AMD EPYC and Intel Xeon

>>61322040
FUD and promises of MCM Soon(tm).

>>61322040
More FUD, more bribes.

File: 87247.png (26KB, 650x525px) Image search: [Google]

26KB, 650x525px

DELID THIS

File: intel ceo JUST.jpg (119KB, 1000x1000px) Image search: [Google]

119KB, 1000x1000px

>>61322443
>a-at least we still got 4S
>battle t-tested

File: Do want.jpg (65KB, 712x687px) Image search: [Google]

65KB, 712x687px

>>61322275
>promises of MCM Soon(tm).
There is no other way they can compete. Their yields on those huge dies must be atrocious. AMD could undercut them further but choose not to. Also, holy shit the 1700 is at $270, I'm really tempted.

File: amada do it for her.png (655KB, 1920x1200px) Image search: [Google]

655KB, 1920x1200px

>>61322470

File: 78012.png (26KB, 650x525px) Image search: [Google]

26KB, 650x525px

NOOOOO

>>61322522
he he
WHAT THE FUCK

But how much does it cost?

>>61322557
>The current Intel pricing draws the first line. If performance-per-dollar matters to you, AMD's EPYC pricing is very competitive for a wide range of software applications. With the exception of database software and vectorizable HPC code, AMD's EPYC 7601 ($4200) offers slightly less or slightly better performance than Intel's Xeon 8176 ($8000+). However the real competitor is probably the Xeon 8160, which has 4 (-14%) fewer cores and slightly lower turbo clocks (-100 or -200 MHz). We expect that this CPU will likely offer 15% lower performance, and yet it still costs about $500 more ($4700) than the best EPYC. Of course, everything will depend on the final server system price, but it looks like AMD's new EPYC will put some serious performance-per-dollar pressure on the Intel line.

File: 1496348874750.png (370KB, 800x627px) Image search: [Google]

370KB, 800x627px

>>61322040
obligatory

>>61322040
Server markets are going to end up being owned by AMD. I don't think anyone can argue with that.

>>61322557
A fraction of what Intel's offering does, that's all that matters. Business applications are all about profit margins, Expect Apple and every other workstation OEM to switch over to EPYC asap.

File: 67413sys.png (29KB, 650x550px) Image search: [Google]

29KB, 650x550px

DELET

File: GOODGOY.png (36KB, 546x526px) Image search: [Google]

36KB, 546x526px

>>61322557

>>61322592
Apple needs to embrace RR for their next Macbook's, for the first time they will not be throttlebooks and they will finally have decent battery life.

>>61322586
Everything oscillates

File: scytheninja4sideview.jpg (182KB, 800x600px) Image search: [Google]

182KB, 800x600px

>>61322040
If you are customer, you can be only glad that the intel monopol is over.

If you are amd fanboy, enjoy while you can. If you are intel fanboy, do damage control and post only single thread benchmarks.

Doeas anybody know how will the cpu coolers look like? Something as massive as ninja 4? Does even something exist for it yet?

>>61322629
Owned != permanently dominated

File: 394820183840.png (81KB, 400x397px) Image search: [Google]

81KB, 400x397px

>>61322443
>gcc btfo
take that stalldrones

>>61322641
this needs to be stickied on the front page of /g/

File: IntelServerFires.jpg (168KB, 759x662px) Image search: [Google]

168KB, 759x662px

Oh dear. . . Looks like Intels TDP means jack shit yet again, with AMD using much less power under 100% load, though the idle and MySQL test looks funny for AMD, might be something going on there. Anyway, EPYC is ~25% more efficient under full load in that test (which EPYC beat Intel), that's a huge fucking deal. When you weigh TCOO, power consumption is pretty much the top of the list.

>>61322040
>server processor
Literally who gives a shit

File: 1492855909643.png (944KB, 1228x1502px) Image search: [Google]

944KB, 1228x1502px

>>61322040
I HATE AMD SO FUCKING MUCH

>>61322692
Leave /g/.

File: 1493381622920.png (6KB, 448x357px) Image search: [Google]

6KB, 448x357px

>>61322689
less initial investment with less operating costs at virtually the same performance more or less.
EBYN

>>61322689
>MySQL test looks funny for AMD, might be something going on there
Lots and lots of hops between the dies/sockets.

>>61322719
Fun fact, unless you own a large-scale company you will literally never benefit from a server

>>61322741

Yeah, that would do it. Looks like an area for further improvement in the future for AMD. If they can get latency+overall power down from constant die/socket hops then they will really crush the market.

File: 1495593910211.jpg (52KB, 640x640px) Image search: [Google]

52KB, 640x640px

>>61322761
>you can't talk about F1 cars in a car forum because you will never need one anyway.

>>61322692
>not buying low end EPYC for your homeserver

>>61322519
that's a pretty qt corpora-tan

>>61322040
WHEN WILL AYYMD RELEASE SOURCE CODE?

I just want to point out that those Anandtech benchmarks are a bit biased in favor of Intel because they are using an old version of Ubuntu with a kernel that doesn't support the zen architecture.
>http://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade/11
>"All of our testing was conducted on Ubuntu Server "Xenial" 16.04.2 LTS (Linux kernel 4.4.0 64 bit). The compiler that ships with this distribution is GCC 5.4.0. "
this is actually hilarious, Intel is legit about to be murdered in the server market.

>>61322959
entitled FOSS-tard plz jump from a cliff

>>61322988
Don't get me wrong AMD used to be great (I still use my librebooted d8 system with 2 AMD opterons) but now they have become just as botnet as intel.

>>61322689
>The power delta for the EPYC system is 177w.
>The power delta for the Xeon 8176 is 263w.
>The EPYC system performs better

This is a fucking bloodbath. AMD has pure gold on their hands.
Their 7nm parts are going to steal the entire data center market.

>>61322984
>Anandtech
>biased
Stop the presses! They can be dicks about it sometimes, but they called out Intel on their new AVX shit.

>>61323068
Well it's going to be POWER9+ versus EPYC2 in the end.
Intel's first 10nm server parts are far far far away due to yields.

>6cores/CCX
4 8 C O R E S

>>61323080
there aren't many sites that test 25.8k$ worth of CPU's. I agree with you but I'm just saying, if they run a newer kernel even on that old version of Ubuntu, AMD will have an even higher performance lead, and probably less power consumption as well.

one can kill a man with that brick

people in the future will look at it like we look at 80's technology

>Baidu
>Microsoft
>A&A
>Dropbox
>Dell
>SuperMicro
>Redhat
>VMware
>HPe

>80% of the server market as day 1 partners

>128 bit AES memory encryption on the fly
>32c/64t starting at $2100 for single socket support

hnnnng

>128 PCIe lanes
>2TB RAM capacity
>on all chips
>even the $500 8c/16t one
LITERALLY RAPING INTEL

REEEEE
HOUSEFIRES
AMDPOORS
PAJEET ON STREET

JUST USE INTEL DAMNIT UGHHHH

>>61323214
Brian, unJUST your fabs.
That 10nm node is a disaster.

>>61323166
pretty much, rape is the only word to describe how bad Intel just got shit on.

2x EPYC 7601 are only 8.4k
2x Xeon 8176 are 17.4k

7601 beating it by 40% in some tasks, while being on par in others, with an older kernel like this anon said >>61322984 , all while using 100w less power. simply EPYC.

>>61323166
Tyan too. They've got the whole market as partners.

File: 1498896782566.jpg (78KB, 396x385px) Image search: [Google]

78KB, 396x385px

>>61323166
HOLY SHIT!!!

>>61323166
It is ogre. Intel will be bleeding money for the coming years. Let us see what will their (((diversity))) hires bring them now.
Fucking kikes.

File: datacenterFire.jpg (34KB, 600x450px) Image search: [Google]

34KB, 600x450px

holy fuck Intel what are you even doing...

"Power draw is recorded every second in our enterprise lab. During the Linux-Bench tests, we captured multiple high-draw bursts that only appeared for one second, cresting at 711W. The same granularity is used during our Linpack tests, but because of the 8176's lower AVX frequencies, we only recorded a 670W peak. "

>>61323355
>all while using 100w less power. simply EPYC.
Using 100w less power, while supporting more PCIe lanes, and more memory, all of which's controllers require power to run.
Honestly, I'm still shocked by that 7551p, fucking 32/64 for $2100

File: 1378555653117.jpg (17KB, 400x400px) Image search: [Google]

17KB, 400x400px

>>61323166
>>61323508
>>61322689
>>61322984
I'm crying brahs

>>61323556
yeah that's going to be one beast-mode workstation CPU that Intel has absolutely no chance of competing against with their shitty 40-44 lanes.

>>61323508
What the...

File: 1494445893840.jpg (162KB, 800x627px) Image search: [Google]

162KB, 800x627px

>>61322582
We need to go wider.

>>61323591
Yeh, between the 7551p and the 7251, Intel have trouble at both ends of the market. Then you've got shit like the 7351, and 7301, which improve on Intel's offerings there significantly, 7301 by up to 70% in SPECint, and that's where a good 60% of the server market alone lies is at that price point

File: image.jpg (29KB, 600x450px) Image search: [Google]

29KB, 600x450px

>>61323508
It's actually more efficient

intel actually says they already sold 500k of new xeons

>>61323738
sure, but only due to it having more cores than the other models tested, its a silly non-factor metric. total system power draw is the only important metric to care about in the server space for obvious reasons.

>>61322519
Where can I find more of this?

Holy shit AMD's FPUs are fucking monsters
If these ever get 256bit pipes for AVX they're gonna eat insane amounts of power.

And this is why I don't give a fuck about AMD's GPU division, their CPUs will make them more money in a year then their GPU made them in 5 years

AMD should only care about making a few pro GPU SKUs and whatever goes into their APUs, rest is just noise.

>>61322837
And what are you going to keep on the server?
Terabytes of furry porn you autistic fuck?

>>61324317
That's how you get $2800 Nvidia GPUs.

>>61324317
I'm interested in gpus for the unified compute platform with hsa and IF. There are rumours floating around of a node with how ever many epyc cores and how ever many vega cores being visible to an external host/controller/software as a singlular compute unit.

>>61324394
Look at me give a fuck.

AAAAAAAAAY 7nm EPYC will devour Intel considering their 10nm part is in 2020/2021 at best and canceled at worst

>>61324419
You're an idiot

>>61324456
Stop assuming I care about fucking graphic cards, I'm running a fucking integrated GPU and if I wasn't I'd be running some <100$ lowend card for the multiple display outputs

>>61322040
delid
dis

>>61324156
http://www.mediafire.com/file/8v26cd117rcscs2/amadaDump.zip

File: 1475211353488.jpg (108KB, 782x782px) Image search: [Google]

108KB, 782x782px

>>61322040
STOP MAKING THESE SHILL THREADS AMD

WE HAVE TO KEEP THE ISRAEL PLANT OPEN

SHREEEEEEEEEEEEEEEEEEEE

File: 1491946472681.png (161KB, 433x484px) Image search: [Google]

161KB, 433x484px

>>61322040
Oy to the vey.

File: 1480362036890.jpg (96KB, 853x1024px) Image search: [Google]

96KB, 853x1024px

>>61324579
not the guy you replied to, but you're a saint.
she's so cute, thank you so much.

>>61324592
>SHREEEEEEEEE
holy shit im ded

>>61323166
RAPING INTEL EVERY THREAD

>>61324340
Hosting furry porn and dank Trump memes

File: Yessss.png (90KB, 222x286px) Image search: [Google]

90KB, 222x286px

>>61324592
>SHREEEEEEEEEEEEEEEEEEEE

>>61324446
>AMD releases 3ghz base clock 48c96t 7nm 2EPYC4U chip
>Brian sits alone in his basement beating off with a belt around his neck silently wishing that tonight will be the night he doesn't wake up after he finishes

Let it happen.

I'm sad they didn't test any VMs the CCXs are purpose built for them.

>>61324695
How so?
I was looking into buying the 16-core EBYN and dedicate 2 physical cores from each CCX to 4 different VMs because I didn't know how they'd perform with cores in different CCXes

>>61324689
>2 Socket AMD Starship servers will have 96c/192t of actual fat x86 cores
I can only imagine the amount of RAM and IO these fuckers are going to have

>>61324488
>wah wah i don't care about gpus therefore therefore amd, being the only other company competing in the market should quit, leaving thousands upon thousands people who do care fucked

Y'all cherry picked hard.
That duel Threadripper system lost in a lot of tests.

The lower price and power usage is it's strongest point.

>>61322040
amd64 is not good choice for server.
POWER 9 much faster, cheaper and power-efficient.

>>61323166
Probably because they got tired of Intel raping them with prices.

>>61324757
Because CCX are mostly self contained and very low latency and can be called dedicated VM silicon, 4 cores per VM is optimal, but you'd get a similar core performance and core bandwidth if you use 2 or even 1 core.

>>61322658
5.4 is fucking ancient though I haven't kept track of how much Zen tuning they've done in 7.1.

>>61324828
Latency between CCX in a single die is still pretty high. How much higher is the latency between CCXes in other dies?

>>61324879
That's why you keep a VM in a single CCX; which is really fast.
Though if you're running something that doesn't depend on other threads, ccx latency, die-die latency and socket latency are all irrelevant.

File: Jim_Keller's_medicine.jpg (156KB, 1600x888px) Image search: [Google]

156KB, 1600x888px

To complete the victory, "Snowly Owl" will rip Xeon D a new asshole. Pic related.

>>61323786
...most likely for Google, nobody else will not bother with AVX-512.

>>61323786
To Google, back in December.

Tleadulippelu sugoiii!

It seems like EPYC is indeed another unmitigated disaster for the server arena, it should have been obvious from the lackluster desktop processors

File: fdaswgwwe.png (532KB, 695x690px) Image search: [Google]

532KB, 695x690px

>>61322040
>this is the current state of (((intel)))

>>61325364
t. Brian JUSTnich

>>61325364
>State of Intel shitposting in 2017
Sad!

>>61325364
Seething.

>>61325364
You forgot your tripcode, namefag.

>>61325386
Lol

File: paźdź.jpg (18KB, 480x360px) Image search: [Google]

18KB, 480x360px

>>61325386
>How Diversity Can Fuel Innovation

It can't! You need SMART people! Not some stupid guy who only got hired for the color of it's skin.

Why do they don't understand that???

>>61325523
>but we have totally unbiased and scientifically sound statistics to prove you otherwise anon :^)
>we won't show you our methods though, why would you want to reproduce our research? are you racist or something?

>>61322040
Sir please delete these thread sir I will have to insist sir delete sir please sir

If only AMD was this good at graphic, and graphic drivers.

>>61325572
I can imagine Brian siting there being like

>Oh so he finish third grade
>but he doesn't know any thing about cpu architecture
>can't design a good cpu
>why would i hire him
>oh he's black

HIRED!!!!!

>>61325572
>>61325523
>>61325691
You know what's funny? They could've solved the "diversity hiring" problem by just removing the race checkbox from the hiring form and hiding the name until they're already on the interview track. Problem fucking solved, you don't find out that it's Pajeet Rajesh until he's already gotten an interview, and then you can figure out from interview recordings whether Pajeet lost the job because he's an idiot or because his interviewer screamed "SHIT ON SOMEONE ELSE'S STREET" and kicked him out.

>>61325523
Panie, kurde jak nie bedzie murzyna w firmie to nam nie bedo procków kupować rozumiesz pan to?

>>61323508
>711 W
Is that when some microcontroller or PnP melts together and causes a resistance spike? How the fuck does the socket even handle that?

>>61325704
But that would mean they would base their job hirings on old and decrepit form of MERITOCRACY instead of DIVERSITY. Bad goy!

>>61325709
Panie, przestań pan pierdzielić głupoty.

>>61322040
Delete this now if you enjoy living.

File: 1496977465049.png (334KB, 679x702px) Image search: [Google]

334KB, 679x702px

SHUT UP SHUT UP SHUT UP SHUT UUUUUUUPPPPPPPP

>>61325723
fuck if I know, but x299 is just as much of a disaster and they dont even have the high core count CPU's out yet.

>>61325730
Sigh.
*unzips capitalism*

>>61325079
Fuck off Serval.

>>61325765
Can you imagine the 16 core intel cpu? Oh the house fire!

>>61325723
They need a bigger Bipline

File: delidthis.png (107KB, 2250x2324px) Image search: [Google]

107KB, 2250x2324px

>>61325822
there are already 24c+ Xeons
although the cores are downclocked to hell

>>61325764
>334kB 679x702
as the person who made this, it triggers me to no end
have the superior 2250x2324 100kb version

>>61325864
fine whatever just delid this thread pls

>>61324404
That is the silicon wet dream. Good luck accomplishing that.

>>61324488
>I'm running a fucking integrated GPU and if I wasn't I'd be running some <100$ lowend card for the multiple display outputs

You'd still be paying more if it wasn't for AMD, if it wasn't for them intel graphics would cost a whole lot more same with the low end nVidia cards.

>>61324788

Lost in lots of cherry-picking applications using expensive software. In the real world, this represent few people.

>>61324818
There are no power9 benchmarks, they changed the architecture again, so there might be regressions compared to power8.

I cant not read in russian
EPУCAЛИM

>>61326232
cyka blyat

>>61326232
>yerusalim
what did he mean by this?
jerusalem=isreal though

>>61322470
Kek, I bought it at 320€
No regret, worth every cent.

>>61322040

It is the Pentium 4 Xeon versus Opterons all over again.

Intel is going to lose massive maketshare in the lucrative HPC and server markets.

>>61322641
Well said

>>61322641
Well, Epyc's coolers will look like any other server heatsink: a block of copper or a vapor chamber with sheet-punched aluminium fins soldered to the top.

Threadripper's heatsinks I think noctua had a couple of prototype models on show at CES, and other than the fuckhuge contact plate they looked like any other heatsink. Threadripper shouldn't have a TDP higher than 135-140W, so in theory if it wasnt for the size disparity between them, any basic tower cooler could handle the stock thermal load.

File: 1484518809365.gif (13KB, 250x194px) Image search: [Google]

13KB, 250x194px

>>61322040

>AMD ebyn having competitive performance
>AMD ebyn consuming less power
>AMD ebyn using smaller cores on an MCM basis for maximum shekel efficiency compared to Intel's monolithic bingbong

Is this the end?

We're hitting a bandwidth bottleneck per core soon, we're unlikely to go over 8 or 10 channel memory unless you want spaceship mobos with prices to match.

On package/die L4 cache will be a requirement for these scaleout processors really soon.

>we're already getting 7nm architectures
Holy fuck how does that shit not melt itself? It's kind of amazing how far miniaturization has gone.

>>61322761
>Fun fact, unless you own a large-scale company you will literally never benefit from a server
But that's not true. Without a server you would not have been able to stain my screen with your shitposting.

Regardless, the discussion itself is interesting.

File: Xeon_Purley_decoder_ring-768x365.jpg (58KB, 768x365px) Image search: [Google]

58KB, 768x365px

https://semiaccurate.com/2017/07/11/intel-launches-purley-aka-metal-xeons/

>That brings us to the first elephant in the room, defeaturing. You might recall AMD went to great lengths to point out that all of their Epyc CPUs are fully featured. Intel on the other hand cuts just about everything useful at one tier or other.

File: 1454541338550.gif (459KB, 294x233px) Image search: [Google]

459KB, 294x233px

>>61327081
>mfw you need a literal decoder ring to figure out Intel's segmented as fuck product stack

File: Purley_SKUs-768x404.jpg (93KB, 768x404px) Image search: [Google]

93KB, 768x404px

It's heartening that Intel's response to actual serious competition from AMD is to double down on the insane pricing and tiers.

>>61326552
If defaults 4.0ghz all cores, expect +200W on 14+ cpus. There is no way you can do lesser, no matter what TDP AMD says.
Thermal transfer will however be much superior to intel.

Ryzen 8 cores at 4ghz already is over 120w.
The only thing bad about AMD cores is the lower frequency, im in the middle of TR or SLX, even with the shit im still prone to SLX due to clocks and single thread performance, but waiting for tr release for final decision.

>>61327101
>yfw shekels for RAID key™ on the HEDT

>>61327127
Well considering intel advantage over AMD is pure clock, since IPC is almost same, EPYC seen to be real nice, especially on systems that can do the same as intel server clocks

>>61323080
They didn't even mention the power and temperature issues SKX has, they're even more biased than Tom's

>>61327192
SKL-X does have better single core, but that doesn't translate to better gaming performance if that's what you were hoping for

File: what-am-i-reading.jpg (37KB, 552x395px) Image search: [Google]

37KB, 552x395px

>>61327127

AYYMD DEDICATED SERVERS WHEN
I want to rent dem 16T servers for <$150/mo

File: Purley_Price_list.jpg (156KB, 581x1011px) Image search: [Google]

156KB, 581x1011px

This is intel's response.

>>61327127
wtf is this clusterfuck?

File: 1499799908951.png (608KB, 1170x582px) Image search: [Google]

608KB, 1170x582px

>Intel flat out lying in their marketing slides

They are fucking piss-pants terrified at this point. Everyone knows Zen was designed as a server arch first and a desktop arch second.

File: smug demon - Ririchiyo.png (309KB, 692x720px) Image search: [Google]

309KB, 692x720px

>>61327475
>8 sockets
kek

>>61327507
>AMD is disjointed
>Image shows disjointed L2/L3 cache on Intel side

Intel, what are you even trying to do?

>>61322522
I'm not even a shill, but that's pretty beautiful. That's how competition should be

>>61327127
>new Xeons contain precious metals
>this clusterfuck of tiers
Guys I think they're going for a periodic table of processors.

>>61327535
Just imagine how shit that must scale.

File: my sides.png (173KB, 640x400px) Image search: [Google]

173KB, 640x400px

>>61327556
>periodic table of processors
Oh fuck.

>>61327507
>flat out lying
I think I'm stupid because I can't spot the lies.

>>61326556
No, Zen2 will be the end.
>>61327507
EPYC supports 2DPC DDR4-2667 though.
What da fuck Intel?

File: blank periodic table of intel template.jpg (224KB, 2015x1261px) Image search: [Google]

224KB, 2015x1261px

>>61327556
>clusterfuck of tiers
read that as clusterfuck of tears
>periodic table
TEMPLATE READY
BEGIN THE LISTING OF THE PROCESSORS

>>61327507
It's funny cause they cherry-picked the numbers, and AMD still comes out ahead on the L3 cache (almost twice as much!), on the maximum memory capacity, on the number of cores, and on the PCIe lanes..!

>>61322658
not only is gcc old but flags are not listed so that's not even a useful graph. icc is intel shill software and optimized for intel hardware only, with or without avx.

worst benchmarks i have seen in a long time

File: IMG_0387.jpg (35KB, 500x332px) Image search: [Google]

35KB, 500x332px

>>61327645
I love you anon

File: 1454162974074.gif (3MB, 365x341px) Image search: [Google]

3MB, 365x341px

>>61327507
>Zeppelin
>desktop die
>basically AMDs Xeon-D is a desktop die
Intel?
Not even the leather jacket man can be such a lying cunt.

>>61327276
I want a single computer to both use vms and game. Its a fucking shit using 7700K to use 3+ vms at same time with 4 cores.

Single core performance is up to par, if you can manage to overclock to 4.7-4.8. So far 7820X fails in this category, but there are many shit from intel: less pci lanes, need to delid.

Threadripper seen all round, but 4.0-4.1 ghz still is bad. At least the 2:1 infinty fabric speed + stepping can add into performance. If so i would jump to one 10-12 cores TR.

>>61327507
>Implementing ethernet and sata further reduces PCI-E lane count
Bullshit. The SATA is integrated into each die (not so sure about the networking) and thus does not impact PCI-E Lane count.

And AMD STILL has more I/O

>>61327714
It's going to be a fucking party trying to figure out what goes where on a table, might need a new thread for this. Wouldn't surprise me if it was more comprehensive as the end result thanks to autistic attention to details compared to what Intel is doing.

File: current zen node vs zen2 node.png (129KB, 1562x332px) Image search: [Google]

129KB, 1562x332px

>>61327730
You wait for Zen2.
You get many ghz, courtesy of IBM nazi science.

>>61327750
The best part is you don't have to get a new mobo for this.

File: [GJM] Urara Meirochou - 10 [7EF256EE].mkv_snapshot_18.02_[2017.03.10_15.00.34].png (656KB, 605x582px) Image search: [Google]

656KB, 605x582px

>>61327730
https://translate.google.com/translate?tl=en-us&u=http%3A%2F%2Fwww.pcgameshardware.de%2FCore-i9-7900X-CPU-266272%2FNews%2FSkylake-X-Spieleleistung-IPC-Stellungnahme-1232703%2F

>Skylake-X: Intel justifies the decreased gaming performance

>>61327750
So so is that what im really expecting, if AMD really delivers Zen2 or Zen+ or whatever fuck name, with higher frequency, it will be a killer.

TR alone if they manage to 2:1 infinite fabric speed will be a great revamp.

For anyone in the market for an HEDT that dont require bleeding edge single core performance, Threadripper seen the very much logic choice. Unless you need AVX512 or single core performance.

16 core threadripper for encoding, compiling , vms, all this is cheaper. Single core (gaming, or modelling like Solid works) still is better on intel.

Im all on hold if i get this year or next, or even make a separated ryzen crappy simple machine to run the vms apart over ethernet.

>Zeppelin
>desktop die
>ZEPPELIN IS DESKTOP DIE ACCORDING TO INTEL
SOMETIMES YOU HAVE TO SHUT THE FUCK UP AND ADMIT DEFEAT. INTEL. PLEASE
SHUT THE FUCK UP.

File: Mysterious Merchant.gif (4KB, 452x523px) Image search: [Google]

4KB, 452x523px

DELET PLS. I WORK AT INTEL. PLSS

>>61323166
>Everything this anon said
>7nm ahead of Intel's 10nm
>K12 coming for ARM server market
>Intel literally has no response for it

This truly is the best fucking timeline lads

>>61327750
Holy shit ryzen clocks to 4ghz on a 3ghz node. That 7nm node claims 5ghz. Zen 2 will be at least 5ghz + ipc improvement, might outperform an 8700k

>>61327839
No, no, don't get too hyped about it. Just follow AMD with their sandbagging tactics against Intel. Remain hopeful but cautious.

>>61327839
>might
Anon it will slaughter it.
Zen1 hits a voltage wall. Fmax for 7nm LP will be really-really-really high, it's a node originally intended for POWER9+.

>>61327772

The cache impact is measurable within some apps and games in particular, but the issue seen to be bounded to current/power limit forced by intel into bioses to keep TDP under expected numbers.
Once power limit is removed. cpu can work properly, but at expense of great furnace.

Essentially, intel is pushing same way as Prescott, but the whole L2/L3 revamp doesnt seen to be very good. It was clearly an enterprise feature shoved in HEDT, a market where will not see direct advantages with it now, or even worse performance. It was required by cloud partners to apps like databases, however, i dont understand how they fit this shit inside HEDT, or intel is simply reusing XEON arch and shoving as HEDT.

>>61327507
oy vey ignore their yields, huge dies are the way to go.

>>61327891
Moar L2 is only worth anything for databases alone and datacenter is not only about databases.
It's like Intel wants to lose server market share.

>>61327862
If they have the same 80% yields as Zen it's going to be a real shoah.

File: Laughing_Boxers.jpg (44KB, 704x480px) Image search: [Google]

44KB, 704x480px

>>61327556
>periodic table of processors.
Holy kek that caught me off guard.

>>61327925
DUV? Close to 60, maybe 70 percent because it's fucking ~80mask steps.

>>61327475
>$7,007.00
>for a fucking quad-core
Holy Shit
EBYN Jewing there, Intel

>>61327829
Its not all 7nm but very similar to intel 10nm process.
Anyway, its a race, all about frequency and power, AMD knows it, INTEL knows it. Its up how manages to release first with consistent quality and yield.

I would still remain on 1151 if intel releases a decent 6 cores coffee lake compatible with socket. +1 year till real changes come.

>>61327962
And you don't even have lotsa L3 anymore since its smaller and exclusive compared to BDW-E.
It's like Intel is thinking with their anuses.

File: Screenshot from 06-07-2017 19_12_35.png (36KB, 571x556px) Image search: [Google]

36KB, 571x556px

>>61327964
Oh boy no, 10nm from Intel is hot turd, since it's so late.
7nm LP is both denser, offers 6T SRAM AND has lower fmin/higher fmax on top of it.

>>61327918
Not only that, but turning L3 into a victim cache non inclusive, all while turning L2 slower and L3 smaller and even more slower. Then intel go cry about ryzen L3.

I dont crossed the numbers in a latency test, but i bet TR/Epyc will end up with better L3 latency. Intel L2 revamp move was certainly bold but it seen a shitty move.

>>61327962
fugg that can't be real.

>>61328007
>lets copy bulldozer cache
>okay brian!

File: 1496469199721.jpg (181KB, 563x821px) Image search: [Google]

181KB, 563x821px

7NM 6 CORE CCX

>>61327999
Yea they are almost same size, however intel is very late to the game, they cant manage to produce it.
If they manage to deliver to desktop/servers in 2018 intel is totally fucked.

>>61328035
8700k will be 6 cores ˙ ͜>˙

>>61328054
Zen2 is H2 2018, and you only need 4 Zeppelin2.0 for one EPYC2.
AMD has no
>muh low yields gibe me money goyim
problem Intel hits.

>>61327507
>desktop die
kek, Ryzen is a desktop version of an enterprise die, everyone knew that from the start.
>disjointed 8 x 8MB L3 Cache
it's more like a disjointed 4 x 16MB Cache, but Intel's L3$ is tied to each core and linked via a near similar "mesh" bus concept as AMD...
>Bisectional Bandwidth
Intel's got the advantage there
>2 channels per die
But there's still 4 dies, all of which are linked via IF, which have similar latencies as memory linked via Intel's Bingbus
>1 DPC DDR4-2667
That's actually true, though
>DRAM capacity
At least they're not lying there
>PCIe
WEEW LAAD
>1S 48+20 lanes = 68 lanes*
But 20 of those lanes are from the chipset, which is linked to the CPU with what is basically four PCIe Gen 3 lanes with added features.
So in reality, it's more like 48 + 4 (up to 20, but not at the same time)
>2S 96 + 20 lanes
I forgot how the chipset QPI works in dual socket motherboards, but don't they share the same total bandwidth as DMI 3.0? So we're still dealing with 4 lanes from EITHER CPUs to the chipset, and they're SHARED.
>Implementing Ethernet and SATA further reduces PCIe lane count
Like it does on Intel. Neat.

>>61327850
FUCKING THIS.

Seriously, I'm a raging, rabid AMD fanboy and I'll even happily defend bulldozer as a worthwhile product on launch but STOP IT WITH THE UNREALISTIC EXPECTATIONS. Cautiously optimistic is the way to go.

>>61328061
2700 will be 12 cores :^)
2600 will be 8-10 cores

>>61326291
same here, no regrets on how much I paid for literally any part of my computer other than the nvme, but even then, the sata alternatives were as much or ball park so not to much regret.

https://youtu.be/B65uPAFGse8

>>61328095
No, I have a feeling that AMD will change the numbering scheme a bit
>R7 2900 - 12 cores (6 per CCX)
>R7 2800 - 10 cores (5 per CCX)
>R5 2700 - 8 cores (4 per CCX)
>R5 2600 - 6 cores (3 per CCX)
>will they even bother with a 4 core at this stage?

>>61328177
>will they even bother with a 4 core at this stage?
Yes, for mobile i.e. laptops.

>>61328190
Why tho?
7nm LP allows you to throw moar coarz into same power envelope.

>>61328199
Think of 80 to 100 dollar shit you can throw at Africa or elementary schools.

>>61328212
Binned RR successor is exactly that.
One 6goarz CCX and Navi-based iGPU.

>>61327192
Entirely depends on binning, do you believe the 1800x is the highest binned cpu or do you think amd is holding higher binnings back for the server/threadripper market?

>>61328177
Where does TR fit into all of this then?

>>61328229
Servers definely get the top crop.

>>61328247
if that is the case, then amd can likely undervolt the chips far further then desktop and use less power to reach same speeds.

>>61328241
24 goarz, 20 goarz, 16 and 12 goarz models?

>>61328054
Logic cell size is smaller for GloFo's 7nm DUV. Z height is lower, so total area is lower.
The 7nm EUV process will further reduce area as well. Intel's 10nm process is quite far behind. Their 7nm node isn't going to bear fruit until 2020 or later.

File: coffee lake wow it's fucking nothing.png (192KB, 1920x1080px) Image search: [Google]

192KB, 1920x1080px

>>61327964
>decent 6 cores coffee lake compatible with socket
Not likely on either account. Keep in mind that for their mainstream platform they have to keep it under 95 watts, which severely limits their ability to crank up the clocks. You'd have to delid and hope for the best.

>>61322443
>icc actually compiles code that runs fast on amd
Inter was too confident.

>>61328288
They are not insane enough to get the second lawsuit right fucking now.

>>61327962

It is because that SKU supports 8-sockets though but even that said it is still a piss-poor value.

A single Epyc would outperform 8 of those 8S quad-core chips with a faction of the platform cost.

>>61327962
Can someone explain what the advantage of 8 4 core CPU's would be? Why not just 4 8 cores, or 2 16 cores?

>>61328380
Software with per-core licenses. Also L3-sensetive software (but SKL-X has shit for L3, why not opt for BDW-E or EPYC 8core SKU?).

>>61328380

There's no advantage at all. You need larger boards, more cooling, higher chassis etc.

Eight-socket system only make sense in HPC but would you be using massive core count chips not puny quad-core non-sense.

No, costumer, you don't understand! I NEED a 60% margin!

>>61324766
I want amd to bow out for 5 years or so just because fuck everyone who only considered nvidia from the 4000-200 gpus,

>>61328463
And I need 16 DIMMs per socket.
Guess what would I choose?

>>61325684
well, they sell GPUs like pancakes to shitcoin miners
:^)

File: SS_2017-07-11_52.png (38KB, 789x492px) Image search: [Google]

38KB, 789x492px

>>61328491
NOOOOOOOOOOOOOO MUH GAYMES

>>61323068

AMD made the right choice dumping their fab, now they can just go to the chinks where Intel is stuck with their own process.

>>61323013

Source on AMD being botnet. AMD don't have RFID in package, don't have DLC, don't let passwordless remote login.

>>61328247
Ryzen Pro are the top crops, actually. Threadripper gets the same behavior silicon as Ryzen, and EPYc gets the same silicon as Ryzen Pro.
I think they're binned differently than what we're used to (longevity versus voltage/stable frequency). So what we consider to be a high-overclocking silicon would not be desirable at all for either Ryzen Pro or EPYC because the silicon/gold traces' durability is lower than what they'd like. Just look at the warranty periods between Ryzen and Ryzen Pro.

>>61328573
>he doesn't know about the PSP

>>61328596
>blindly regurgitating things you know literally nothing about
Keep on trying, shitposter.
The only criticism of AMD's use of an ARM TrustZone coprocessor is "Well, it could be bad, maybe, in theory!" There is nothing to hold against them because its fundamentally different from intel's directly internet connected Hypervisor. AMD needs 3rd party drivers/firmware and software to allow remote management because they're using a 3rd party solution. If you don't want to utilize it, then you don't have to utilize it. There is no remote management functionality unless you put everything in place. Intel's solution is self contained and can be silently enabled right out of the box.

The PSP isn't even turned on in AMD's consumer chips, only Pro models. Its not like a KNOX device where you can just download the KNOX software and start using it. I can't utilize the PSP on my Carrizo system at all, and the same is true of all their newer CPUs as well.

File: 1445721726647.gif (2MB, 250x172px) Image search: [Google]

2MB, 250x172px

>>61328660
>intel's directly internet connected Hypervisor.
wat

>>61328693
Intel's coprocessor for remote management uses in house solutions for everything, and everything the coprocessor needs to have network access is all right there. The difference between having an active listening device in your system and having it segregated from the internet is only a BIOS setting.

>>61328741
That's not what a hypervisor is, you idiot.
You're talking about Intel's ME and AMT/vPRO capabilities, which are physical components on the motherboard and the CPU.

>>61328798
>I don't understand what any of these things actually do
The coprocess is a hypervisor, just like TrustZone coprocessors are. They are capable of secure execution, or execution prevention.

>>61328693
he's a retard with how he explains shit but the basis is that intel's security controller connects to the internet if you are whether you like it or not.
Even shit like qubes OS where your OS itself doesn't have internet connection but rather a vm inside it wont stop it from connecting

>>61328840
>that intel's security controller connects to the internet
>security controller
>a physical component within Intel's chipset
And that's not a fucking hypervisor

>>61328857
where did I call it a hypervisor you turbo sperg?

>>61328878
Not you, this moron >>61328660
>>61328813
A co-processor is NOT a hypervisor

>>61328904
These coprocessors are in fact Hypervisors. They can control execution, they're Hypervisors.

Just fuck already.

File: 7672_PIA21327.jpg (285KB, 1020x902px) Image search: [Google]

285KB, 1020x902px

> It will happen when the weather cools

>>61328255
It was probably not designed to reach more than 4.1 unless insane voltages.

>>61327507

Can't seem to find the source on this image anywhere. Mind posting a link to the marketing slide?

File: 1499822169843.jpg (155KB, 1020x902px) Image search: [Google]

155KB, 1020x902px

>>61329039
/pol/ is always right

>>61328273
I dont care about power envelope when i can delid and overclock the shit of it, as long as the silicon can hold higher frequencies, in case 14nm+ can very well with reasonable voltages.

It will be cheaper than purchase another full TR/SLX system while i expect next year decent upgrades

File: 2bqUm9L.gif (1MB, 400x225px) Image search: [Google]

1MB, 400x225px

>>61329084
>highly binned 6 core CPU
>mainstream platform
Pick one.

>>61328199
Even more efficient / lower power usage.
Acutel and whatnot will buy them up like crazy.

>>61329107
They have to make a new die for this shit, so its probably getting the same 7700K treatment, ie, top bin of the mill for the specific die. Worse goes as i5 coffee lake without HT.

>>61328857
>it can shut down or start the computer remotely
>it can monitor all things that the computer does by setting up a KVM without the user's knowledge
>b-but it's not a hypervisor because I said so
It's shit. It will continue to be shit. Especially if you can access it without a password.

>>61329132
Is this why it keeps getting delayed?

>>61327507
> 8c Zeppelin is a *desktop* die

If 8c is lowly desktop territory, Intel, what level of garbage have you been selling us in the i3/i5/i7 series for the last 10 years?

>>61329310
Goyim Gruel tier CPUs.

>>61325704
I think GitHub did this and ended up only hiring white dudes because it turns out they tend to be the most qualified.

>>61329367
Well, then the issue is somewhere up the chain. Either colleges are fucking it up, or high schools, or middle schools, or the kid's parents screw their kid's future.

Theoretically, the necessary skills for software development are achievable by anybody. It's basic fucking logic. You can google 99% of it. The most complex part of it is the trig that you do when doing 3D vidya and physics and most game engines do that for you. If a NEET like me who failed at college and struggles to get IT certs can put together a basic game, then the rest of humanity doesn't have an excuse for webdev and server work.

>>61327507
>repurposed desktop dies
Actually, desktop Ryzen uses server dies with some hardware disabled.

>>61322593
hmmmm really makes me think

>>61328288
No, AMD just designed a CPU to run super fast on ICC. Now Intel is fucked, they'll have to gimp their own CPUs!

File: 1496480240772.png (673KB, 4500x4648px) Image search: [Google]

673KB, 4500x4648px

>>61325864
This is the second time I've seen someone gloating about making this meme picture and posting a higher res.

Shut the fuck up about it.

>>61322040
>process node advantage
It's not advantage in low-clocked server/mobile chips, only in xXx_ProGaymuRR_xXx desktop ones, like 7700k

>>61324818
>POWER9
> power efficient

>>61322522
oh my

>>61327507
why would you ever try lying to actual technology people that make their bread from server efficiency?
One or two would switch to amd just because of insult.

File: 20170712_075745.jpg (2MB, 3264x2448px) Image search: [Google]

2MB, 3264x2448px

>>61324592
>plant
try plants. two plants and 4 research centers. source: am israeli

>>61329055
https://www.computerbase.de/bildstrecke/78687/

>>61330810
>bigher power does not require bigger power for cooling
>electricity is free for servers

File: 1468771937055.png (134KB, 1500x844px) Image search: [Google]

134KB, 1500x844px

is this a real thing? I'm doubting my sanity here.

>>61327679
>higher is better
every fucking time, you fucking burgersharts can't even read the simplest charts

>>61331618
It's total bullshit because typical server workloads deal with multisocket well, multidie is not a problem.

>>61322040
how much does EPYC cost? somehow i don't think the average /g/tard will be running dual EPYC (what a silly name anyway), and it would only be useful for multithreaded tasks, for general desktop use you're still better off with a higher clocked lower thread count cpu

>>61331553
we need proofs
post nose

>>61322761
The 16c pricing on Epyc goes down to 650 USD, 8c is below 500 USD. It's in the same bracket as an enthusiast part. It's absurd to compare these to the exclusivity of an exotic car.

>>61331648
no, I mean did intel actually made those slides?
Why are they shooting themselves in the leg with it?

File: spurdoEBYN.jpg (1MB, 2048x1536px) Image search: [Google]

1MB, 2048x1536px

>>61326556

>>61331658
to be fair, enterprise motherboards are expensive.

>>61331656
8 core variant will cost 400-500$ and will have all the uncores of 32c ones (128 pci lanes, 8 channel RAM, FUCKHUGE cache, 8mb of l3 per core)

>>61331656
The top of the line SKU (32c/64t everything enabled) is $4200.

Wikipedia has the lineup with pricing: https://en.wikipedia.org/wiki/Epyc

>>61331683
>everything enabled
fun fact: all EPYCs have full feature list enabled.

>>61322040
Take your medicine you Jew fucks. I hope Intel loses billions.

>>61331691
the P-suffixed models are 1 socket only though.

File: Image 001.png (30KB, 712x449px) Image search: [Google]

30KB, 712x449px

>>61327679
U have the flags (pic) and link to test's source code
https://www.cs.virginia.edu/stream/

>>61331661
Yeah, they're real. It's pretty funny and sad at the same time.

>>61331661
I can't believe these are real marketing slides produced by Intel. If so was a sales rep i'd be ashamed to present them. If I was a CTO being presented these I'd be unable to contain my laughter.

What proof do we have that these are real?

>>61331761
The "I can't believe how stupid this shit is" slides are mixed in with the other more normal ones. >>61331567

>>61331761
https://www.computerbase.de/bildstrecke/78687/

>>61331779
So some of the slides are joke slides? I'm being serious. Intel is a multibillion dollar company. I'm in sales and if asked to present these slides to a CTO I'd refuse to. You can't present these without losing all your credibility.

>>61331821
It really does seem too dumb to be true. I don't see any evidence that someone slipped in a bunch of fake slides to make Intel look like stupid assholes.

>>61331821
No, these are real, 100% pure slides made by H1Btel.

Wow it's almost as if intel is doing exactly what I predicted: Shit on AMD for not having support or being reliable.

Which is both nonsense but intel hasn't told the truth for decades.

>>61329468
>Well, then the issue is all of this bullshit that deflects from reality
Whites and Asians as an aggregate are the smartest and most technically skilled/competent people on the planet, and women are in fact not as capable as men.

It's simple, it's scientifically provable, and true.

This is literally worse than Lavalake-X launch.

>>61329468
>Theoretically, the necessary skills for software development are achievable by anybody. It's basic fucking logic. You can google 99% of it.
t. non-programmer

File: microexpressions-disgust.jpg (28KB, 524x336px) Image search: [Google]

28KB, 524x336px

>>61331850
>>61331863
Having worked in both marketing and sales I can honestly say I will never purchase an Intel product again . My disgust is now eternal. These abominations are a disgrace to marketing and cruel to their salesforce. As soon as my 4770 shows any sign of needing an upgrade Intel can fuck right off forever.

>>61331953
That's H1Btel for you.

someone needs to post the statement of intel after their epyc annc

literally /g/ tier shit writter on it

Fuck, how could they tell?

>>61332020
BATTLE TESTED

>>61332020
http://www.barrons.com/articles/amd-reveals-epyc-details-intel-vows-to-top-it-1497997334

I guess the whole "Epyc" name rubbed the salt in the wounds a little too deep.

>>61324788
{Needs citation}
>Inb4 Intel sponsored test

>>61322522
RIP Intel.

File: amd-s-veteran-cpu-developer-jim-keller-leaves-the-company-492160-2.jpg (25KB, 570x340px) Image search: [Google]

25KB, 570x340px

How does he keep getting away with it?

Intel BLOWN THE FUCK OUT
L
O
W
N

T
H
E

F
U
C
K

O
U
T
AHAHAHA THERE IS LITERALLY NO WAY TO SUPPORT INTEL WITHOUT LOOKING LIKE A COMPLETE RETARD RIGHT NOW
Intelfags must be curled up crying in a corner because there is NO actual way TO SUPPORT INTEL AAHAHAHAAAA goddamn. Will intelfags ever receive acknowledement again?

that's a big chip

>>61322522
2xEPYC is the price of a single Xeon. lel

>>61332351

This man is a fucking magician. Helping on ZEN, giving AMD basically 1 arc to do everything, super scalable, high yield, higher perf/watt than Intel's offerings and literally better in every single way.

>>61322603

Those price comparisons, holy fuck. Intel is over.

File: 1468475454256.png (228KB, 590x387px) Image search: [Google]

228KB, 590x387px

>>61322040

>>61332454
>>61332411
that just means intel made a shit ton of money while amd had nothing, amd is too late with these to make much off of it, same story with their gpus

>>61332550
damage control intensifies

GloFo's "7nm" is 8.2nm while intels "7nm" is 6.7nm

But no one really cares.

No one really cares that the current 14nm finfets are fucking 16nm with 22mm characteristics either.

>>61332605
Oh look, its another IDF fanboy blindly talking out of his ass.

>>61332605
Intel's 7nm is so far away that GloFo will be fabbing 5nm GAAFETS by then.

>>61332605
Gloflo's 7nm is 1.5 years away. Intel's 7nm is 3-4 years away.

>>61332770
1.5 years away at worst case scenario*

Does anyone honestly think that the diversity meme is killing companies? Only thing Apple does now is fucking gay health stat apps, le useless thin watch. They no longer invest in come up with breakthrough things like the iphone announcement, unibody aluminum laptops, retina screen macbooks. Even now that quality is still unmatched. OS X was pretty good until after mavericks. They are just no longer investing in developing OS X, even that apple file system was delayed like 2 years. Because they no longer have the core brain power, they are replaced with diverse people.

Now we have homo touch bar, whoa what an innovation. I really wonder whose idea was that.

Same goes for intel, they only care about hiring diverse people, that's it.

File: 6566.jpg (70KB, 330x319px) Image search: [Google]

70KB, 330x319px

>>61332605

>J-JUST W-WAAAAIT

>>61332605
Also
>BEOL is everything
the fuck is that meme?

>>61332820
Apple's problem is and was Steve Jobs. Steve Jobs didn't run a tech company, he ran a fashion company that sold tech. He was militant, overbearing, a perfectionist. He berated people for their failures, and his direct charges were afraid of failing because of it. Henry Ford operated the same way.

In his absence there is only a company riding on the wave of being fashionable. They're no longer pushing their own paradigm, there is no demand for perfection in form and aesthetic. There are people filling out the shell and maintaining the image. They were the perfect example of well structured authoritarian efficiency.

Diversity at its core is nothing but PR. A company ran by competent people only wants to be as vibrant as necessary to win social brownie points with lefties. A profit driven entity still wants profits above all else. Apple is one of the most white companies in Silicon Valley for what its worth. The upper management is all goys.

Can Intel please stop with this
>BEOL is everything
meme?
Their 10nm node is an abortion performance-wise.

>>61332947
what is BEOL again?

>>61332947
>b-b-b-but density!
Its just desperate fanboyism. There is one super pervasive autist on the internet who spams this everywhere. He goes by the name Weitkin. Hes a lifelong forum dwelling degenerate on Anandtech, and I never thought much of it, but I've seen him spamming all around the web. Hes in the comments on EEtimes, siliconsemicon, RWT, and a dozen other sites. I'm sure he posts the same nonsense here.
He acts like defending intel is his actual job, and I'm starting to think it might really be.

>>61322470
if you don't game it's a fine cpu

>>61332961
Back End Of Line.
The underside metal layers of a chip that include interconnects, basically everything important aside from the actual switching logic and memory cells on the front side.

>>61332976
For that matter intel's 22nm process was wholly lacking in density, by a great margin to industry 20nm nodes.
The narrative constantly changes. Fanboys always zealously defend their favored brand no matter what.

>>61328199
Not every laptop is a 35-45W piece. I'm pretty certain AMD wants a decent single-digit-wattage mobile/embedded CPU.

http://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade/16


Shit, what a comeback.

What's gonna happen when Zen2 comes out and most software is fixed?

>>61333008
But even for gaming it's bretty gud.
>>61333247
So 7nm they should aim at 7w?

>>61333265
I presume AMD will buff up their AVX implementation aswell. From a half-AVX256 to a half-AVX512.

>>61333305
amd doesn't need memex, they have real gpus for that kind of computing workloads

>>61333305
AMD has no reason to support 512bit AVX instructions. Intel is only pushing this because they are selling nonGPU vector accelerators in the HPC market.
AMD offering high 512bit throughput would legitimize intel's push, and help further adoption of Xeon Phi coprocessors.

AMD does not want this. AMD wants to keep selling their CPUs and GPUs. They want GPUs to be the primary HPC number cruncher.

>biggest launch of server gear since SB xeons
>stock barely moved

Those slides don't inspire shareholder confidence

>>61324317
The only reason AMD is here today is their GPU division acquiring them multiple console contracts for their semi custom business, keeping the company afloat while they were developing zen.

>>61333332
and after epyc launch amd's stock shot up 15% in 2 days

>>61333342
And AMD will continue to have a strong grip over Sony and Microsoft, because all other options for cheap custom APUs by the console companies have been exhausted.

>>61333342
Yeah, and now they should stop wasting money on consumer GPUs, double down on pro GPUs and a full CPU lineup in every segment from fanless to Sahara-approved

>>61328596

>he doesn't know what he's talking about

>>61330723
This looks blurry and upscaled.

>>61333865
some retard probably ran it through waifu2x for no real reason