How can Intel even compete?

>>61034147
By cranking up their fabs, expanding their current platforms and taking a hit of their margins.

They still have the superior product, they are just jews about giving good value.

>>61034284
They already said they would be willing to sacrifice marketshare to keep their 60% margins.

>They still have the superior product
I can't think of a single thing Xeons have over Epyc.

>>61034147
> $3400
Alright /g/, what Xeon is competing at that price range?

>>61034284
>By cranking up their fabs,
Thats not happening. All their future product lines heavily depend on their 10nm line, and the process is still horrible. If they weren't having massive problems with it then we'd already have Cannonlake mobile parts. Instead what happened is that Cannonlake server parts got taken off their road map.
Intel is in a bad position right now.

>>61034313
m-muh $9000 e7s

Man why even bother with these threads? None of you losers are ever gonna get your greasy fat fingers on one of these, ever.

>>61034339
Reminder:
It is 2H 2017 and we have no 10nm parts from intel.
Intel's 7nm foundry building is still under construction.
Their foundry lead is gone.

By not exclusivly selling to green people?

>>61034375
kek

>>61034357
It's /g/ - Technology, not /g/ - Technology I Can Personally Afford.

>>61034357
>Why would you have threads about supercars on /o/
>Why would you have threads about tanks on /k/
>Why would you have threads about Half Life 3 on /v/

>>61034284

Is it superior?

It's a niche product in the datacenter and cloud server markets. Really the only thing it absolutely bests AMD at this point, is memory latency, AVX512 and absolute single core performance.

10nm will help coral the heat and energy, when, if, it ever arrives. But Zen 2 is already at 7nm moving on to 7nm+ the year after that completely decimating the clockspeed advantage while still offering better performance per watt.

Then there is the yield issues. AMD is at 90+% of every wafer they stamp out the door. Intel is CONSIDERABLY less. It is superior in a niche like Bulldozer was.

>>61034440
>>Why would you have threads about Half Life 3 on /v/
Too far dude, too far.

>>61034362
10nm just began with the 835

Everything will be delayed

This is the last hurrah for silicon designers

I suspect 10nm will get into more mobile SoC, then ULV like Core M with broadwell, then BGA, and trickle down to HEDT

>>61034440
Dammit I want a tank

And no, HL3 is never ever happening

>>61034313
single thread performance
if you have a sql server intel is still the person to go with.
the E5 2699 wrecks it in a lot of tests
and finally some of instruction sets differ like avx

not much else however, ryzen has the i/o and value proposition.
>>61034339
they are still the fab leaders by a good margin, 2-3 years ahead of anyone else.
the processes from GF, Samsung and TSMC are all very different too which makes comparison difficult.

>>61034449
>But Zen 2 is already at 7nm
Calm down anon, who do you think is going to fab 7nm chips for them?

>>61034313
the naming is better, I guess

>>61034570
I don't know... Glofo? They're even being helped by IBM

>>61034556
At the same clock speed AMD is on par with Intel, so in the server space Intel doesn't have the single core advantage because AMD's parts are not limited by thermals. You are correct that some parts with AVX512 will have an advantage in specific workloads but AVX512 adoption is still in early stages.

>>61034595
AVX512 is a total fucking meme, the only good thing about it is moar registers.
t. Linus B Torwalds

>>61034570
GloFo has 7nm DUV on the map for Q4 17. They're supposed to be fabbing 7nm products next year.

File: 1498068938542.jpg (17KB, 233x217px) Image search: [Google]

17KB, 233x217px

reminder Ryzen gets analrayped by a 5yo i5 3570k in gaymen:
http://www.pcworld.com/article/3176100/computers/amd-ryzen-7-1700-vs-a-5-year-old-gaming-pc-or-why-you-should-never-preorder.html

>>61034556
>they are still the fab leaders by a good margin
Not even remotely close. Not even in the right ballpark, pal.
Intel's 14nm Trigate process brought them nothing but constant issues and poor yields which is why it took them so long to start producing volume levels of high TDP chips. Samsung had their 14nm LPE lines producing chips by the tens of millions inside of 6 months. Back end area scaling metrics mean nothing for the switching performance of the front end of line.
Intel doesn't have anything going for them, especially since Global Foundries now has the full wealth of IBM's process IP.

>>61034642
>muh gaymen
last thing to resort to heh?

>>61034642
>old gayman
>gayman
Who cares?

>>61034643
This, Intel's insane boner for BEOL scaling to preserve their fucking margins is killing them with 10nm.

>>61034449
Do you have any clue what the fuck you're talking about?

From 14nm and under you don't just skip a process.

The 835 is 10nm, but we know nothing about its reliability. We might see an A11 chip take on 10nm later this year. That would fund further fabrication facilities at 10nm.

But 7nm? First you need an ULV in 10nm. Then you need some lower end GPU, maybe even mobile APU. Then we finally get to BGA. After that, of the process is going well, you start seeing socketed CPU. In 10nm, not 7nm.

7nm is at least 5 years off, more like 7. 14nm began in late 2014 and we still haven't seen anything Vega, and that's 14nm.

Stop smoking shit, and take a puff of this

File: ayymd.png (126KB, 640x360px) Image search: [Google]

126KB, 640x360px

>>61034643
>the processes from GF, Samsung and TSMC are all very different too which makes comparison difficult.

>>61034688
>7nm is at least 5 years off, more like 7.
*laughs in Nazi science*

>>61034693
Anon, do try to pay attention here.
>Back end area scaling metrics mean nothing for the switching performance of the front end of line.

Intel has nothing going for them. They're behind the curve.

>>61034688
Mass production of 7nm wafers is on track. q2-q3 2018.

File: 1496444107365.png (123KB, 240x286px) Image search: [Google]

123KB, 240x286px

>>61034711
You're actually retarded if you think intel is behind, that's all I'm going to say.

bring yourself up to date, anon.

>>61034706
>>61034718
and IBM is already fucking around with a 5nm node

>>61034718
Mass production of WHAT? Vapor?

It's been pushed to H2 2018, and nobody has even mentioned WHAT is being made first. Guess what folks? Nothing at all! You won't see a god damn thing next year in 7nm, guaranfuckingteed.

File: wow.png (181KB, 1476x1394px) Image search: [Google]

181KB, 1476x1394px

Is pic related the most threads Intel can deliver?

>>61034748
Ahh but shit, I guess some guy holding a wafer that may as well be a rock, IS THE NEXT BEST THING

MOAR COOOOOOORS

>>61034741
IBM revenges. Intel fucked IBM supplying 386 to everyone.

>>61034755
I believe they got an 28 core somewhere, but i'm not sure

>>61034588
Neither is as good as Opteron.
Xeon is a fucking gas.

>>61034755
28 core was paper launched. 32 core coming next year. (48 core Epyc Starship is also coming next year)

>>61034774
20k usd with <1% yields.

>>61034796
>48core on 7nm about a year away
Anyone who doesn't buy AMD stock now (or on the likely dip on Q2 earning report) is a braindead moron.

>>61034755
I believe they have a 28 core but it's like $9k.

>>61034738
IBM is a company with countless millions of dollars to fuck around with and they spend stupid amounts of it on R&D they may or may not ever use for the purpose of building a base of patents. GloFo has access to all of IBM's silicon IP now and they're making use of it by getting the jump on Intel. You're right, their 7nm isn't actually 7nm by Intel's standards, but it is very close to what Intel is calling 10nm AND they're going to use finFET transistors, which puts performance somewhere in the ballpark of 7nm. Intel has been fucking around on 14nm now for too long, and they're about to lose out on the jump to 10nm because of it.

>>61034817
and when AMD launches 7nm cpus, intel will make a statement saying "It's not a real 7nm cpu"

>>61034865
>muh BEOL scaling
They can fuck off then.

https://en.m.wikipedia.org/wiki/14_nanometer

Read the history. Do you know how fucking long it takes to go from 'taping out' to in your hands?

A long fucking time! Read!

>>61034887
holy shit man, stops being so asshurt because we'll see 7nm cpus in 1.5 years

>>61034901
1 year from now. Samsung and IBM assraping intel.

>>61034147
if zen yields are so great then why are the clocks so shit?

>>61034944
things that are not related: the post

>>61034960
Its perfectly related. If Intel settled for shit clocks then they would have great yields too.

>>61034323
lots of them
https://ark.intel.com/products/93806/Intel-Xeon-Processor-E7-4850-v4-40M-Cache-2_10-GHz


https://ark.intel.com/products/93793/Intel-Xeon-Processor-E7-8860-v4-45M-Cache-2_20-GHz


inbetween those two

>>61034865
GloFo's 7nm process is denser than intel's 10nm, and is built on IP far more advanced than intel's 3rd gen Trigates.

>>61034944
Are they shit?
Hitting 3.5ghz below 1.1v isn't bad. A characteristic of the process is higher voltage with lower amps. The only drawback of Samsung's 14nm LPP is limited clocks above 4ghz.
Yields and clocks are not the same thing, anon.
GloFo's 7nm node targets a drastically higher fmax.

File: 5453235235235.png (26KB, 711x533px) Image search: [Google]

26KB, 711x533px

>>61034944
Because amd doesn't want to burn your house down.

>>61034990
https://ark.intel.com/products/93809/Intel-Xeon-Processor-E5-4650-v4-35M-Cache-2_20-GHz

none can compete intel is dead

>>61034545
A tank isn't *that* expensive. If you got a good job and devoted yourself to saving up you could afford one.

>>61034357
fuck off kid I'm gonna buy one just to piss you off

>>61035014
It's impossible to buy modern tanks tho

why do I have a feeling that that guy talking about how 7nm is 7 years away is making an enormous post to prove he's right?

>>61035129
o boy I guess I better stay in this thread

>>61034997
https://www.youtube.com/watch?v=r32LcBqiv7I

File: time.gif (1MB, 626x462px) Image search: [Google]

1MB, 626x462px

>>61034147
THIRTY
TWO
FUCKING
CORES

>>61035192
48 cores end of next year

I'm confused, what does the 1 socket price mean? Why is it cheaper than a 2 socket price?

Anyway it's still cheaper than hot intel trash.

>>61034545
YOU SHUT YOUR WHORE MOUTH

>>61034147
i don't think you can recover from green skin. looks pretty serious,

>>61035228
those 3 processors only work on single socket configurations

AMD Promethean(24c/48T 6 core per CCX) in the x399 and AMD Pharos(48C/96T) in the servers.

File: 1411177440158.png (68KB, 500x500px) Image search: [Google]

68KB, 500x500px

>>61035168

kek'd

>>61035264
those names look badass

but performance is still king

>>61035264
Another leak? WTF.

>>61034944
Fucking hell mang.
>if i like vanilla ice cream why dogs bark?
But you did make me laugh so thank you.

>>61035284
AMD Kyzen(zen2 IPC+new process)

>>61035276
Could you please avoid the antisemetic imagery?
I know it's a joke an all, but it's really offensive towards posters of the judaic faith.

Thanks in advance, hope it doesn't come to me having to report you.

Regards,

>>61035339
Now with IBM's nazi space magic 4 way SMT!

>>61034440
>Why would you have threads about supercars on /o/
You don't. It will just get shitposted to death by benchracers.

>>61034796
xenon is a gas
even if it was called xenon, that's actually a pretty good name, since it shines when energized and it's a noble gas making it very stable

>>61034357
Dumb nigger, I'm buying the 7401P as soon as it releases to replace my dual Xeon E5-2687W server

>>61035040
Stupid nigger, an M60 export Patton for $65K is good enough
>b-b-but muh 120mm smoothbore
Fuck off, nigger. You can't even afford the ammunition and propellant charges. If it makes you feel any better, you can always cram ~15 .22LR barrels into the main gun and pretend you're using scatter shots.

>>61035520
Is this an allegory to the "you only need 4 cores goy"?

so what xeon should I look out for on ebay when AMD destroys intel and all of the chips are sold for 1% the retail price on ebay?

I already see some 18 cores on ebay for $250

>>61035531
No, it's an allegory for "quit bitching about not being able to afford a $8K 28-core Xeon and buy a $500 12-core Threadripper instead"

>>61034981
No, they wouldn't?
Do you have any idea what you're talking about?

>>61035595
Haslel-EP and their E5-2690W v3

>>61035352
out

>>61035604
>2690W
damn that does look like a nice core

is the v4 ok too?
https://ark.intel.com/products/91770/Intel-Xeon-Processor-E5-2690-v4-35M-Cache-2_60-GHz

I don't see any W in any of their names

>>61035631
v4 is Broadlel
v3 is Haslel
W means Housefire

>>61035642
o damn
idk if I want a housefire even though my server-rack is fire proof

>>61035658
If your 2011v3 motherboard is rated to handle 160W TDP CPUs, then you're fine with using W SKUs.

>>61035667
no mobo
was gonna replace my 7 year old server when everything drops and is being thrown away on ebay
only thing I'll keep is probably my seasonic 430w dual rail dual cpu connection PSU that is invulnerable

>>61035279
>but performance is still king
Yeah, about that .. >>61031565

>>61035693
I know anon, I know, I'm just saying that AMD wins even if they have a shitty corny name, given good performance too

>>61034147
Xeon price drops when?

>>61035849
when intel files for bankruptcy, because those shareholders literally obligate intel on giving 60% margins

but you'll probably see some cheap xeons on ebay after companies dump them out for epyc

>>61035427
You're right. I'm dumb.

File: 1496678794383.jpg (13KB, 255x255px) Image search: [Google]

13KB, 255x255px

>>61035352
Sure!

>>61035849
Never.
Intel will sell via marketing and underhanded tactics. Investors and Intel want their 60%+ margins to maintain brand image even at the cost of revenue.
Also because they know if they offered reasonable value, then if they managed to make a better arch and process they'd then be expected to continue to offer better value to compete with the past chips.

>>61034147
>16-cores starting at $650/$750
What are your predictions for Threadripper prices?

>>61036152
>underhanded tactics
FTC is watching this time, and AMD clearly has OEMs, partners and vendors on their side.

>>61036168
About $3.50

>>61034147
I am an intel guy but honestly they might have killed themselves.

Refusal for years to increase cores-no experience in high multicore manufacture
Refusal for years to deal with rising tdp-bad thermal margins
Trying to cheap out like kikes-bad thermal paste shit tier design having to delid 1500 dollar chips
Cant keep up with where high multicore chips are now-rebranded i9 trash


Intel fucked itself over. Too much greed too much arrogance.

When threadripper drops it may well destroy them. Very few people are interested in buying the new intel chips at their msrp. They are just too shitty and too expensive especially with thermal issues.


Intel is fucked.

>>61036236
You forgot to mention seperating market brackets in the most obnoxious way possible to ensure nobody can get exactly what they want without also getting something else they don't need for a higher price.

>>61034284
>They still have the superior product

Epyc obsoletes Broadwell-E overnight for nearly everything, and it's actually even better for most non-HPC server use cases than Skylake-E.

AVX512 support means needing 128B datapaths to L1 and L2 cache, and that shit sucks down power fast, even if you're only doing non-intensive int workloads.

Furthermore, Intel can't sell all that many 30+ core Xeon chips thanks to imperfect manufacturing yields and needing to fuse bad cores off, whereas AMD gets 80+% of the tiny Zeppelin dies fully intact and can just glue the little fuckers together with a 99%+ success rate into as many 32c Epycs as the market feels like buying.

AMD will be able to sell millions of 2*32c 2S servers, whereas no amount of "cranking up the (existing) fabs" would get Intel half as many of their top core count chips out the door.

>>61036316
You could have just said Phenom < Core 2 quad

10 years later

>>61035168
oh shit this was in the MK2 movie im sure

>>61036316
Almost everything (absolutely everything?) done with AVX512 should be done on a fucking GPU instead.

>>61034642
>day 0 benchmarks
>relevant today
Surely you can't be serious?

>>61036573
Seeing as the image is of a certain crazy person who publically assaulted someone and lied about it, then was subsequently arrested, I'd like to think it's all an elaborate ruse. But you can never tell who's trolling and actually retarded these days.

>>61034570
Have you been in a cave for the past six months? GloFo is on a roll, and Intel has been stalled for years.

>>61034147
They can't compete with AMD's BBC sized cores.

File: jew killer.png (400KB, 816x1020px) Image search: [Google]

400KB, 816x1020px

I want to see those clockspeeds, anyone has any sauce?

Thanks Keller.

File: 1497990702380326529779.jpg (2MB, 3968x2976px) Image search: [Google]

2MB, 3968x2976px

>>61036772
>simply ebyn xDDDDDD

http://www.anandtech.com/show/11562/amd-epyc-launch-event-live-blog-starts-4pm-et-

>>61036772
the top tier one has 32c/64t, 2.2 GHz base, 2.7 GHz all core turbo, 3.2 GHz <12 core turbo

>>61036862
>32 core
>3.2 ghz boost
MUH DICK

>>61036862
why didn't they compare it with E7's though?

>>61037109
>comparing two and eight-socket systems
Wat.

>>61036456
AVX512 only exists because Intel wants to try to stave off absolute 100% GPU domination of the HPC market, not because it is actually a good idea.

>>61034313
>They already said they would be willing to sacrifice marketshare to keep their 60% margins.

They can do that. But if AMD's offerings are as good as we think they are (and even get better very soon), intel cannot do that indefinitely.
I think their strategy is right now to delay this decision until 2020 when they can come out with their new arch earliest. But in my opinion, by then they will already have had to cut into their margins

File: 2sperf.png (65KB, 620x326px) Image search: [Google]

65KB, 620x326px

>>61034323
Nothing.

>>61035355
>4 way SMT

The fuck is this?
4 threads per core? Why?

>>61035595
>>61035849
>>61035871
When does Epyc actually ship in quantity? Which datacenters/IaaS providers do you think will jump on it?

>>61034642

>old program falls back to shitty defaults when it doesn't know the new arch supports everything it wants

>>61035040
You can buy Centurions and Chieftains pretty cheap m8.
Chieftains are about £60k, though getting an engine that will work may cost nearly as much.
If you're set on a *tank* specifically, and not just an armoured vehicle than yeh, it can be alittle pricier. You can get a Scorpion for about £26k

>>61039971
>When does Epyc actually ship in quantity?
Dell, HP, and SuperMicro have 2H 2017 set out, so *around* that time more than likely.

>>61040153
fug I hope that Dell throws them in workstations.

>>61039971
>Which datacenters/IaaS providers do you think will jump on it?
1&1 and DropBox are already on board.
Microsoft and Amazon have announced support for EPYC too. Baidu are getting in on it as well

>>61040267
I thought Amazon had a massive contract with Intel that also involved some semi-custom silicon? Seems like Intel wouldn't be happy about them jumping ship.

Could EPYC bring down their rates for servers even more? as it is they are piss cheap.

>>61040295
No doubt Amazon have scrutinised the details of their contract with Intel. But I doubt Intel could afford a legal battle with Amazon.

File: POSTTHISWHEN7NMPRODUCTSCOMEOUT.png (8KB, 1200x125px) Image search: [Google]

8KB, 1200x125px

>>61034748
screencapped so I can remind you how retarded you are when we get 7nm products from AMD next year

When are they going to show off the Threadripper?

>>61035264
Aren't those called Rome and Milan (7nm and 7nm+).

>>61040691
These are codenames. He mentioned platform names, just like TR's platform is named Whitehaven.

>>61040685
They showed TR at computex. https://youtu.be/NeuamYdfU4w?t=16m

>>61034357
The cheapest ones start at 400$. With 8 memory channels and all the PCI lanes.
>MUH SERVER CHIPS MUST BE EXPENSIVE
Nope, you're thinking Intel.

>>61035264
>Promethean(24c/48T 6 core per CCX) and Pharos(48C/96T)

sauce?
and really 2x 6c CCX? please babby Jesus let this be true and not AMD taking the easy 3x 4c CCX route.

>>61041466
Going for bigger CCX is the only reasonable way.

>>61035264
>6 core per ccx
AMD is killing quad cores too.

>>61041510
might hurt yields=>more expensive
definitely not next generation

>>61041760

They are already at 90+% yields. In six months every chip sold will be nearly pure profit until they switch nodes to 7nm.

>>61036675
That guy is obviously shilling and spreading Intel FUD.

>>61041807
just let me know when they get off their asses and sell a true SoC BGA on ITX/uATX to compete with Xeon-Ds.

I can understand launching Epyc first, but for there not to even be any chatter at this point just feels weird.

Maybe the onboard 10GbE in the A/B steppings is bugged or something?

>>61041854
>Maybe the onboard 10GbE
There are none on Zepplin dies. They would need to either make a new die just for the integrated MACs, or use an external controller for 10G ports (highly likely, given how many shared PCIe lanes a 16-core can have)

File: pic3_090916.jpg (48KB, 270x220px) Image search: [Google]

48KB, 270x220px

Intel is actually moving away from the CPU market. AMD is having it's time now but this era is the end of the desktop computer and the market is going to shrink. Intel tried to move to mobile CPUs but it's pretty evident that ARM is going to dominate the mobile market fairly soon. Intel is focusing on other markets now.

When ARM takes over the mobile CPU market neither Intel or AMD are going to have the technology to compete. AMD could easily end up a niche product line for hobyists if it doesn't compete in other fields too.

Intel simply doesn't want to fully invest in CPU chips anymore. I'm not even an Intel shill, my PC is Ryzen.

>>61041913
>When ARM takes over the mobile CPU market neither Intel or AMD are going to have the technology to compete
>AMD are not going to have the technology to compete
Except that fucking wrong
https://en.wikipedia.org/wiki/AMD_K12

>>61041913
>le ARM will dominate everything meme
just no

>>61041952
It will eventually. Intel is trying to delay the inevitable with that x86 emulation lawsuit against Qualcomm and Microsoft, but it's going to happen in the mobile and portable computing segment.

File: 1492801783942.jpg (134KB, 954x605px) Image search: [Google]

134KB, 954x605px

>>61041899
there's actually 4 10GbE MACs, nigger, just not apparently used in any product so far.

the PCS is native 10GBASE-(K/L/E/S)R 64b66b compatible IIRC, so it should be possible to plug SFP+ shit virtually directly into the CPU.

>>61041952
ARM has its visions clearly set for the future. All of it's processors have always been made for mobile use on low power.

>>61041974
>just not apparently used in any product so far
...Which means that they clearly don't exist anywhere except in your head, AMDrone

File: 1494090152782.png (495KB, 1070x601px) Image search: [Google]

495KB, 1070x601px

>>61042053

>>61042072
Show me where the PCS you mentioned is indeed 10Gb capable

File: handholding for hebrew shills.png (479KB, 849x2009px) Image search: [Google]

479KB, 849x2009px

>>61042053
can you not read a diagram?

>>61042096
>not a single vendor partnered with AMD for EPYC has mentioned this capability
>AMD overpromises and underdelivers once again
Huh, really makes the nogging itch

>>61042086
the PCS is going through muxed 12Gb PHYs. what do you think?

>>61042133
That AMD is too incompetent to make 10GbE work on their SoC when Intel's been able to do this on Broadwell-D for years.

>>61042117
Hi JIDF

File: goalposts.jpg (27KB, 300x240px) Image search: [Google]

27KB, 300x240px

>>61041899
>There are none on Zepplin dies.
>>61042053
>they clearly don't exist anywhere except in your head, AMDrone

...

>>61042142
>AMD is too incompetent to make 10GbE work on their SoC


for (((you)))

>>61034147
Slightly cutting their insane margins, cranking up their fabs. The fact that Intel survived fucking Itanium means they'll survive anything.

>>61042142
Lisa's going after Xeon-D with Zeppelin, too.
Maybe not next month, maybe not even the next stepping, but it's certainly coming this product generation.

>>61039971
IBM even has 8-way SMT for both POWER8 and POWER9. That's just how they roll.

>>61041913
ARM isn't going to be competing in the high-end server market any time soon.

File: 1495723291837.png (52KB, 960x540px) Image search: [Google]

52KB, 960x540px

>>61042464
>AMD can only manage 2-way SMT, and doesn't even pull that off very well
Guess it really is shit at everything, makes one wonder why Dell, HP, and Supermicro are even bothering with AMD again

>>61042583
Microsoft and Baidu too.

>>61042583

Isn't AMD's SMT way better than Intel's HT?

>>61042607
20-30% better.

>>61042607
No, HT has had almost a decade of refinement and optimization, while AMD shat out their SMT after their last attempt with CMT was dead on arrival.
>>61042620
>he actually believes this

>>61042627
Could you please stop shitposting?

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

4KB, 452x523px

>>61042627
>hey guys intel is better. I work at intel, pls buy intel processors.

>>61042627
>No, HT has had almost a decade of refinement and optimization, while AMD shat out their SMT after their last attempt with CMT was dead on arrival.
It says a lot about how bad Intel's current crop of engineers is that something AMD just "shat out" is better than Intel's equivalent that's been polished for a decade.

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

119KB, 1000x1000px

>>61042583
>>61042627
Go away, Brian.

Hey, just a question for you guys:

Why do you constantly flame over Intel vs. AMD? Or Nvidia vs. AMD? Does it really matter who fabs the best processor? Shouldn't you just want the best value per dollar for your budget?

I get the whole "team mentality", but what do any of us have to lose from one manufacturer one-upping the other? Like, if I've been an Intel guy my whole life (for whatever reasons), and AMD puts out a CPU that has a higher performance per price mark, shouldn't I just go with that CPU since it is objectively better? What do I have to gain from sticking with Intel? I get that you guys will say shit like "muh cores" and "muh thermal footprint" and other shit. Those are, of course, valid points to make when deciding between two processors. But are they really worth having a flamewar over?

Why can't you guys just get whatever thing suits your needs and shut the fuck up about it?

>>61042788
>Why do you constantly flame over Intel vs. AMD?
Because I am paid by Intel to shit on AMD so that I can buy more shares in Qualcomm and AMD

>>61042788
>Shouldn't you just want the best value per dollar for your budget?
Yes. Which is why competition is important. So naturally a lot of people are getting excited when AMD makes return after Intel hasn't had any real competition for years.

>>61042788
>Shouldn't you just want the best value per dollar for your budget?
You should see some of the dipshits around here yelling perf/dollar doesn't matter, and neither does perf/watt. Apparently every Intel/NVidya owner is a big dick money spender that tosses thousands at PC hardware like it's a joke.

>>61042811
I wish this was why. Unfortunately, the userbase of /g/ is too inconsequential for any company to spend real astroturfing dollars here. That is better suited to something like slashdot.

>>61042822
I completely agree that competition is good. And I am very much on board with the whole concept of getting excited because AMD is starting to kick it up a bit. I guess my problem comes from the inevitable reaction to that. Why do people think a lack of competition is a good thing?

>>61042840
As a confession, in my last PC build, I went with a 5820k and two 980ti. Granted, I have a bit more money to spend than most, so I wasn't extremely particular about which specific CPU and which specific GPUs I got. But after seeing the Ryzen product launch, I'm pretty impressed by some of the benchmarks, and have been thinking that I might give AMD a try again with my next build.

There are those who would call that stupid, and offer absolutely no tangible reason why.

File: 1497381152408.jpg (48KB, 600x329px) Image search: [Google]

48KB, 600x329px

>>61034357
>Nobody on /g/ works a job where they buy servers or server hardware.
>Starting at 475 USD.

Got me to reply.

>>61034642
>gayming

lol ok cool

File: ok.png (165KB, 2640x2200px) Image search: [Google]

165KB, 2640x2200px

>>61034147
its all numbers. no one needs even 16 core CPUs. it falls back to program makers not making good use of even 4. if the raw horsepower of each core is higher, THEN intel is in trouble.

>>61043960
>no one needs even 16 core CPUs.
Meanwhile here in the REAL world Intel has been increasing the core count of their server CPUs with each generation, with up to 24 cores in Broadwell and 28 in Skylake. I guess they must have been doing that for no reason, huh.

>>61043960
>server software can't use more than 4 cores
t. retard

>>61043960
Welcome to servers. Your 4 core "muh clocks" CPUs are far, far away now.

>>61043960
Can you not identify a server thread even when you're actually in one?
Most enterprise CPUs sold nowadays are for "scale out" crap like web or VM farms, file servers, etc.

>>61042788
Because it's fun. Same reason why people cheer for sports teams or engage in console wars.

No one needs 3.2GHz boost frequencies on their server farms. 2.8GHz on half the available cores is adequate enough.

File: tXcQa7j.png (602KB, 921x543px) Image search: [Google]

602KB, 921x543px

>>61044208
It's 3.2 GHz on 12 cores. All core boost is 2.7 GHz.

>>61044208
>needs
>enough
fuck off with those communist words

>>61044281
Where did they say max turbo is good for up to half cores active?

>>61044802
https://arstechnica.com/information-technology/2017/06/amds-zen-steps-into-the-server-room-with-epyc/

>ike Ryzen, Epyc can boost clock speeds depending on usage levels. The top-end 7601 part, for example, has a base speed of 2.2GHz, with an all-cores boost of 2.7GHz and a maximum boost of 3.2GHz. Ryzen's maximum boost is very limited, only applying with one or two cores active. Epyc's is a bit more versatile; that 3.2GHz can be reached with up to 12 cores active.

>>61044847
So Ryzen = 1-2c/die, and Epyc = 3c/die then. Thanks.

I wonder where TR will fall regarding this behavior.

>>61044802
>>61044847
The power consumption throttling is handled by each individual die afaik. This is probably the most effective option for load balancing. Can't confirm but maybe a reviewer will check if it's three cores from each die when they post a per core clock chart under load.

>>61044917
Probably 2-3c/die

File: 1471010470924.jpg (154KB, 900x657px) Image search: [Google]

154KB, 900x657px

>>61045013
>>61044917
I'd bet on at most 2c per die, since workstation loads are more dependent on single thread performance than server.

File: amazaryzen.png (68KB, 766x668px) Image search: [Google]

68KB, 766x668px

>>61045055

>>61034313

Don't they have on the die fpgas. I'm sure those fit a niche

They're already fucked, but the enxt year might be the nail in the coffin with 7nm http://www.anandtech.com/show/11558/globalfoundries-details-7-nm-plans-three-generations-700-mm-hvm-in-2018

>>61034738
She is so smug and for good reason.

File: amadaexciteULTRA.png (210KB, 422x537px) Image search: [Google]

210KB, 422x537px

>>61045096
Oh shit.

>>61034814
I am so poor man. Maybe I should just ask family or some shit. Q2 will hopefully make it dip since desktop isn't going to suddenly make them billions. What I want is the Q3 and after. AMD is going to get a huge boost with TR and Epyc.

>>61034284
what you mean superior product? there isnt a single intel cpu out there that the uncore speed is tied with the ram speed..

what if amd decide to create a new ecc ram for server running on 3200? amd will literally destroy intel

>>61034545
HL3 is never going to happen you say ?

google jiira files leak

>>61045228
DELET

man you know intel shills are so butthurt when they say that intel is superior to ibm..

File: 1496702263772.png (162KB, 633x900px) Image search: [Google]

162KB, 633x900px

>>61045228

>>61045252
They gotta get with the times and start shitposting that Intel is better than IBM, Samsung, GloFo and AMD combined.

File: 1486851399777.jpg (200KB, 1024x1024px) Image search: [Google]

200KB, 1024x1024px

>>61045281
AMD can fab at TSMC as well.

>>61041913
kek, this is hilarious. Samsung is obliterating Intel on storage, ARM is obliterating on mobile, and AMD will obliterate them on servers. Things are not looking great at all for Intel.

>>61042788
Might as well ask why not go to Walmart since they have the best prices. Intel are fucking cunts. I'm not saying AMD is the good guy company sicne they all want our shekels, it's just that they're less shitty than Intel.

File: 1487669618201.jpg (35KB, 544x365px) Image search: [Google]

35KB, 544x365px

>>61045359
>Samsung is obliterating Intel on storage,
This is incorrect. The IMFT NAND is a superior process on both a density and cost basis. The "meme" 3D Xpoint also BTFO's everything in existence at low queue depths.

Everything else is right though.

Also can someone shoop an i9 into the fireplace?

>>61045435
>The "meme" 3D Xpoint also BTFO's everything in existence at low queue depths.
We'll see when it hits officially and not those retarded 32GB dims. Also, Samsung is apparently working on a similar technology named Z-NAND so Intel's xPoint might have trouble later on as well.

>>61045502
the future is memory channel flash

>>61045502
Z-NAND is years away from hitting mass production, and is clearly a worse implementation for filling the DRAM>Storage gap. 3D Xpoint hasn't even begun to reach it's max potential or density yet either, who knows where it will be by the time "Z-NAND" is actually more than a concept.

>>61045560
>storage at similar speeds to RAM
muh dick

>>61045570
How about read/write cycles that aren't complete shit?

>>61042788
Because there are filthy fucking liars who would present erroneous information from a position of authority and expertise.

By combating those degenerates who lie all the fucking time who need to prop up their business we now have an equal voice.

File: delet_this.jpg (13KB, 480x480px) Image search: [Google]

13KB, 480x480px

>>61034440
>>Why would you have threads about Half Life 3 on /v/

>>61045096
Wew lad.

>>61045302
Too bad TSMC's 7nm is low power mobile SoC trash.
>>61045228
And Intel is still silent about their 10nm. And when will they start inserting EUV?

File: GntGYw.png (38KB, 499x338px) Image search: [Google]

38KB, 499x338px

>>61045624
>This semiconductor product coming to market in 3 years has been promised to offer improvements over what's immediately available today.
Well shit, I'd sure fucking hope so. What do you think will happen until then?

>>61034748
7nm parts have already taped out. At most it'll be the 1 half of 2019

File: [HorribleSubs] Gabriel DropOut - 05 [720p].mkv_snapshot_10.19_[2017.02.25_01.06.39].jpg (31KB, 360x360px) Image search: [Google]

31KB, 360x360px

>>61045096
>7nm DUV in 2H 2018
How would it make it into Ryzen 2 then? Or am I being dumb and misunderstanding something?

File: 0GSCodm.jpg (693KB, 2550x1400px) Image search: [Google]

693KB, 2550x1400px

>>61045979
Not to mention AMD can risk server chip production earlier in the process cycle because even with a 12c die, it could end up being smaller than Zen 8c dies now. In comparison, Intel is forced to do their large chips only on a fully mature process due to monolithic dies. Just look at the release time from Zen > Epyc vs laptop > server parts for Intel.

>>61046130
This image needs updating. It's still using that silly leaked blurry as fuck dieshot.

>>61046088
Dumb sataniaposter.

>>61034147
Look Intel's upgrade path, they're not the type of company to ramp up core counts for shits and giggles, better processors will come when there is a real-world need for them. Right now it looks like AMD just pulled out its 12" dick, except that it has issues with getting an erection, which is what matters more than anything.

>>61046312
>Look Intel's upgrade path, they're not the type of company to ramp up core counts for shits and giggles
Literally the only reason they're not releasing a 32-core CPU to compete with Epyc is because they can't.

>>61046195
And virtually everything it labeled ended up being wrong.

File: dinner with slaves.png (61KB, 276x577px) Image search: [Google]

61KB, 276x577px

so what's this about MUH 60% MARGINS? i never understood what it meant

>>61034796
I want Opteron back :(

>>61046312
>issues getting an erection
but it crushes all the xeon e5 line by a long shot anon, I'd hardly call that shitty performance

but since you're a blind shill and not really here to make an proper conversation, fuck off

>>61046312
Yet another retard that doesn't understand anything about CPUs besides "muh clocks" 4 cores. Fuck off back to /v/, faggot.

>>61046411
intel is forced by their shareholders to make $1.6 for every dollar they spend making a processor

>>61046486
*at the very least $1.6

>>61046486
>>61046507
and this paired with their shitty way to fabricate processors turns into really high prices for the end user

>>61046522
define "shitty"

>>61047083
Not, in a sense, "modular" like AMD's process with CCXs, they have to make large monolithic dies for high core count CPUs.

>>61047083
I'm not sure how it applies elsewhere, but HCC Xeon models have kind of crap yields thanks to their large die sizes.

AMD's Voltron trick with TR/Epyc uses a bit more power than an ideal monolithic chip, but it's certainly possible to make a ton of them much cheaper than Intel's 24+c chips.

>>61046312
all the serious players came on board and shilled for amd
all of them said they have already servers with epyc for them
amd annc that they got ALL the players on server market
it has literally the best power consumption by far
it has the best perf by miles
it has the best price by light years
its literally the whole package
and the good part is amd can afford to sell them at even lower prices give the yields and die usage LOL

>>61047083
when intel makes processors with high core counts, they need to make one big chip that contains all the cores inside, if one of the cores doesn't work, it goes to the trash

This means that they can only make use of small fractions of a wafer (<10%)

when AMD makes processors with high core counts, they just have to "glue" together 2 or 4 chips. each chip has 8 cores in total and are considerably easier to make

this means that they can make use of large fractions of a wafer (>80%) for fully working 8 core chips, while the chips where one core or more doesn't work can be used in processors with 6 or 4 cores per chip

>>61047238
this allows AMD to sell chips with high core counts for really cheap, while having a larger profit margin compared to intel

>>61047238
>>61047248
i assume core-interoperability between CCXs suffer as a redult

>>61047273
It's a drawback, yes, but only noticeable in specific workloads with constant inter-thread communication

>>61047290
with proper thread scheduling this problem would be even harder to notice

>>61047308
Linux had all the necessary patches merged well before Ryzen was released.
Only Wangblows had problems with scheduling, and not for the first time - this exact same shit happened to HyperThreading when Nehalem was introduced.

>>61047392
let's see if microsoft changes their minds, seeing that they're supporting AMD in the Enterprise market now

>>61047181
>Epyc uses a bit more power than an ideal monolithic chip
HE DUN KNOW ABOUT BINGBUS

>>61047417
what about the ebinmail?

>>61047427
Ebinmail is the worst of two worlds: shit latencies and meh scaling. Better than bingbus for HCC probably.

File: big_intel-xeon-platinum-8176-cinebench-r15-benchmark.jpg.jpg (281KB, 980x942px) Image search: [Google]

281KB, 980x942px

>>61047427
>>61047532

>>61047559
I want to get off Intel's ringbus ride.

>>61046312
kek. I feel bad for Intel's Xeon line. Those must have atrocious yields and AMD can even undercut them to hell, but they're not even going all out.

>>61047591
Skylake-E ditched ringbus for the very much immature mesh.

>>61047616
if AMD went for 60% margins on the server chips they're destroy every hope of intel selling server hardware ever again, those things are dist cheap to assemble

>>61047616
Prepare for imminent Intel marketing onslaught about why 4S/8S is utterly crucial for every business now.

File: 1468871826010.jpg (147KB, 1024x1024px) Image search: [Google]

147KB, 1024x1024px

>>61047637
Which actually was a fucking meme in itself because it's only going to improve performance on 12c+ SKL xeons and regress performance on everything else in some situations. Unfortunately those are some of the most popular chips in 2P server space at the moment.

>>61047637
Mesh is literally moar bingbus

>>61047637
Is this mesh shit ever going to improve?

>>61047660
>4s/8s
4/8 socket? you mean 4 or 8 processors per motherboard?

>>61047660
but their stuff scales even worse on 4S/8S

it's like 2.5x on 4S

>>61047680
both

>>61047660
Intel scales like hot turd on 4S/8S. They literally have no chance against EPYC.

>>61047660
>>61047680
>>61047684
4s/8s is mostly giant database servers from my understanding, and in rare cases large working sets of memory to serve out to coprocessor workloads.

>>61047679
Probably. But whether or not it does, it still means Intel is stuck with monolithic dice for the time being.

>>61047708
aren't universities who buy them mostly?

>>61047732
Yes. The rest of the world uses fucking 2S and soon will use 1S.

>>61047746
it's sad to think that most people with 2S are only using it because of the amount of lanes available

>>61047758
Well, that's Core for you. Too bad Zen is about to demolish it so fucking hard it'll be painful to watch Intel damage control for the next 4 years.

>>61047783
what's intel's new arch all about again?

>>61047791
Chopping off parts of x86 to make it "faster" hopefully without resorting to oversized vectors.

>>61047809
>hopefully without resorting to oversized vectors.
This means it's going to resort to oversized vectors, isn't it?

>>61047809
that doesn't sound all that good, intel slept way too much

>>61047532
>E B I N M A I L
Holy shit, truly epyc name. Lol'd, have a (You).

>>61047842
Yes. Probably. Its Intel you know. If Oregon team has anything to do with their new uarch it 146% means moar oversized vectors.

>>61047809
>>61047842
>oversized vectors
AVX1024 when?

>>61047844
It means that x86 needs a mercy killing by the hands of RISC-V. With the help of AMD of course.

>>61047873
can you do a tl:dr on RISC-V?

>>61047890
It's a new modular royalty-free RISC architecture without being a housefire like OpenPOWER or being irrelevant piece of shit aka OpenRISC.

>>61047913
but why is it so much better than x86?

File: NXT0grZ.jpg (223KB, 1920x1274px) Image search: [Google]

223KB, 1920x1274px

>>61047859
Why have AVX1024 when you could have AMADA?

>>61047925
No Intel.

>>61047944
Vega10 can't into 1/2 FP64 tho.

File: 1471268429567.jpg (55KB, 800x800px) Image search: [Google]

55KB, 800x800px

>>61047962
>This body pillow doesn't have 1/2 FP64.

>>61047946
eventually intel will end up selling x86

>>61047994
x86 would be long dead by then and you'd need to license -64 part of it from AMD.

File: m1q9yw7ugp2z.png (10KB, 583x344px) Image search: [Google]

10KB, 583x344px

>>61047946
2bh valid argument if you look up what Intel "did" while it reigned supreme

>>61048015
>Intel spends years and billons of dollaridoos to develop abortion known as XPoint
>AMD with their peanuts R&D budget invents fucking HBM
Really makes you think.

>>61047913
An ISA can't be a house fire. It's just a bunch of specifications and manuals

>>61048046
>implying all that money was actually going to R&D
It was probably going to other (((people)))

>>61048092
Only P.A Semi managed to make non-housefire PowerISA chips, and P.A. Semi was absorbed by crapple. Maybe Google will make something.

>>61048105
No, Intel spends fuckton of money on fabs while failing to deliver new nodes because of their insane boner for BEOL scaling. At least they make good SRAM and NICs.

>>61048132
can't find shit about it besides news trowed around

what is BEOL scaling?

>>61048046
>intel
>have bilions in the bank
>buys mcafee
HAHAHAHAHAHAHAHAHA

>>61048176

At least Altera acquisition was sane enough.

File: 1234810246786.jpg (28KB, 432x288px) Image search: [Google]

28KB, 432x288px

>keep silent for years
>Make your opponent lazy
>Come back with the biggest blow it literally crippled your opponent.

Should the Radeon division do the same too?
Instead of try to catching up just let Nvidia meme them self like Intel

>>61048205
Navi will come with MCM gpus.

>>61048205
If they managed to make MCM work on GPUs, something similar but not so strong will happen to nvidia

we'll see on Navi

>>61048205
Huang already made a second thermi with 815mm^2 die.

>>61048221
>>61048222
hivemind

>>61048234
sauce?

>>61048248
GV100 is 815mm^2. I wonder if the yields will be lower than 1.7%.

>>61048221
>>61048222

The concept of MCM (as it applies here) for a gpu is considerably more difficult to implement that it is for cpus. The sort of latencies gpus deal with might mean that physically having two dies that far apart won't let you signal fast enough between each die without crippling performance.

>>61048275
nvidia has a history of lying on TDP too?

>>61048317
What do you mean by
>lying on TDP
TDP != power consumption.

>>61048315
>that far apart
Physically they are going to be very close, sitting on one big meaty silicon interposer.

>>61048330

>TDP != power consumption.

/g/ doesn't understand this at all which is why this board is pants-on-head retarded when it comes to power draw figures of electronics.

>>61042627
ok? AMD's SMT is still 20-30% better than intel's garbage shit

intel does threading worse than every company and architecture except ARM

>>61048337
>/g/
You mean /v/? Consumer hardware thread are 90% /v/.

>>61048315
I know, that's why i'd rather wait and see instead of guessing stuff wrong

this is just basic speculation for me (don't know about the others)

>>61034147
>thinking AMD is relevant
Intel competes with Qualcomm and Apple soon, not some small ass company that only exist due antitrust laws.

>>61048528
>STILL didn't give up
holy shit man, don't you have anything better to do?

>>61048543

Not while his 7900x is cooling down.

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

173KB, 640x400px

>>61048528
>Intel "competes" with Qualcomm
By completely failing to break into the mobile market and getting their ass kicked? Then getting pooper peeved when they start emulating x86? Major competition going on right there.

>>61048543
I didn't even read the rest of the thread, nigga.

>>61048571
My desktop got an old ass FX 6300.

>>61048577
That's the point. Qualcomm and Apple actually have the chance to really fuck them over and they don't seem to have a plan for it. Whatever AMD does is irrelevant. They could kill them at any point by lowering their margins, just it'd lead to lawsuits.

>>61048617
>lowering their margins
I don't think losing 90% of their shareholders will help them

File: doitfgt.jpg (85KB, 835x703px) Image search: [Google]

85KB, 835x703px

>>61048617
>I didn't even read the rest of the thread, nigga.

>>61036236
yeah no shit
386 > pentium > pentium III > q6600 > 980x

each of these progressions is like 5x the performance and I've been stuck at 980x for 7 years because I refuse to buy a cpu that isn't at least twice as fast, overall. Not even talking single core or IPC so that has to tell you something about CPU evolution these last years.

Kudos to AMD, looks like I'm getting threadripper very soon

>>61048617
>Whatever AMD does is irrelevant.
You stupid useless faggot. They're poised to recapture some serious marketshare in the server space, where Intel gets most of their money from. Epyc is no joke.

>>61034796
2xStarship would be pretty nice. 96 threads 8)

>>61048742
96 cores 192 threads you mean right?

>>61048645
It's not like it'd be noticeable for their bottomline, you really underestimate their size and overestimate AMD budget.

Then again, it's all hypothetical bullshit, they can't afford AMD to die. As long their market share stays similar, having them offer a competitive product doesn't matter at all.

It's the new players that are a real danger.

>>61048711
>some serious marketshare
Serious for AMD dimensions. Peanuts for Jewtel. Most of the shit is secure due long ass running contracts either way.

>>61035218
I just ordered a 44 core desktop for myself at work

>>61048770
>dual socket for 44 cores
kek

>>61048767
nice job waiting for bump limit before coming to shit up the thread

>>61048767
>It's not like it'd be noticeable for their bottomline
You're literally retarded.

>>61034147
anyone have the intel roadmap image that had us on 7nm like 2 years ago?

>>61047238
That isn't true. Intel has a few models and she bins damaged dies. How you think this i5 and i3 and lower xeon skus appear?
What is different is that because the die is smaller there is a lower chance the de can't be used for any bin. The more massive a die is harder to get a perfect litography.

>>61048831
End result is still the same. High core count Xeons cost a zillion bucks.

>>61047637
Mesh performance and worsen l2 we mean for enterprise apps so Intel don't give a shit for anyone using it on hedt.
This whole avx shit need to die.
I still don't understand how Intel create those shitty chips with gpu. Either create a competent apu or just ditch it completely from a die like Zen does.
This whole shit of adding useless functions need to die.
I'm yet to see anyone using avx on i5 or i7

>>61047944
Can you actually buy this pillow cover anywhere?

>>61048808
You're pretty retarded if you think it'd be more than single digits. The gap between them is ridiculous, and customers would never give AMD a chance unless there is a huge benefit in performance AND price, and Jewtel got all of the OEMs on their circumcised cocks to ensure nothing major happens in the first place. And over 98% in the server market.

Stop being a retarded fanboy. Enjoy Epyc being not shit, buy some shares, but calm the fuck down. Nothing major will change.

>>61048907
>unless there is a huge benefit in performance AND price
There is you stupid faggot. You'd know this if you read the goddamn thread.

>>61048132
Millions on Rd
>Larabee
Dead
>Xpoint
Shit
>Mobile CPUs
Dead
>Millions to advance nodes
Still failing year after year releasing reviews of same node since haswell

This is some dumb way to invest

>>61048907
>customers would never give AMD a chance
literally 80% of the server market is going with amd and EPYC
>unless there is a huge benefit in performance AND price
yes, there is
>Jewtel got all of the OEMs on their circumcised cocks
not anymore

>>61049310
>literally 80% of the server market is going with amd and EPYC
Got any source for that?

>>61049416
Read the damn liveblog of the launch event. They have a lot of major players onboard already. http://www.anandtech.com/show/11562/amd-epyc-launch-event-live-blog-starts-4pm-et-

>>61048907
>Jewtel got all of the OEMs on their circumcised cocks to ensure nothing major happens in the first place
AMD has OEMs on their side, and if Intel tries the same slimy shit they did last time, the FTC is watching.

>>61049543
>>61049558
Offering AMD stuff is a huge step from actually having customers to choose AMD stuff.

>if Intel tries the same slimy shit they did last time, the FTC is watching
Last time they got a slap on the wrist so laughable, even Nvidia joked about it. Why would it be any different now?