JUST SEGMENT MY SHIT UP
>>61333757
its called fragmentation you stupid leaf
>>61333757
I need that 7401P
NEED
>>61334077
y tho
>>61334101
Because why not
>>61333757
What the fuck is wrong with Intel?
What's the point of like half of these SKUs?
>>61334121
#bettered by threadripper
>>61334126
8 memory channels tho.
>>61333757
glad to see intel does the retarded $10 price difference for 300mhz turbo on a 35w higher tdp chip in the >$3k price range too, I thought they were only cucking their consumer market there for a second
>>61334123
They have better gaming performance than ayyyyyymd
>>61334155
Is this official?
>>61334155
How many fps in CSGO can I expect from a 28 core xeon chip?
>>61334155
they're a fucking monopoly
>>61334155
Lmao, this is /v/ tier shitposting, just scare tactics, they spend more time talking about EPYC than their product.
Ans their stock reflects this.
>>61334173
Soon to be crushed.
No monopoly lasts forever.
Why didn't Intel ever stop to reconsider their retarded plans once they realized Zen is actually good?
>>61334166
yes
>>61334317
Man I ask local Intel shills?
WHAT THE FUCK IS THAT?
INTEL IS LITERALLY SHITPOSTING.
IN OFFICIAL SKYLAKE-EP SLIDES.
WHO THE FUCK APPROVED THAT?
>>61334317
If Intel had a gpu worth shit they wouldn't even think about avx
>>61334317
>login.php?session_id=
SOMEONE ALREADY CALLED ON THE BULLSHIT INTEL SAID
https://www.techpowerup.com/235092/intel-says-amd-epyc-processors-glued-together-in-official-slide-deck
>On SMT implementation between AMD's SMT and Intel's HT, Intel is basically comparing a $2,200 8-core Xeon to AMD's usually $499 Ryzen 7 1800X
>And it's certainly a coincidence that for Intel to achieve these SMT implementation scaling numbers, which paints them in good light, they had to down-clock the Ryzen 1800X to 2.2 GHz
>>61334895
>$499
Isn't it $469 now?
>>61334895
>Here, Intel is comparing their server-grade processors with AMD's Ryzen, desktop processors gaming woes, which really, is one of the best examples of comparing apples to oranges that I've seen in a long time.
>So AMD's server platform will require optimizations as well because Ryzen did, for incomparably different workloads?
>Putting things in the same perspective, is Intel saying that their Xeon ecosystem sees gaming-specific optimizations?
>>61334914
Intel creates new market, datacenter gaming market.
>>61334944
Maybe Intel plans on resurrecting Onlive
>>61334914
>Here, Intel are telling us how much better for the customer it is to be hard-locked to Intel's ecosystem for virtualization, since "VMs running on Intel Xeon processor compute pools can only live migrate to other Intel VM Pools"
>It's like they're saying "just imagine the amount of work you'll have to migrate these to AMD. Better remain with us."
>>61333757
A literal decoder ring to know what CPU you have.
>>61334995
Intel will sell it to you.
8156 is my favourite.
>4 cores
>3.6GHz
>105W TDP
>$7007
You can't make this shit up
Meanwhile in AMD land, $7000 gets you 64 cores at 2.6GHz.
>>61334978
>>61334895
>>61334914
Lisa Su and AMD will have a great laugh at this desperate attempt from Intel
>>61335012
This is for L3 optimised workloads but they gutted L3 so it makes zero sense.
I don't understand Intel.
>>61334171
>CSGO
maybe like 120
DELET
>>61335160
An actual mesh of Ring Buses!
>>61335160
looks like JUSTnich is the new Todd Howard
>>61336131
Nothing? Because a $1000 server CPU with all the bells and whistles isn't very expensive
>>61336167
makes me think about how much threadripper will cost
>>61335190
You're not funny.
>>61333757
>>61334069
Someone needs to defragment that product line.
>8156
>4 cores
>$7007
I checked ark, this is real
>>61337953
But why
>>61334182
Exactly.
When will /g/ understand this.
FYI I'm a 7700k owner.
http://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade
Yeah but oh man, wouldn't you just love to get one of those EPYCs, baby? Ooooh!
>>61334155
Please tell me this is real.
> Our Xeons are better because
> Look at all these companies using our cloud collaboration service
> Because of this your software is guaranteed to somehow work better on Xeon
> It's easier to deploy applications because "Robust Ecosystem reasons"
> Now look at AMD
> Look at all this RED! It means danger! DANGER!
> Look at this questionmark. Don't you feel insecure and scared already!
> Software is not "optimized" for 4 die / socket even though that shit is abstracted away on a hardware/kernel level!
> You will have to rewrite all your software because their x86 is somehow different from the rest of the worlds, trust us!
> Don't believe us? Here have this link to a consumer gaming site proving that AMD worse because "gaming performance"
Holy shit Intel. First that press-release and now this. What ARE you doing.
>>61338278
it gets better
>>61338307
>WCCFTECH
>MOTHERFUCKING
>WCCFTECH
>>61338307
> Socket change is innovation
> Not having to buy a new processor every year means you'll stagnate!
> Look at this trusted source for proof
> * Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors
Holy fucking SHIT what am I reading
>>61338431
damage control
>>61338307
>software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors.
JUST
>>61333757
>that 7281
>16 cores for $650
What a massacre.
>>61338307
>>61338307
>What is Rome
>What is Milan
>implying at least 10-15% IPC uplift plus frequency uplift from 7LP wouldn't destroy Intel on their god awful 10nm
Some idiot goin to buy $7000 4core overclock and game on it
>>61334173
This was supposed to be the beginning of their total monopoly. They clearly had Skylake SP planned for a while, and had no time to backpedal from it once Epyc was announced. So instead of tightening their grip on the x86 server market, they end up lookinh retarded.
>>61333757
Both Intel and AMD CPU's have segmentation, you fucking retard. They're both x86.
>>61333757
guys come on this argument is pointless
If your CPU doesn't put a hole in your wallet and performs good enough then
>why do you bother
>>61338794
>LinusTechTips
>>61338010
Because you need server quality
>>61339276
>What to buy when building 10k$ gaming rig? *Click here*
>Well, 7k$ Xeon is just the most robust choice when compared to glued together processors. Babies can glue together their fingers. PLZ BUY
>Intel fuckboys think they can compete
>Intel delayed 10nm until 2018
>AMD releasing 7nm in 2018
>AMD taping out Zen2 in November
Where were you when Intel became the poorfags?
>>61338010
this is what their binning looks like
>>61339474
>mfw Intel shill can't respond to this post
>>61339236
>50+ skus
>12 skus
fuck off, you can't even install the full 1.5TB of RAM on all those xeons, some are limited to half that amount, intel is imposing artificial limitations so they can charge more for specific features
>>61338743
Try 40%. It's a 50% die shrink.
AMD will be on 7nm while Intel is dicking around on 10nm in 2018/2019.
The jew is kill.
>>61339522
Proof that CPUwarfags don't know shit about actual technology. They just sit there and argue their favorite brand like the fanboy cucks that they are.
>>61339522
>1.5TB of RAM
What use cases have this much RAM? Genuinely curious.
>>61339563
servers
>>61339573
I know, be more specific. Is it file servers, databases?
>>61339563
In-memory databases are the latest meme in enterprise.
>>61339563
>actually resorting to "why would anyone need that?"
wew
but in case you're not baiting, people on the enterprise market like to put their massive shiny databases loaded on ram so it doesn't perform like a snail, image/video processing also uses a lot of ram, just like any workload that has to deal with a lot of data at the same time
>>61339557
Proof Intel shills can't respond to this.
>>61339543
>>61339515
>>61339474
>>61338743
>>61339585
Imagine having a multi-terabyte Oracle database entirely loaded in memory.
>>61339260
We're discussing server shit. And Intel's server shit will put a hole in your wallet, bank account and mortgage your house.
>>61339543
OHHHH GOD IM CUMMING SO HARDDDDDDDDDD!!!!!
>>61339585
SAP HANA
>>61339618
>Y-you cause i-i-instability when going s-so low. O-our God did not meant u-us to tinker w-with 7nm. PLZ BUY INTEL!
Suddenly after this post all the Intel shills disappear.
It's almost as if Intel actually does pay for shills on /g/ and can't respond when they get absolutely btfo.
>>61339543
40% IPC uplift? No way. 40% frequency uplift? Sure. They will want to keep the TDP in check, though. Imagine a 6 core per CCX 48c/96t monster EPYC.
We also don't know the yields on that 7LP yet. Yeah it's IBM, but it's also GloFo. Would love it if it's great, though. These Intel Press Workshops slides are fucking hilarious. 2017 best timeline
>>61339791
6 core CCX would be the next logical move but AMD needs to be careful they don't go ludicrously dense per CCX otherwise you get to the skylake-x problem where you can't move heat from the die fast enough to keep the cores cool. Half the reason why EPYC and threadripper can actually be cooled is their packages are massive which makes them really easy to cool.
>>61339785
that first response to epyc some weeks ago and now these slides just prove that they actually have sponsored shills here, the arguments they make are way too close, just a few days ago there was one guy saying that threadripper would suck because it has 2 numa nodes/domains or whatever
>>61339791
https://www.globalfoundries.com/news-events/press-releases/globalfoundries-track-deliver-leading-performance-7nm-finfet-technology
Boy did that Global Foundries and Samsung technology partnership pay off big time.
Exciting times.
>>61339832
they won't have that problem because they actually use solder, meanwhile skl-x cpus have the die at 90°C and the ihs at 30°C
>>61339563
I wish to understand what you can do with a xeon 4 cores that support 768gb ram and cost 4k because memory support is the only difference from 7700k (lower frequency too)
>>61339881
I wasn't trying to start an argument. Reply to someone else.
>>61339881
it's $7000
>>61339785
Intel is doing their jobs for them, now.
Reminder: A Jew never changes.
Intel will always have unethical and questionable business practices.
>>61339842
Exactly. The thing about this is I own a 6700k and 7700k. Intel should of did one of 2 things. Focus on the mobile x86 architecture or churn out more investment for the desktop architecture.
Intel did neither and will suffer. They keep talking about these "technology features" yet they never make it to the average user and are only available for highend Xeons that cost 2k+.
I'm rooting for team red to gain 50% of the x86 market.
>>61339881
Load a database with low amount of queries into RAM? I don't know shit about RAM databases but this is only thing I can think of. Also maybe a backup server?
>>61339846
IBM, not Samsung. lern2read
>>61340067
its all three actually. Samsung will be using a variant of the same process that they all collaborated on, though Samsung won't be offering a 7nm node until they have full EUV integration.
You can read about it on EE Times.
>>61340067
We were both right. IBM and Samsung.
>>61340140
>pic related is Intel
>"By biting the bullet and skipping 10nm, GF opened up the technical bandwidth to attack 7nm head-on. Others have been dividing their resources and going for half- or even quarter-nodes."
Everyone points and laughs at Intel.
>>61340213
Intel will just lie through their teeth and claim they're superior no matter what. The consumer will believe what intel tells them to believe.
Thats how they operate.
>>61340250
Ryzen sales tell otherwise.
Wasn't everyone except Intel cheating with how they count number of nms?
So an Intel 10 may be equal to other's 7 for example because Intel doesn't cheat?
>>61340273
How are those x299 sales doing bud?
>>61340273
Oy vey! Don't you see how inconsistent Ryzen is! Its glued togetha!
>>61340279
glofo's 7nm is denser than intel's 10nm, but it isn't what intel would consider by their own metrics real 7nm
>>61340279
No, 7nm LP is both denser and offers higher fmal/lower fmin.
Also lower mask count.
How the hell did Intel managed to have higher mask count with inferior node?
>>61340291
Did anyone ever buy these housefires?
>>61340279
You shills said the same about Zen 14nm being a lie. Is that what you think? It's a process revamp. It's a full shrink.
https://www.globalfoundries.com/news-events/press-releases/globalfoundries-track-deliver-leading-performance-7nm-finfet-technology
>>61340317
the richest of dumb /v/ kiddies probably did
>>61340250
In consumer spaces there's going to be a few diehard Intel fans no matter what. General consumers just buy whatever is on the shelf at Best Buy. In server spaces, they're a lot smarter about what they plan on rolling out.
>>61339474
7nm would be at best released at the end of 2018 or early 2019, the process won't be ready until mid 2018.
>>61340309
The jew doesn't consider it "real".
Those sales figures are gonna feel pretty real in a year and a half.
>>61340279
Not exactly cheating, but naming scheme has nothing to do with feature size now. It's just marketing now.
>>61340355
Anon HVM starts H2 2018.
>>61338451
I like how Intel didn't mention how thanks to AMD's Memory Encryption, that all VMs running on an EPYC are completely isolated from eachother and individually encrypted on memory
>>61340375
delid this
>>61340375
Because they don't have comparable feature.
Why would they advertise their competitor?
>>61339691
>>61339971
You know it makes no sense. 4 cores for a database. Even running Oracle in memory doing that with a shitty low core is stupid. I'm just guessing because I can't find usage for a chip so costly with 4 cores.
>>61340279
This is just a horridly misguided myth that keeps getting regurgitated. No one in the industry is lying about anything in regards to process metrics.
The marketing name for process nodes is based on a few things, none of which are the specific gate length of a transistor, though at times they may line up.
The min line that the fab can expose and etch is often used for their specific marketing name. However ASML accounts for all of this, and has a set of guidelines for what denotes a node of a certain class.
The whole notion that anyone was lying about anything came from a couple intel fanboys on forums like Anandtech. With 14/16nm TSMC and Samsung created what they called "transitional" FinFET processes wherein they used their existing 20nm BEOL and a new smaller FinFET FEOL.
Intel shills of course took this and started spamming around the web
>HURR NOT REAL 14/16nm!!!!
Conveniently ignoring the fact that intel's own 22nm trigate process was hilariously far behind industry 20nm nodes in density across the board.
>>61340355
Did you read the article? >>61340334
The jews got caught off guard and are still trowing heavy cash at 10nm. Looks like the Poos pulled a fast one with Samsung and IBM.
>>61340394
>Why would they advertise their competitor?
Yes.
Why did they spend half a product launch talking about how their competitor is 100% guaranteed not to compete?
If they say nothing, they are simply showing that they live in denial of the inevitable, if they say anything, then they're giving AMD the spotlight.
>>61340401
imagine a REALLY big database that people don't have to access a lot, but when they do, it needs to be instant
>>61340402
You have to admit it's misleading. Two wrongs don't make a right.
>>61340394
>>61340473
>intel's 10nm is both less dense and offers less performance but needs MORE masks
WHAT THE FUCK ARE THEY DOING?
>>61340473
*Estimates
Holy shit, this is biblical levels of shilling to avoid the approaching process leapfrog.
Someone archive this thread.
>>61340453
Foundries shouldn't have public facing marketing names at all. The average consumer, even the average tech enthusiast knows literally nothing about the industry. Its just giving clueless people a number to be confused about.
Misleading? I wouldn't say so, though it is pointless. No one familiar with process metrics is being mislead here.
>>61340430
Considerable you will pay more for 768 gb of ram than a cpu, but scaling things if I was in this position I would get at least 8 cores. I still fail to see a reason for a 4 cores zeon so costly.
Usually you end up with lower count xeon servers on small companies but this product isn't aimed at then so it's look like a fucked product.
>>61340495
they need a lot of money for those diversity hires so they aren't seen as bigots, didn't you know all the big and trendy companies are doing this?
>>61340495
>7nm EUV reduces area further
>it reduces power further
>it reduces mask count, complexity, reduces variability, and increases yields
>intel will still be on their 10nm node
Let that sink in for a moment.
>>61340522
That is bollocks they pay shlit for niggers and pajeets. Money go to executive bonus.
>>61340495
They like to use as little transistors as possible. They are penny pinching so hard milking the core architecture.
Their new architecture better be out of this world amazing.
>>61340545
Exactly, a high yield 7nm Zen2 architecture Intel shills can't deny.
This is biblical.
>>61340545
Zen already makes the jews screech.
What will happen when Zen2 launches?
>>61340619
>Rabbinic howling intensifies
>>61340545
Better yet GF has 5nm already in the works for the future. Intel is still trying to solidify 10nm on target for 2h 2018. Even funnier Intel was planning on sticking to 10nm until 2020.
>>61340797
They will stick with 10nm until 2020, and probably later.
5nm will also take a looooooooong time to hit the market.
>>61340840
That's why GloFo has 7nm LP+ with EUV.
>>61340840
Intel's 7nm facility in Arizona will still be under construction for at least a year. It'll take over a year to get the facility clean enough to start handling wafers. They're a very, very long way off from moving beyond 10nm.
By the time intel is producing 7nm chips with their new arch AMD will be on a new arch of their own with 5nm GAAs.
>>61340840
IBM is already grasping at 5nm with EUV lithograph (the same technology GF is adopting for production.
AMD is sprinting while Intel is still on the line.
Sauce.
https://arstechnica.co.uk/gadgets/2017/06/ibm-5nm-chip/
>>61340908
I'd say GloFlo/Samsung/IBM will reach working 5nm cpu's one full year before Intel reaches their 7nm.
Daily reminder.
>>61340935
Nice broken clock you got there, Intel.
>>61340928
IBM already hitting 5nm with EUV. >>61340914
Even further down the road IBM invested $3 billion in Graphene nanotechnology manufacturing over five years. 7nm graphene circuits in 2020. Nanomachines 2025?
>>61341069
>IBM already hitting 5nm with EUV
Yes, making transistors. I talked about full CPUs, full products.
>>61341069
http://www.datacenterdynamics.com/content-tracks/servers-storage/ibm-invests-3bn-in-graphene-semiconductor-research/87785.fullarticle
>>61341084
They made a working 5nm GAAFET chip.
It was a POWER of kind.
>>61341102
Did it work? Of course not, it was just a proof of concept.
>>61341118
Schlomo please.
Stop doubting Nazi science.
>>61341084
Chips. 5nm chips by IBM with EUV. GF also invested in 5nm with EUV. Intel 7nm facility isn't even build yet. Slotted for 2020+.
>>61341143
Truly this is the best timeline.
>>61341128
All I want is Intel to crash and burn, or at least stop being jews.
I'm just being realistic about IBM nazi magics.
>>61341171
IBM nazi magicians manufactured working 5nm GAAFET POWER chip.
The next Intel announcement should just be the Intel CEO coming out and saying just fuck my shit up senpai and ending the press event.
>>61339861
The use of solder (or lack thereof) isn't the deciding factor in skylake-x woes. It simply is too many cores in such a small surface area because clocks are pushed too high to get the desired performance. I bet if anyone bothered to do a proper underclocking analysis skylake-x power draw would drop like a stone if back down to low 3ghz range or high 2ghz (i.e where the xeons actually run).
>>61334155
well, it seems that intel is indeed fucked and they know it. Good times ahead.
>>61339236
can you even use segmentation in long mode?
i thought you had to make all of your segments span the entire address space?
>>61341278
No I disagree. You can't overclock skylake unless you delid it. Otherwise it just throttles.
>>61333757
AMD
>16C/32T @2.4GHz base
>128 lanes
>750$
Intel
>10c/30t @2.2GHz base
>48 lanes
>$773 and a more expensive platform
AMD is just steamrolling over the low segment with no mercy.
There's no way they will still release Threadripper as a middle ground, that would literally be another shoah.
Intel will have 50% market share by 2022 mark my words. I'm setting up my robbinhood account at dropping 5 grand on AMD stock. Gonna make 20 large easily.
>>61341457
>20 large
calm down tony
>>61341457
>he hasn't bought $AMD at $2 per share
>>61341278
SLX runs hot and has massive power draw no matter what, but there's something seriously wrong with the IHS and TIM. There's simply no heat transfer unless you're running a custom loop, the CPU gets hotter and hotter while the AIO stays cool and doesn't raise proportionately. Intel massively fucked up not using solder, there's no way around it. FX 9590 was a housefire but it was manageable since you could still rely on the solder to transfer heat effectively.
>>61333757
>CS, DS, ES
Nostalgia, anon. Intel is still assblasted over flat memory models.
I have never seen a company get absolutely pummeled in every direction from a competitor like this before.
>>61341457
pretty conservative estimate given that their server selection already looks like dogshit compared to epyc and you're looking at a 20-30% performance gain across the product stack from Zen2 at a minimum
>>61341682
I have some disposable income should I drop some cash on AMD stock?
>>61341384
>>61341828
>10c/20t
Doesn't change the fact that AMD looks a lot better.
Also, is Threadripper is still going to be a thing, or was that just a codename for Epyc?
>>61341854
I think it was a codename for the lower core EPYC cpus.
>>61341854
Epyc is AMD's srs bsns server chips. Threadripper is their X299 equivalent.
>>61341854
Ryzen: 4-8c Normal desktops
Threadripper 10-16c HEDT
Epyc: 8-32c Server cpu
>>61341894
>Threadripper is their X299 equivalent.
Except without the spontaneous combustion.
>>61341774
they're at a near year-high right now so it depends on whether you're looking for a long term investment or to speculate like a jew. you won't see anything like the returns people were making last year and it's pretty volatile on a day-to-day trading basis but a bankable money maker down the track
>>61341922
Guess we'll just have to settle for superior performance then. That is the problem with going AMD - you get superior performance and less housefires and this is why they will remain a second-class brand.
>Zen3 will be glorious 7nm EUV
>higher IPC
>higher clocks
>lower power
What an incredible time to be alive.
>>61342103
>Zen3
No anon Zen2. I repeat Zen2.
>>61341926
I have 2 grand and I'm thinking about doing it for a long term investment.
>>61342173
7en2 is 7nm, but won't use EUV.
>>61342103
>>61342173
>>61342255
>>61341069
>IBM Nazi Wissenschaft magic can make Graphen memes work
>Intel unironically shitposts in marketing slides
What a time to be alive
AMD, please use Smug Su during Threadripper launch :3
>>61341921
Too bad they will all have different sockets. But at least I hope we can expect them to last more than 1 generation..
>>61342445
Technically Threadripper and EPYC technically share the exact same socket. Just EPYC is designed to run without a chipset.
>>61342319
Seconding this. The memes will sustain me.
>>61342319
You'd think Intel would be pretty much invinsible after ten billi after ten billi coming in quarterly but instead they do a parody/show true face of capitalism.
>>61342311
Ayymd is going to crush Intel with Zen2/3
>>61344185
If AyyMD won't be able to grab significant portion of CPU market using Zen architecture then there is little hope for the industry
>>61342487
technically?
Technically speaking, AMD could ride Zen with incremental improvements all the way until they hit 64c/128t CCXs as a single die. I mean, for fucks sake, Zen scales at nearly 97-98.9% up to 64 cores across 4 sockets, before performance starts degrading.
And once they hit 64c/128t CCXs, they just have to come up with Infinity Fabric 2 that can handle 4 die @ 64c/128t CCXs and repeat. They literally have a black magic uArch on their hand that can utterly slaughter Intel on every front, with a linear scaling paradigm.
Its insane. The jump to 7nm on the IBM/Samsung/GloFo is going to be a huge eye opener at the real, raw potential of what Zen can do and offer; and then the jump to 5nm EUV is going be an escalation of that "OH MY FUCKING GOD WHAT?!" moment.
>>61339618
>>61339557
>>61339543
>>61339515
>>61339474
>>61338743
>>61344185
>>61342103
Fake news. Good luck taking over the market.
People will still prefer Intel. They have brand recognition for server market. AMD who?? Oh the Zen chip is targeted for gaymen.
>>61345299
ayy lmao.
https://www.semiwiki.com/forum/content/6879-exclusive-globalfoundries-discloses-7nm-process-detail.html
>>61345330
Opteron was a thing. Also, lol @ "gayman" targeted Zen when their own stupid slides have some comparison saying it's crap at gaming based on day one benchmarks. >>61334155
>>61345330
>Oh the Zen chip is targeted for gaymen
>>61341854
EPYC codename was Naples, Threadripper is Snowy Owl
>>61341478
>tfw didn't fall for the buy amd stock meme
>>61334069
What did the text on this meme say, originally?
>>61347446
Never mind, it was "an Albanian"
>>61342201
Wait until VEGA reviews come in first.
>>61341854
>Also, is Threadripper is still going to be a thing, or was that just a codename for Epyc?
It's still a thing. Same socket/package as Epyc (= 4k pins) but only 2 die and half as many DDR4 channels and PCIe lanes. Also probably no buffered RAM support, so 4c*2dpc*16GB/DIMM = 128GB memory limit compared to absolutely nuts 2TB per socket limit for Epyc.
>>61335004
Only $1499 for an Intelâ„¢ Xeonâ„¢ Decoder Ringâ„¢ goy
>>61345030
No way. They'll tap out like Intel at Skylake X core levels where they bunched all the cores into 1 die. CCXs are great but AMD has to be careful with how they scale them out with power consumption and heat in mind. I think the max that can happen is probably the current Ryzen setup on 1 die as a CCX, 8 cores 16 threads, and on 7nm/5nm.
>>61345030
you're out of your mind.
there won't be nearly enough external memory bandwidth to sustain anything close to that.
Ryzen/TR/Epyc sockets are frozen until DDR5 in ~4 years, and you're not gonna see 16c Ryzen 3.0s with 2 DDR4 of 3.2-3.6 GHz memory.
12c/24c/48c sockets will be the biggest we see until we get platforms with more exotic stuff like on-package HBM3 or similar.
>>61333757
>that l333T post
>>61348457
A 6 core CCX is a possibility. We could wind up with 6c/12t mainstream Ryzen2, and 12c/24t Threadripper2.
Intel will be increasing core count in their SKUs consistently, and AMD can't afford to be left behind yet again.
Zen's issue regarding memory isn't bandwidth. The IMC is remarkable when it comes to throughput efficiency, and DDR4 speed will be getting consistently higher in the coming year. Zen's issue with memory comes from the inter CCX bus in the Infinity Fabric being bound by a 2:1 ratio with DRAM speed. Faster memory directly improves cross CCX performance.
AMD could afford to add cores to the CCX design, all they have to do is tweak their IF ratio a bit. 2400mhz DDR4 is still some of the most common sold today. Maybe a year from now 3000 to 3400 will be considered average, while 4000mhz+ kits are reasonably priced high end. So long as the average bandwidth available continues to increase AMD doesn't have any memory issues to sort out.
>>61348578
Jesus I just had a fucking stroke.
6 core CCX, not a 6c complete chip. 12c/24t Ryzen. 24c/48t Threadripper2. One of the early rumors was that their next gen enterprise chips would be 48c and 96t as a step up from the current 32c/64t chips.
Staring at this AutoCad grip is doing something to my mind.
>>61339563
Simulations, render farms, training neural nets and analytics are some that i know of
>>61348668
Also poorly optimized clusterfuck databases for every other company's shitty CRM/ERP shitware.
>>61345330
Any infrastructure fag remembers the days when AMD was the best bang for the buck for compute, and those days look like they're back, thank fuck. Hopefully enough of those fags are CIOs now to give AMD a break into the market.
>>61334155
this chart was made for dilbert's boss
What is Intel smoking? I want some of that.
>>61334317
>up to
>>61334978
it reeks of desperation actually
>>61338307
get this to newton and leibniz
>>61350091
They inhaled some of that NVIDIA Graphworks(tm).
6 years in paint, voila.
>>61350359
Delete this, goy.
>>61333757
Why don't they round those numbers? FUCK
>>61350479
intel floating point is so accurate that they don't need to
>>61350479
They literally couldn't if they only had to use rounded numbers there'd be an even longer list of bullshit.
>>61334753
Gonna crosslink sql hijack the fuck out this shit
>>61333757
intel 8156, 4c/8t skylake $7007 >LOOL
>>61342311
>EUV light sources
>Either use a laser to vaproize tin droplets
>Or build a synchrotron
>>61351948
Or they could use a free-electron laser as the light source tuned to EUV wavelengths. But I think one strong enough to hit the required power and frequency level would be even bigger than a synchrotron, and would require one for each machine that needs EUV light.
>>61351948
Yes, EUV is finally going mainstream in ~two years.
>>61335012
an 1800x would literally outperform that chip. if manufacturers actually release real ECC ready boards for ryzen it'd be a nobrainer
>>61333757
Is this real? 8 channel ram for $475? Holy shit sign me up Daddy!
>>61352281
Yes, AMD gives you 8 channels of RAM and 128 (one hundred and twenty eight) PCI-E lanes for $475.
You even get 4megos of L3 per coar.
>>61348578
>Maybe a year from now 3000 to 3400 will be considered average, while 4000mhz+ kits are reasonably priced high end. So long as the average bandwidth available continues to increase AMD doesn't have any memory issues to sort out.
JEDEC peaks out at DDR4-3200, and everything else will always be unofficial overclocking territory.
Zen 2/3 will cap out at 2*6c CCXs and not 2*8c or higher because most tasks can't just sit resident in L2/L3 cache and actually need healthy amounts of memory bandwidth.
Also, there won't ever be 3 mem controllers per die, since a 19" wide server chassis just barely fits 2 Epyc sockets' and 32 DIMM slots' width.
>>61336493
Fuck off
>>61352350
There's always HBM2 for on-package memory.
>>61352281
No, that's for 2 socket systems. So you'd need to buy two of them.
>>61352365
It's 1S/2S.
P SKUs are 1S only, the rest are 1S/2S.
>>61352281
Yep, and no arbitrary memory or PCI-E lane limitations either. You wanna get that $500 chip and slap 2TB of memory on it? Go right the fuck ahead, it will do it.
Intel? You're limited to 768GB unless you shell out bigtime.
>>61352365
He's talking about Epyc, which is 4x 2-channels per processor.
>>61351965
FEL can really crank up power higher than the tin drop bullshit, but the dielectric mirrors for EUV apparently already need water cooling to put ~100W on target.
No need to make fuckhuge machines that will just vaporize your mirrors etc. faster.
>>61352383
Yep, that's the other problem. I wonder how chilled the water being fed to the mirrors has to be to keep photonic/thermal damage to a minimum?
>>61352468
>>61352468
Of course.
Do you have proofs of your statement? Except for Intel marketing slides
>>61335035
>This is for L3 optimised workload
Kek. No.
The absurd price is for the scaling up to 8 sockets, so that you can have a large set of working memory which is used to offload onto coprocessor based nodes for HPC. This is actually one of the better uses of the seemingly bizarre segmentation.
>>61352429
the magic of the photoelectric effect means it doesn't actually fucking matter too much.
the chilling keep the mirrors from getting soft and deforming, but every single photon absorbed in the dielectric stack still smacks around electrons with the same amount of energy, regardless of the overall light intensity.
EUV is absolutely the last stop in optical lithography.
You just can't make, guide, or stop even soft x-rays in any sensible way, and EUV will be rough enough as it is.
>>61352365
dumb nigger, it's 8 channels/socket
>>61352498
We take all competitors seriously, and while AMD is trying to re-enter the server market segment, Intel continues to deliver 20-plus years of uninterrupted data center innovations while maintaining broad ecosystem investments. Our Xeon CPU architecture is proven and battle tested, delivering outstanding performance on a wide range of workloads and specifically designed to maximize data center performance, capabilities, reliability, and manageability. With our next-generation Xeon Scalable processors, we expect to continue offering the highest core and system performance versus AMD. AMD’s approach of stitching together 4 desktop die in a processor is expected to lead to inconsistent performance and other deployment complexities in the data center.
>intel shits on amd because it has 4 numa nodes/cpu
>does the same thing
HAHAHAHAHAHAHAHAHAHA
>>61352678
>the kikes literally tell you to cosplay EPYC with your shiny new Xeonâ„¢ Scalableâ„¢ Familyâ„¢ Processorsâ„¢
JUST
>>61345030
I AM ERECT
>>61352678
LMAO
>>61352678
heh
AMD were really stupid for not making a workstation epyc SKU. 2P xeons are still the only option due to AMD being lazy. If you're not concerned with power use 2p xeons come in high TDP versions while epyc 1p systems only let you run a single 2ghz 180w cpu.
With a proper motherboard and and unlocked multiplier the 32 core epyc would likely run at 3.4ghz and beat the xeon 2p options.
>>61353113
2.7 boost for 32 goarz is not enough for you?
>>61353113
>implying that pcie lane amount and memory speed aren't more important than raw cpu performance
>>61334317
I like how one of their slide monkeys let the wordbubble overlap the line between AVX2 and AVX512.
>>61353113
do you really think 2p xeons can compete against a 32c epyc?
even if it gets only 1200 points for each 8 cores at 2.7GHz (all core turbo), this is already 4800cb, and let's be realistic here, very few 2p xeons can score higher than this
>>61340334
Wow, so 7nm Zen before Q3?
Nice, I'm going to pass on Ryzen this year then, and just wait till 7nm gets released.
Can I play Arma 3 on ultra with it?
>>61353340
no because arma 3 has the worst optimization ever
if you can find a single core cpu that runs at 30GHz, maybe it will do the trick
>>61333757
>8ch ddr4
Damn. I'd get 8 4GB ddr4's and be set.
>>61333757
>those xeons with no turbo clocks
HAHAHAH
>>61353356
Mostly true, but remember instructions per clock (which is hard to measure) means you cant really compare clocks across many cpus. Generally only do it within a certain nm of cpu
>>61353113
>AMD were really stupid for not making a workstation epyc SKU.
What is the single socket 16-32 core EPYCs, anon?
>>61353448
Fucking Threadripper also makes a decent werkstation.
>>61353460
yes, but if you really need the additional memory channels and lanes, you buy epyc
>>61353460
This, I think ECC will be board specific instead of market specific like EPYC, but ECC is there and TR does support 512GB of memory.
Double of the Skylake-X
>>61341278
Indeed if you keep lower frequencies (less than 4ghz) it is very very efficient. The problem is, under 4ghz you better go ryzen since it's almost same ipc.
The only Intel advantage is clock albeit it come at expense of logarithmic increase
>>61353487
>TR does support 512GB of memory
what? never heard this anywhere
>>61353485
4 channels and 64 IO is nothing to scoff at, that's the same amount of channels and more I/O than 20+ core Broadwell Xeons had, they're definitely not lacking.
>>61353487
TR is 128gb max?
It has no RDIMM support afaik.
>>61353506
During some DELL interview, Robert Hallock mentioned that TR supports 256GB compared to Intel HEDT at 128GB.
Lemme find it
https://www.youtube.com/watch?v=Bk8hU9U7Ryo&t=9m42s
>>61353526
OH GOD
IT GETS BETTER
3k CINEBENCH PTS
NOT A FIRE HAZARD
256 GIGS OF ECC MEMORY MAX
YAY
>>61353526
>>61353550
if this is true then I'll have some fun with ramdisks
>>61353526
>area 51 threadripper will come with factory overclock
muh dick
>>61353526
WEW FUCKING LAD
>>61353336
>>61353526
9:40 for the guy saying max ram
>>61353736
>t=9m42s in the URL
NO shit
>>61353816
sorry, when I played it it started from the beginning
HAHAHAHAHAHAHAHAHAHA
http://www.anandtech.com/show/11636/amd-ryzen-threadripper-1920x-1950x-16-cores-4g-turbo-799-999-usd
>>61353834
>$999 FOR THE 16C/32T
HOLY SHIT
>>61353286
>even if it gets only 1200 points for each 8 cores at 2.7GHz (all core turbo), this is already 4800cb, and let's be realistic here, very few 2p xeons can score higher than this
Is it confirmed to do 2.7ghz on 32 cores? There's a video of dual 32 core epycs only scoring 5446 in cinebench which would make the 32 core 1p version slower than thread ripper in 3d rendering. Unless the single socket version clocks much higher in turbo.
>>61355433
sauce on the video?
>>61357582
>>61357601
did you really just bump this thread?
>>61357582
>>61357601
wtf are you doing retardo tripfag plz go
explain this bullshit
>>61357756
they have to sell their shitty bins anon
and let's not forget 60% margins
>>61357756
>$7007
wew
>>61357756
It's "moar L3$ per core" part but since Intel gutted L3$ even 7251 is better for cachebound software.
>>61355683
I dont get how performance is so bad compared to threadripper. Performance per watt is good but they have left a gapping hole in their cpu lineup. Highend workstations will have no choice but to continue to put up with 2p xeons. Performance per watt isn't critical so 400 watts of xeons are going to beat a 180 watt 32c epyc.
https://www.youtube.com/watch?v=32hTAooG98c
>>61358306
>performance per watt isn't important in the datacenter
>TCO isn't important
anon pls
>>61358306
CB15 is not the best thing to test server CPUs with.
>>61358306
Plz read an actual review. http://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade
>>61358306
that thing is scoring a lot lower than it should
that score is close to what a single epyc 32c should/would look like
>>61358425
CB is shit at multi-socket.
>>61358446
Doesn't Intel multisocket just scale like shit in general?
>>61358479
Yes, it runs into QPI saturation very fast.
>>61358446
yes, but these scores are just awful, it's giving an extra 10% performance increase over what a single 7601 should get and not pic related
>>61338307
>A socket change a year, keeps the goyim in fear
>>61358535
What part of
>CB is shit at multisocket
did you not understand?
Also EPYC is bretty new.
>>61358546
intel's multisocket performs better on cinebench than amd's
>>61358554
Because Intel's QPI is old as shits and IF is new (and is a blackbox for a good reason).
>>61358562
do you think they'll fix this?
>>61358570
Yes? Maybe?
No?
Anything buy SPEC is mostly irrelevant.