shoah

Intel really, really doesn't wanna lose the halo product crown.

But they fucked themselves over this time, it's still 2011LGA, meaning they can't go over 18 cores on it, meanwhile AMD thought ahead and used the 4094 pin socket that their server chips use.

I read somewhere that it would be 2.5GHz @180W, but that's still way better than the 1.4GHz 28c/56t @205W

It's LGA2066 though.

Intel's next move is to 10nm, which even they say, is currently slower than their 14nm nodes, and would be for some time. Not to mention
>1.7% yields

Whereas AMD's next move is to IBM's 7nm node (IBM's tech, Samsung's fabs)
Even then, corearch wise, there's very little Intel can do, hell, there's never been anything Intel can do, even their current arch is simply just a refresh of what IBM designed for them 12 years ago.

Minor difference, 50 pins won't help them scale, their Skylake server is around 3600 pins.

>1.4GHz 28c/56t @205W
It's 2.5 though.
>Samsung's fabs
GloFo will fab 7nm HP though.

>GloFo will fab 7nm HP though.
Ah yes, quite right.
Got caught with AMD's current fab deals with GloFlo + Samsung

kek, not even the 220W fx 9590 motherboards needed 2 cpu connectors

what the fuck is intel doing?

Why is amd bullying intel

>what the fuck is intel doing?
We're going back to the old times, heck, 18core SKL-X is literally P4EE 2.0.

meanwhile at amd

segmentnext.com/2017/05/22/amd-threadripper-benchmarks-score/

This was a fake, and low clocked too.
It's likely going to end up with threadripper having 3.5/4.0 frequency

...

AMD likely wanted to hold out a 24+ core consumer EPYC for next year's refresh, guess that's not happening.

>Yes, but it's modified socket without the ability to use xGMI links, thus no 2S mobos.
Please elaborate.

What's "IF" user ?

xGMI is the protocol AMD uses for socket to socket communication.

SP3r2(threadripper socket) doesn't have that ability.
While SP3(naples/EPYC socket) does have it, but physically both sockets are the same.

Also as a footnote, SP3r2 is a 180W+ socket while SP3 is a 200W+ socket

infinity fabric?

well, the actual performance in cinebench can be inferred with simple multiplication, seeing that the architecture is the same, it will get similar performance at similar clockspeeds

so if single threaded performance on a stock clocked 1800x (3.6GHz all cores) is 156cb and multithreaded is 1618, it means that with SMT enabled it performs like ~10 "raw" cores

1618/156 = ~10.37

so 16c/32t at would perform as ~20 "raw" cores, given the same clocks, we can calculate the following:

156*20 = ~3120cb

but we also need to take into account that IF scaling is not absolutely perfect, so that score drops to ~3000cb for multithreaded

Reminder that for 6 channel memory you need 1596pins, so you're not getting that on LGA2066

Infinity Fabric, a literal marvel of hardware engineering.

p4EE managed to copmete with the best athlon while pic related is pure embarrassment

everyone already knows that Skylake-X is quad channel user

Thanks. I guess this way of manufacturing allows cores without compromises.

The thing is you don't know how IF scales.
Secondly, you don't know how much of a effect 100GB/s infinity fabric has.
Thirdly, you don't know how quad channel memory will affect this.

>p4EE managed to copmete with the best athlon

Yeah, but Skylake-EP is 6 channel, and Intel cannot go higher on this socket.

Intel cannot go higher because they would need fucking 4th bingbus for 8 memory channels.

so the calculations I did can give at least the worst case scenario for the 16c/32t @3.6GHz, and even then it performs better than intel's 24c/48t available today

>pcgameshardware.de/Skylake-Codename-259478/News/Bis-zu-18-Kerne-bestaetigt-und-nicht-verloetet-1228994/
> Zudem wurde uns bestätigt, dass die CPUs entgegen der bisher gängigen Praxis bei der HEDT-Plattform nicht mehr verlötet sein sollen.
> Translation: Skylake-X will not be soldered
JUST

As someone who's been interested in computers since the IBM XT with it's 8088 was cool... I sure hope so and it's' about time too.

The 80286 replaced the 8088 and was cool and new, the 386 was really good compared to the 286 and the 486 was a bit cool and new too. The Pentium was faster but it was a bit of a meh and then AMD came with the Athlon 64 which was really cool. The race to 5 GHz was on at that point but we never got there. Things just stalled out at 4 GHz so they started offering dual-cores. And then quad cores.

The Athlon II had a quad core variant and ran up to 3.5 GHz. This chip was released in 2009.

I realize there's been some improvements in IPC since 2009 but still: the last decade has been a total yawn. A total and complete boring pile of nothing. There's been actual improvements and innovations in the mobile space but PC CPUs have been a huge failure in progress.

We should have had 4GHz 64 core CPUs years ago.

Thank you AMD for the fact that we got some progress with Ryzen and that this forces Intel to do more than just offer the same shit in a new wrapping with a meh 0.5% improvement.

While being expensive and hot, it did on air. 9590 require WC block and shits the bed vs 4.5Ghz sandy on air.

I mean they can't go from 4 to 6 channels on this socket, I'll leave that synchronization embarrassment Intel calls a ring interconnect for another time.

HAHAHAHAHAHAHA

They don't use bingbus anymore in server Skylake

>Skylake-X will not be soldered

Are Intel out of their minds? This has got to be fake

Why would Intel feel threatened by AMD when their old 4 cores knocked the wind out of AMD's 8 cores?

The state of AMD is so bad that Intel could just relaunch Sandy or Ivy Bridge processors with a mild overclock and sell them at $100 a pop and still be the winner

KNL mesh interconnect is mini-bingbuses. Nothing changed.

>Skylake-X will not be soldered
fermi in CPU form incoming

Care to show us?

You ready for threadripper?
Look at the size of that fucking base.

...

Sadly your average ruby/node hipster dev has no idea how to multi thread properly anyway, so for home users this means cheaper CPUs at best :/

Fuck 7nm we do live in the fucking future and I love it.
We should all honesty just use 2cores that's all we need.

...

I can attach this to my forehead when I get a fever?

So it's KNL mesh? Lame.

nice 1% yields

>You know what happened when intel decided to step their game. Finally amd has pushed them enough for us to expect a new core 2.
Not going to happen, intel still had a near monopoly on high end fabs back then. Now they're a bloated company and competitors spread fab costs across high volume ARM chips. Zen + ARM servers are going to hurt intel, they won't die but they're never going to be the same again.

Probably for compatibility with *Lake accelerators.

Wonder how it'll perform, can't possibly be any worse than the ring bus

These are mini-ringbuses. At least they got rid of switches, lmao.

Reminder that Skylake-X was originally supposed to only be 10 cores.

Thank you AMD
But too bad for Intel one Zeppelin die costs like $40 to manufacture, they can't even begin to compete.

Sauce in case anyone saw that thread and went "I want to see boobs jiggle"

Something like 2169 cores

Thank you mr. Huang.