INTEL Xeons literally BTFO

INTEL Xeons literally BTFO.

Other urls found in this thread:

ark.intel.com/products/92986/Intel-Xeon-Processor-E5-2620-v4-20M-Cache-2_10-GHz
spec.org/benchmarks.html
www3.intel.com/content/dam/www/public/us/en/documents/specification-updates/desktop-6th-gen-core-family-spec-update.pdf
twitter.com/SFWRedditImages

PLS DELET. PLSSS

woah that's a EBYN cpu :DDDDDD

Source.

Incoming shoah.

Epyc specs.

BWAHAHAHAHHAHAHAHAHAHAHAHAHAHAHHAHAAHAHAHAHAHAH

HOLY SHIT
NO POINT EVEN COMPARING 2 SOCKET SOLUTIONS FROM AMD, ELSE YOU'LL JUST GET 128 CORES OF FUCKED UP SHIT NIGGAS

AMD

The NDA on this material ends on June 20th. Here are some buzzwords from the slides:

up to 32 High-Performance “Zen” Cores
8 DDR4 Channels per CPU
Up to 2TB Memory per CPU
128 PCIe Lanes
Dedicated Security Subsystem
Integrated Chipset
Socket-Compatible with Next Gen EPYC Processors

No compromise 1-Socket

Right-size underutilized servers
Optimize storage and heterogeneous compute
Low power consumption
Up to 20% lower cap-ex

The EPYC 7000 series processors feature 128 PCIe lanes and 8-channel DDR4 support (up to 2666 MHz). Some parts are listed with two TDP values (right now I’m not sure why). EPYC CPUs are available with up 32 cores. The cheapest part should be available under 400 USD and the most powerful EPYC 7601 processor will be sold for around 4000 USD. The highest clock speed is 3.2 GHz in turbo mode.

>$400 8C/16t comes with the 8 channel memory and 128 PCI lanes
Holy Fuck.

I wonder what the prices will be... the price bracket on the left is for the Intel chips.
But we can largely assume that if Ryzen is about ~350, and TR is ~850, then doubling up again the 32-core chips should be about $1800 or so... less than half of Intel's top-end Xeons for 50% more performance.
Intel is pretty much done in the server space, at least for a while.

I hate and for being shit, but I'm actually looking forward to this.

How does this part make any sense? What are 8 cores going to do with all this memory and PCIe bandwidth?

Might there be standard ATX boards designed around this?

IO server?

...

C E R T I F I E D S H I T W R E C K E R

J E W K I L L E R

CONTRACT KELLER

KISS ME, KELLER!

And people called me crazy for wanting an 8core threadripper with 4memeory channels and 64pcie lanes.

8core eypc looks quite cheap. Can probably saze a lot of money on a motherboard for it by only wiring one memory channel to each die.

>tops out at 180W

Sad, I was expecting 205W like with Intel.
Though the TDP on that 8 and 16 core is hilarious, especially compared to Ryzen and Threadripper.

DELID DIS

Cancel that. The 8core is only for 2p systems. Doesn't make sense when 1s is going to have higher IO density, the only gain it will have it the double ram size when it wastes all the io lanes for intersocket.

Synthetic benchmarks! Not legit! DELET

>those scores
>those low base clocks

Is AMD actually using turbo for its all turbo clock on server parts? Makes a lot more sense, else a EPYC wouldn't demolish Intel like this.

Being that these are multiple zeplin dies, with a lot more PCIe connectivity and memory controllers, the increased TDP is sort of warranted.

THE SECOND HOLOCAUST WAS AT THE HANDS OF A SINGLE HUMAN BEING

>64 threads for 180W

AMD is finally back.

I could imagine the top chip to boost up to 2.8-3Ghz as it has been speculated before.

Look at the >400USD comparison EPYC 7251 vs e5-2560 v4
Epyc: 8 cores, 2.1/2.9
Intel: 8 cores, 2.10/3.0
ark.intel.com/products/92986/Intel-Xeon-Processor-E5-2620-v4-20M-Cache-2_10-GHz

Yet it's 23% faster, all core turbo of the e5 is 2.3GHz, exactly what 2.9EPYC part would do.
If Turbo clock = all core turbo for these parts, this would make sense.

This is not hapening! My 7700k looks so shite after 6 months!

There's no reason you won't be able to use a chip intended for 2P systems in a single socket config, just like you can use any Xeon in any socket it fits, even X99.

Wow, I thought it would be 2.9/2.8GHz at 200W, this is even better.

AMD has outdone themselves.

>SPECint
>synthetic
Lmao, SPEC is around 85% of what datacenter buyers look at.


spec.org/benchmarks.html

What if israels intel plant shuts down because of this?

OY VEYYY

Jew Keller

>8 channel memory and 128 lanes from EIGHT TO THIRTY TWO CORES

Intel just got demolished, their entire lineup, bronze, silver, gold, poo, only the Platinum lineup has any chance of competing.

AMD has done it

>Platinum lineup has any chance of competing.
With 20k USD cpus? HAHAHAHA

I think he meant Pentium not Platinum)))))))))

Hence the "chance"

Some don't care about price at all, RAM costs more than CPUs
So only the Platinum lineup can offer any competition, rest are practically pointless.

True I'd looked over that. But there might be some configuration thing with the inter socket IO using PCIE.

I know there's more drag from the extra server junk. But zen 8cores at 120W could probably be pushing 3.5ghz+ like the desktop parts. At the least it's probably slammed at 2.9ghz and maybe it's added xfr?

Wow, and I thought base clock was misleading, am even more confused now.

I wonder why vendors even use base clock anymore, when these parts will have an all-core turbo so far above it base clock might as well not exist?

Just sandbagging? "look our chip is so low clocked and its performance is amazing!" kind of deal?

...

E5-2699A v4, 22 cores - 2.4/3.3 GHz - all core turbo 3.0GHz
EPYC 7601, 32 cores - 2.2/3.2 GHz

Yup, this pretty much proves that stated EPYC turbo = all core turbo in regular x86.


Fucking hell AMD

jim "the contract" keller
motherfucker has yet to disappoint.

To TDP's probably means one of them is stock in one of them is with an XFR like setting. Or you looking at stock with minimal PCI E lanes in stock with all the PCA lanes saturated

MEGAHERTZ MERCENARY

There is a market segment that Intel's not tapping, there's a very limited number of people who really need a really powerful CPU but there's a fuck load people who need a lot of I/O, and AMD's giving them what they want a single socket server that can saturate their needs. Intel only does this if you buy a two socket server which cost of fuck load if you only to put one CPU in it.

AMD's hitting everybody's market segmentation the same, not fucking any of them like Intel does. If you need a CPU and I/O AMD's got a CPU with the I/O you need, unlike Intel who makes you settle for something you don't need or settle for something you don't want.

Base clock is probably for AVX* or with fully loaded PCIe lanes and memory controllers.

if you're rich as fuck already and want something that scales beyond two sockets, sure

but for every one of those they'd sell to some hft operation, amd's probably selling 20 epycs to everyone else

>tfw can now count jews 161% faster

>AMD's hitting everybody's market segmentation the same, not fucking any of them like Intel doe

um excuse me but where is ayymd's answer to the all important pentium market

>that EPYC 7281
Wait what? 16 cores for $600-800 WAT

Given the PCI E lanes in the memory, the CPU might be relatively cheap at that motherboards going to be expensive.

I wonder what the performance per watt is like, since that's more important to server hosters.

I wonder what single core turbo is.

AMD wanted to get out Ryzen Threadripper and epic as fast as possible, there Pentium equivalent or their around lineup isn't going to hit until the back to school market comes up

That way doesn't give Intel a chance of lowering the prices of their CPUs to be compelling. Or at least not lowering their prices of CPUs fast enough. We've seen it in the OEM market before where Intel will drastically cut their prices to stave off competition if they think they're going to have to whether something bad.

At least that's the way I see it and what they seem to be doing given everything they said.

They'll be four four core CPUs, these are throw away dies at best that they're able use for segmentation purposes. Same with the eight core version.

To follow on from what you're saying; Intel literally can't cut prices on their high-many core xeons much at all since each one costs many more times to make than a threadripper or Epyc.
I think their strategy is going to be robustness campaigns. They're going to spend a few million dollars on an engineering team to QA the fuck out of some AMD chips looking for execution bugs and such, and then they secretly forward this information to journalists to make AMD processors look 'unreliable', so they can maintain marketshare with the 'premium' branding.
As well as the usual bribing OEMs to not buy AMD and the usual dirty israeli tricks.

...

Every single die Intel makes comes from a silicon that costs $6000-$15,000 to produce bare minimum, onset die they get may be 40 to 100 shots at a perfect die, they can easily segment the parts so they cover the cost of making them, they could severely reduce their profit margin and come up squeaky clean, but heads are gonna roll if that fucking happens.

There likely whether five bad years of CPU sales just so they can keep their margins. This isn't the first time Intel's done it.

Long story short they could compete price wise but they be taking a very very big hit long-term if they did that, especially if there next CPU architecture actually is good, as it'll take 4 to 5 generations for them the even come close to bringing back those margins.

FUCK NO IT ISNT SUPPOSED TO BE THIS WAY

WHY IS THIS GUY STILL ALIVE HE ISNT SUPPOSED TO BE ALIVE STILL

STOOOOOOOOOP

>They're going to spend a few million dollars on an engineering team to QA the fuck out of some AMD chips looking for execution bugs and such, and then they secretly forward this information to journalists to make AMD processors look 'unreliable',
So, what you're saying is. Intel would rather invest millions into teams QA'ing AMD processors, meanwhile they do very little QA on their own?

Terrifyingly enough, this sounds probable.
But at the same time. all AMD has to do is publish Intel's Errata sheets, which grow longer and longer with every CPU launch.

I don't get how a 2s server with only one cpu in it helps anyone? People say it's for the extra pcie lanes, but those lanes are fed from the second cpu which isn't present. Is it just 1s not having enough motherboard designs or something.

"You can have quality Intel Xeon, or you can have... did this fall off the back of a truck? I'm not sure you can trust this garbage. There's gotta be something wrong with this if they can make something that much cheaper than a real company can"

Nah they'll keep their own QA up to the usual high standards. (Their standards are actually pretty high when it comes to QA). It's about the smear campaign.

because you don't need each pcie device to be going balls out on full pcie 16x to be useful, look at most pcie ssds, they are 4x or 8x, i'm unsure if there is even one 16x, and if that's the case, it may be better to raid the pcie device at a lower pcie multiplier.

granted when I say this, i'm talking about a far lower end use for most applications.

Didn't Intel fire most the quality assurance team? Isn't that why the Atom processors they shipped off to China are all fucking failing hard?

Also keep in mind back in AMD's heyday they were the server CPU, there's no amount of it Intel's jewerry that will get people use their inferior products unless contractually obligated.

38 pages of errata m8
38 pages
www3.intel.com/content/dam/www/public/us/en/documents/specification-updates/desktop-6th-gen-core-family-spec-update.pdf

Intel binned a fair number of their experienced QA team members.

QA on atoms is a different sport than QA on server parts. You piss off your business customers, they ain't comin back. You ship some garbage atoms and no-one cares.

J I M K E L L E R

E

W

K

I

L

L

E

R

They come back if you are the only game in town, and till epyc they sure as fuck were just that.

Nah in server land they're competing with ARM.

not an adequate replacement for most applications.

Just don't overclock them 7700K's goy

I'm more thinking that one socket is going to supply 40lanes for whatever and the chipset with sata/usb. But if there's a 2s system, if there's a second socket with nothing in it, the pcie lanes connected to it just aren't going to be doing anything at all. And it's come up a few times recently where apprently 20% of all 2s systems are running 1cpu

You can't buy Pooripper because AYYMD sold all of them to Alienware and you have to buy a very overpriced Alienware shit to get it

TOP KEK, AYYMDPOORFAGS

FTC is watching, good luck.

A smear campaign is a different ordeal to Intel literally blackmailing and bribing OEMs to not use AMD products.

Huh?
>trusting trump's agencies
Those places were super corrupt before trump.

Fuck off, retard. Alienware has OEM exclusively, not part builders.

Ok, seriously now.
What's the point of paying more for slower and worse Intel products? Anyone?

This failing c2000 atoms are in loads of popular SMB file servers now. I'm suprised this hasn't become a bigger thing. The offered fix is horseshit, installing a resistor to the motherboards but that only delays the fault onset to ~5years instead of a guaranteed 18months as it is now. What happens if a mobo is refurbed and sold as a fixed version but it still close to failure as it was running at the damaging voltage for over a year or something.

Kind of null and void offering a replacement on a CPU that's currently failed in orbit.

Intel is better goy! Buy Intel Xeon GOLD and Platinum processors.

Don't think this FUD would work if AMD beats Intel in every category.

Seems to be what the OEMs are doing though. No one is rushing out the c3000 or anything else to stop this. I predict some major shit storm when the systems all start to fail.

>Intel literally can't cut prices on their high-many core xeons much at all since each one costs many more times to make than a threadripper or Epyc.
sounds like Intel gotta git gud

>just like you can use any Xeon in any socket it fits, even X99.
I'm pretty sure you can't, anymore. At least Skylake Xeon e3s are supposedly not compatible with consumer chipsets.

JIM "KABY KELLER"

skylake e3s required a different chipset but you could use e5 v3/v4 xeons in x99 (and earlier generations in x79, like those e5-2670s everyone was going nuts over a year ago)

i thought the newer skylake xeons used a different socket to the hedt stuff though (because 6 channel ram or whatever)

>xeon platinum
>still less memory channels while costing a lot
>competing

Skylake e5s and e7s aren't even out yet. Skylake(and kaby lake) e3s share the 1151 socket with the consumer desktop chips, but require the more expensive workstation/server chipsets. And they still only use dual-channel RAM, too.

None? Why do you think they're collectively shitting themselves?

T
HE CANT KEEP GETTING AWAY WITH IT
E

F
UNSTOPPABLE
C
K
I
N
G

MAD
A
N

Wair for Coffeelake
Wait for Cannonlake
Wait for Intel's completely new arch in 2021

I did say older E5s (E7s too? but who the fuck has those)

I thought the Skylake stuff had been sold to a few of the major buyers already but it wasn't coming out for the rest of us for a while yet

Originally you said ANY Xeon, user.

Also skylake-x x299 is pretty much the x270 chip on their hedt boards as the increased I/O from the cou itself. Xeons are less likely to work again as it's desktop/consumer chipset again.

In that vein isn't x399 the same as x370 in terms of io?

don't trigger my autism

i called out the haswell and broadwell generations specifically (v3 and v4 in jew numbering)

>"Intel's most Powerfull HEDT CPU - 7740X"