Quantcast
[ 3 / biz / cgl / ck / diy / fa / g / ic / jp / lit / sci / tg / vr / vt ] [ index / top / reports / report a bug ] [ 4plebs / archived.moe / rbt ]

Due to resource constraints, /g/ and /tg/ will no longer be archived or available. Other archivers continue to archive these boards.Become a Patron!

/g/ - Technology


View post   

[ Toggle deleted replies ]
File: 1.28 MB, 1920x1080, Screenshot 2018-12-04 16.12.40.png [View same] [iqdb] [saucenao] [google] [report]
68787521 No.68787521 [Reply] [Original] [archived.moe] [rbt]

WHY DID YOU MAKE ME DO THIS INTEL

https://www.youtube.com/watch?v=PCdsTBsH-rI

>> No.68787634
File: 10 KB, 234x216, download.png [View same] [iqdb] [saucenao] [google] [report]
68787634

THIS CAN'T BE HAPPENING

>> No.68787637
File: 324 KB, 882x758, 1506977173618.jpg [View same] [iqdb] [saucenao] [google] [report]
68787637

>>68787521
NONONONONO PLEASE BUY INTEL

>> No.68787665

>>68787521
Big if true
This could make Intel's 10 core CPU look like a shitty joke

>> No.68787684

>>68787521
>6c/12t for $100
I'm just going to bask in this for a bit, don't mind me.

>> No.68787689

>>68787684
r5 1600 is already like $130

>> No.68787702

>>68787521
I hope they stomp my 9700K into the pavement. Fuck Intel.

>> No.68787707
File: 288 KB, 552x661, 1515411243361.png [View same] [iqdb] [saucenao] [google] [report]
68787707

IT WAS OUR TURN

>> No.68787709

>>68787689
It retailed at $250, and that wasn't even two years ago.

>> No.68787722

big if true

>> No.68787724

>>68787521
is this the same leak that had the 64c 5ghz base meme?

>> No.68787738

>>68787724
The video doesn't mention anything about EPYC, so presumably not.

>> No.68787751

>>68787724
no

>> No.68787767

>>68787521
And DDR4.
Looks like the ol' 3570K is getting replaced with a new 3*** part.

>> No.68787778

>>68787521
This'll stir up some shit if real, hot damn.

>> No.68787782

>>68787724
He compares them in the video, and remarks that seeing the 64 cores at 5GHz should indeed make you go wat

>> No.68787792
File: 57 KB, 329x329, 1505147990486.jpg [View same] [iqdb] [saucenao] [google] [report]
68787792

>>68787521
b-but. . . m- more . . . b-b-bingbus. . . MO. . .RE HOUSEFIRE!!!!

>> No.68787801

>>68787778
>16 cores for fucking mainstream
Bloodbath.

>> No.68787817

>>68787801
If they work. Remember the 2990WX is slower than a 2700X in like 90% of tasks, and loses to the 9900k in 99%.

>> No.68787821

>>68787801
Don't forget the ECC memory support. We're potentially looking at Xeon slayers here.

>> No.68787822

>>68787521
Quad cores will still exist though, 2 cores from each 8 core chiplet on AM4

>> No.68787835

>>68787817
Different layout, on AM4 you will have two 8 core chiplets access the same I/O die in the middle.

>> No.68787837

People won't forgive him if he is wrong. Hope it isn't fud material or RIP

>> No.68787843

>>68787521
I'm still remaining sceptical, but hey, CES is only around the corner w so I guess we might find out soon. Either way, I'm glad I haven't invested in a high end Coffeelake system the last month.

>> No.68787855

>>68787822
Don't think so unless they decide to bring a monolithic binned laptop APU to desktop as an Athon.

>> No.68787857
File: 255 KB, 462x492, want.jpg [View same] [iqdb] [saucenao] [google] [report]
68787857

>>68787521
>them gpu speculations

>> No.68787917
File: 3.70 MB, 448x251, 1521901955555.gif [View same] [iqdb] [saucenao] [google] [report]
68787917

>>68787521
>affordable 8c/16t
>those tdps
>that pricing on everything
>$99 for a 6c 12t cpu
Are we finally entering a second golden age of cpu advancement after the stagnation that bulldozer/intels complacency caused?

>> No.68787959
File: 1.83 MB, 700x5000, 1539898335826.jpg [View same] [iqdb] [saucenao] [google] [report]
68787959

Reminder.

>> No.68787975
File: 244 KB, 1800x850, 1543685951637.png [View same] [iqdb] [saucenao] [google] [report]
68787975

>>68787959
Indiantel is here.

>> No.68787990
File: 1.05 MB, 1196x676, 1527619034298.png [View same] [iqdb] [saucenao] [google] [report]
68787990

>>68787959

>> No.68788003

>>68787959
Not even a shill, I have a 6700k, but holy shit I want 1080p to fucking die. It's a baby resolution eternally inferior to 1200p and it needs to fucking go.

>> No.68788008

>>68787959
>everything is gayming
>>>/fa/cebook

>> No.68788009

>>68787521
Extremely skeptical but this would make Zen 2 one of the best CPU lines released in the past 20 years if it ended up being true, not even Sandy Bridge was this good

>> No.68788013
File: 110 KB, 750x600, 1337029848462.jpg [View same] [iqdb] [saucenao] [google] [report]
68788013

>>68787959
Crashes premiere right?
amirite?
Do they pay you per post or per word?

>> No.68788021

I think - assuming this leak holds true - that the 3700x will be THE cpu to get for top tier AM4 perofrmance when cost is factored in. Particularly given it falls within the 105w TDP which means (assuming mobo manufacturers bother) the older x370 and x470 boards can handle that even with a potential overclock.

Its chips like these which are part of the reason I bought a C6H for my 1700 as I wanted a board capable of handling monster chips years later.

>> No.68788039

>>68788021
>105w TDP which means (assuming mobo manufacturers bother) the older x370 and x470 boards can handle that even with a potential overclock
Exactly. X470 owners will be replacing every 2600

>> No.68788042

>>68788021
Too bad it's not real. Maybe next year guys, oh well

>> No.68788043
File: 806 KB, 1001x823, happy.png [View same] [iqdb] [saucenao] [google] [report]
68788043

>>68787521
poorfag here with an shitty i3 6100 Every single day I want to get rid of this piece of shit. if that ryzen 3 3330 is true im goona be so fucking happy.

>> No.68788058
File: 55 KB, 655x919, 22aRaTn.jpg [View same] [iqdb] [saucenao] [google] [report]
68788058

>>68788042
This is you

>> No.68788061

Still waiting on how 3rd gen TR performs. I hope they have another 16c part; ECC DDR4 can really only feed 4c/channel.

>> No.68788065

>>68787521
How is this any surprising at all? Probably half of Intel’s top engineers have went to Apple the last few years. It’s not AMD catching up, it’s Intel slowing down.

>> No.68788073

>>68788058
And she did win the popular vote

>> No.68788078

>>68788042
Just wait for Zen 2+

>> No.68788079
File: 71 KB, 400x388, 1540395672071.jpg [View same] [iqdb] [saucenao] [google] [report]
68788079

I feel uneasy about CPUs with core/thread counts that aren't powers of two, and I'd rather get a 3600X than a 3700X even if they were almost the same price.

>> No.68788100

>>68788079
>get 3700X
>turn off a few cores
>enjoy the extra frequency

>> No.68788104

>>68788079
I'm you but the opposite.

>> No.68788116

>>68788065
>half of Intel’s top engineers have gone to Apple
So that's why they haven't been innovating for Moses knows how long. Riiight. It makes so much sense now. Thanks for clearing that up.

>> No.68788118
File: 38 KB, 436x402, 1454450833116.jpg [View same] [iqdb] [saucenao] [google] [report]
68788118

>>68787521
>R3 3300
>99$
>6C/12T
>overclockable
is this real life?

>> No.68788124

>>68788079
Same. I have a weird autism about everything in my computer NEEDING to line up with powers of 2.
>1, 2, 4, 8, 16, etc terabytes of storage
>1, 2, 4, 8, 16, etc gigabytes of RAM
>1, 2, 4, 8, 16, 32, 64 cores on the CPU
>64CUs in the GPU (I can only live with a Fury X/Nano or a VEGA64/FE)

>> No.68788127

>>68788079
I only buy CPUs with core/thread counts that are powers of three. Still waiting to upgrade my Phenom X3

>> No.68788143
File: 49 KB, 450x571, 1288367321063.jpg [View same] [iqdb] [saucenao] [google] [report]
68788143

>>68788073

>> No.68788149

>>68787521
this is fucking insane
are you saying that this summer i could be using 16core 32 thread 4.5GHz+ CPU on ALL cores with haswell+ IPS? Is this real life?

>> No.68788157
File: 26 KB, 325x243, 1494814430892.jpg [View same] [iqdb] [saucenao] [google] [report]
68788157

>>68788127

>> No.68788160

>>68788143
You will be the butthurt one when this turns out not to be true. Actually, you will probably pretend like you never believed it

>> No.68788161

Everything honestly seems relatively reasonable with regard to the numbers given. Only thing that seems iffy is the clockspeeds, and the possibility of not having enough memory speed to feed over 12 cores or more with dual channel. If a memory problem exists though it's likely it could be solved by DDR5 since AMD was trying to get some research done on that.

>> No.68788163

>>68788118
>50 watts
AYYYMD CPOOS CAN'T EVEN HEAT YOU ON WINTER
THANK YOU BASED INTEL FOR TURNING MY ROOM INTO A BONFIRE

>> No.68788165

>>68787521
Provided that these "leaks" are true... holy shit son, 6c/12t for a Benjamin? This would have never happened with Intel in the lead.

>> No.68788168

The wojak posts have to be a bot or a paid amd shill.

>> No.68788172
File: 31 KB, 638x716, 1517628412870.jpg [View same] [iqdb] [saucenao] [google] [report]
68788172

>>68788118
>will be able to overclock it to hell and back at 50tdp and an aftermarket cooler
>$99

>> No.68788182

>>68788160
You're being >>68788058 again

>> No.68788186

>>68788168
Probably

>> No.68788195

>>68788172
>Ryzen
>overclock
>not just letting the integrated boost do its almost flawless magic and focusing on providing it with much needed RAM overclock instead

>> No.68788196

>>68788186
I agree

>> No.68788199

>>68788149
>Is this real life?
That is indeed the question.
>>68787837 has a good point

>> No.68788200
File: 279 KB, 540x540, 1542837269452.gif [View same] [iqdb] [saucenao] [google] [report]
68788200

>>68788149
They increased ipc by up to %29 depending on workload in addition to the core count/clock speed increase.

>> No.68788207

>>68787521
>launching an 8-core 20CU APU for $50 more than an equivalent AMD GPU
>8 7nm cores for $180

You people are fucking retarded and out of your god damn minds.

>> No.68788211

>>68788161
I'd like to see HBM L4 before we see DDR5. More bandwidth, better latencies, and huge capacity right next to the die for very little extra TDP. The biggest gain would actually most likely be gaming. Not averages, mind you, just 1/0.1% lows in purely CPU bound scenarios. They would basically disappear completely, as there would be VERY few L4 cache misses (assuming between 1 and 8GB of HBM cache).

>> No.68788221

>>68788168
Or OP being 12

>> No.68788222

>>68788195
>not overclocking a $99 cpu just for fun because you can
>not seeing how good you can make it

>> No.68788223

>>68787521
>tfw my 2600 is deprecated by a $100 3300
should have waited.

>> No.68788228

>>68788013
That or he's just from a shitty broken family turning him into an attention-seeking shitposter.

>> No.68788242

>>68788021
>3700x
>not just shooting for the moon with a 3850x
I unironically wanted a 9900k but 16core at 5ghz+ is too hard to pass up

>> No.68788255
File: 2.46 MB, 320x240, 1513431470984.gif [View same] [iqdb] [saucenao] [google] [report]
68788255

>>68788223
>got a 1600 on launch
>waited because didn't see the point in zen+
>will be able to get a comfy 8c 16t zen 2 cpu for <£250

>> No.68788258

>>68788199
I want to believe tho

>> No.68788259

>>68788207
Just let the AMD fanboys have their moment before reality comes crashing down on top of them

>> No.68788276

>>68788207
r7 1700 is $170 faggot

>> No.68788278
File: 73 KB, 700x565, 8kw7ryfyfrfz.jpg [View same] [iqdb] [saucenao] [google] [report]
68788278

>>68788168
NOOOOOOOOOO THEY CAN'T BE REAL!

>> No.68788286
File: 91 KB, 700x525, 1336779853646.jpg [View same] [iqdb] [saucenao] [google] [report]
68788286

>>68788223
mfw I got my 2600 for -60€ because I called amazon to return the 1600 and rather than just send me another 1600 they just gave me my money back to buy a replacement myself and when I went to buy a replacement there was a flash deal on a 2600 for 140€.

>> No.68788296

>>68788200
The +29% IPC increase comes from Float-heavy workloads, since (as I predicted last year) Zen2 has fully 256-bit FPUs
Serial/Integer IPC is probably up 11-15%

This is legitimately AMD's Nehalem > Sandy Bridge evolution and it will be glorious.

>>68788276
It launched at $320 you retarded nigger.
Tell me why parts won't be at price parity with older parts that have the same performance when it's been that way from the beginning?

>> No.68788306

>>68788296
Because the 8 core Zen chiplet is 72mm^2 thus the yields will be good. Are you retarded?

>> No.68788316

>>68788195
do we know if they're giving XFR to the Ryzen3? I don't see why not, but entry level CPUs usually have some downside with OCing

>> No.68788322

>>68787521
I don't even care if they don't get to 5GHz
4.6 or above is an instabuy for me

>> No.68788327

>>68787521
Fake and gay.

>> No.68788338

>>68787975
>Germanistan

>> No.68788344
File: 339 KB, 682x792, 1515221586890.png [View same] [iqdb] [saucenao] [google] [report]
68788344

>>68787521
onononononononono

>> No.68788346

yuo waitfags ITT ar babbis
watch this
>2020
>7nm+
>Socket TR5
>2 8-pin CPU plugs, 2 8-pin PCI-E plugs
>DDR5 w/ Registered/Buffered ECC
>32c @ 5.4GHz, 4 chiplets
>16GB HBM3 L4
>Navi successor 64CU on-package
>32GB HBM3 Navi memory
>8 memory channels
>4 DisplayPort 1.5 ports
>8tb memory
>every single one of the 128 PCI-E 4.0 lanes saturated with as much Optane as can fit
I fucking love waiting for this

>> No.68788348

>>68788322
kek
AMD can't even reach 5GHz. Enjoy your shitty house fire CPU

>> No.68788349

>>68788306
Evidence? Statistics? Figures?

Usually when fabricated SKU lists come out someone has leaked invoice manifests from the packaging plants. We haven't seen any of that.
I'd also like to know where you get your data for TSMC's 7nm defect rates.

>> No.68788350

>>68788296
That guy from work with a nehalem i7 and a GTX 480
It was like a little space heater in his office

>> No.68788351

>>68788211
Why was HBM so expensive? 1 or 2 suppliers with low volume runs? What's stopping AMD from just ordering them from TSMC or GloFo? they could use that for their entire product lines

>> No.68788355

>>68788338
Hire more indiantel yieldshitters Intel, your shitposts don't work.

>> No.68788358

>RX3080 = ~RTX 2070 for $250
I can't wait for AMD to use 7nm to kick Nvidia's teeth in with their fat trash dies that are good for machine learning and other professional applications but might as well be dark silicon for rasterization in gaming. AMD doesn't have a stock surplus of RX570/80s that they need to get rid by keeping prices high post-mining bust either.

>> No.68788392

>>68787521
fake viral marketing

>> No.68788399

Imagine if these leaks are true, a budged low power and almost silent gaming pc with the lowest tier navi with no extra power conector and a ryzen 3 3300. For poorfags that are still fine with 1080p it's going to be amazing.

And that navi equivalent to vega64/rtx2070 seems perfect to throw my gtx 970 in the dumpster where it belongs.

>> No.68788407

>>68788351
HBM is expensive because of low volume. It is legitimately the best non-persistent memory solution we have as far as balancing power usage, density (how many bytes/mm), and speed (both latency and bandwidth are vastly better than DDR5 with HBM2). It isn't widely used outside of supercomputers and GPUs because packaging it onto CPUs or memory sticks is a massive engineering hurdle we haven't done yet (unlike SRAM and DRAM), and would disrupt current standards. So, because it's hard to do and relatively new, it's low volume and will remain as such for the forseeable future.

>> No.68788437

>>68788399
>Imagine if these leaks are true
they aren't

>> No.68788453

>>68787521
>affordable 6c/12t cpu + gpu chiplet
either you are being massively b8ed jim, or next year is gonna be pretty interesting
with that said, go fuck yourself jim, go advertise your shit on plebbit

tfw i still have to decide what music im gonna play on intels wake ceremony

>> No.68788456

>>68788399
There are a lot of nice IPS high-refresh freesync 1080p monitors as well.

>> No.68788457

>>68788399
>Imagine if these leaks are true
they are

>> No.68788458

>>68787521
If this is true INTEL BTFO!!!!!!!

>> No.68788459
File: 25 KB, 300x300, 1189215622091.jpg [View same] [iqdb] [saucenao] [google] [report]
68788459

>>68788259
>>68788327
>>68788348
>>68788392
Your anguish sustains me

>> No.68788462

>>68788306
70mm2 is not small for 7nm. Look at the xtor count in Apple's A12 and A12X. A die somewhere around 80mm2~ can fit 7billion xtors. That isn't a small die. Given this extreme density, the fact that this process uses quad patterning everywhere, it being a new node, yields aren't going to be high. Its pure fanboy retardation trying to rationalize this any other way. Even with EUV yields won't pick up for some time until pellicles

>>68788351
Its expensive because its complex, its stacked DRAM slices. If anything goes wrong when joining slices via their TSVs you get a dead stack thats worthless.
Its not like Hynix and Samsung are only making it in low volume because they feel like it. They yield what they can.

>> No.68788470

>>68788407
I see, thanks
>because packaging it onto CPUs or memory sticks is a massive engineering hurdle we haven't done yet
Felt like it would be too dumb if the high cost was only due to production runs

>> No.68788474

>>68788116
Apple literally opened an office right next door to Intel, offered them 3x the salary for working on A-series processors that ship every year and make great leaps in performance and process every year.

If I was an engineer at Intel, being held back by process delays year after year, I don’t see any reason not to switch to the more exciting and better paying job that doesn’t even require me to relocate.

>> No.68788475

>>68788407
I'd just like to interject for a moment.

HBM definitely has density issues. At best we get 2GB per stack and 8 stacks, and every HBM2 chip needs 1024 wires to connect to its processing element.

4 stacks give 64GB and 4096 wires that need to be routed and connected. It's an absolute nightmare.
HPC demands as much RAM as can fit into a board at times. 256GB per socket is common now and it's not unusual to see 512 in database heavy clusters.

HBM, again, definitely has density issues when compared to the bandwidth it supplies.

>> No.68788485

>>68787521
>5.1 boost clocks
kek. either housefire or fake

>> No.68788492

>>68788437
Even if they're fake, the numbers don't seem that far-fetched when you have very small chiplets with an incredibly high first-gen yield percentage.

>> No.68788497

>>68788349
Because it just works.

>> No.68788507

>>68788474
I was just saying that the process delays and lack of innovation were there before those engineers started to leave

>> No.68788513

>>68787521
Maliciously fake and flagrantly homosexual. This is just a reformat, with the more obviously pulled-from-ass BS tweaked, of equally-fake table that's been doing the rounds for the last week.

>> No.68788514

>>68788492
it won't be chiplets.

>> No.68788534

>>68788507
Of course, but the brain drain has probably further delayed it considerably.

>> No.68788537
File: 769 KB, 1221x768, 1522587307134.gif [View same] [iqdb] [saucenao] [google] [report]
68788537

>>68788462
Yeah but unlike intel's 10nm retardation, TSMC 7nm just works.

>> No.68788552

It just works.

>> No.68788554

>>68787959
>paying Jewtel shekels
>to gayme
It's true - you Intcels have absolutely no life. I'm not even entertained by you. I pity you now.

>> No.68788582

>>68788124
You're going to hate me and my system:
>12GB RAM
>3TB hard disk
>Vega 56
But my CPU is quad core, so there's that.

>> No.68788589

THIS IS NOT FAIR!

What are we gonna do Intelaviv bros?

>> No.68788591

>>68788554
It's mostly indians with core2duos, /g/ is their designated shitting loo.

>> No.68788593
File: 2 KB, 184x156, 1312093346312.png [View same] [iqdb] [saucenao] [google] [report]
68788593

>>68788513
Well that's what we're gonna find out about isn't it

>> No.68788595
File: 71 KB, 579x411, 1334737117601.jpg [View same] [iqdb] [saucenao] [google] [report]
68788595

>>68788457
>>68788437

>> No.68788607

>>68788595
don't ever reply to my comment ever again

>> No.68788609

>>68788595
My dad taught us not to be ashamed of our yields.

>> No.68788627

>>68788582
You're right, I hate you and your system.

>> No.68788637

>>68788399
It just works.

>> No.68788647
File: 1.46 MB, 500x298, 1449936331959.gif [View same] [iqdb] [saucenao] [google] [report]
68788647

>>68788346
>still runs AotS, Far Cry and MS FSX like shit

>> No.68788651

>>68788462
7nm HP has a much lower density than the LP process used by Apple.
Presumably yields will be much better.

>> No.68788661

>>68787521
3600X is looking juicy for gaming with.

>> No.68788662
File: 99 KB, 2048x1175, 1527105407832.jpg [View same] [iqdb] [saucenao] [google] [report]
68788662

That R7 3700x looking like my pick.
i5-3570k to Ryzen 7 3700x gonna be a huge upgrade.

>> No.68788665

>https://en.wikichip.org/wiki/amd/microarchitectures/zen_2

Key changes from Zen+

Security
In-silicon Spectre enhancements
Increase number of keys/VMs supported
I/O
PCIe 4.0 (from 3.0)
Infinity Fabric 2
2.3x transfer rate per link (25 GT/s, up from ~10.6 GT/s)


OH NONONO GOYIM

>> No.68788675

>>68788627
You should only 75% hate me and my system.

>> No.68788677

>7nm
>not waiting for picometre

>> No.68788682

>Ryzen 9
i came

>> No.68788685
File: 1015 KB, 841x745, 1334737161450.png [View same] [iqdb] [saucenao] [google] [report]
68788685

>>68788607
>>68788609

>> No.68788693

too make yourselves appear less stupid in the future (I am screenshoting and saving this thread) I advise you to start each post with

>If these leaks were true

>> No.68788707

>>68788462
>Its expensive because its complex, its stacked DRAM slices. If anything goes wrong when joining slices via their TSVs you get a dead stack thats worthless.
>Its not like Hynix and Samsung are only making it in low volume because they feel like it. They yield what they can.
While technically correct, that's still a matter of how much of DRAM is dedicated to go to HBM stacks. There's more than enough memory printed to have high-volume, low-price HBM if we switched all current DRAM production lines to HBM. The fact that the yields are lower IS a problem, but one that could definitely be solved by throwing engineering at it. DRAM is an INCREDIBLY mature and high-yield memory type, HBM is still new and hasn't had as much work put into it. I'd guesstimate that with a few more years, these problems aren't going to be as bad.
>>68788475
The ROUTING is the real nightmare, and why it hasn't penetrated the market. HBM is incredibly difficult to integrate into currently existing market types. GPUs already had fat memory busses, so switching to HBM wasn't as big of a feat as tying a CPU to it. And even then, they use massive, low-yield passive interconnects printed not on PCBs, but on silicon. While, again, this is a problem that could certainly be solved by throwing sufficient engineering talent at it, it is still the primary reason it hasn't been integrated into other market categories.
I wouldn't say it has density issues, though; 4 8GB stacks are on-package on the new MI60. Per mm^2 space, HBM is a massive improvement. It really is just the routing and packaging issues that need to be solved, which are problems with any new technology. I still want HBM to become the mainstream memory solution, and I think it's feasible; just not easy.

>> No.68788732

>>68788651
Look up what makes a transistor favor lower vs higher clocks. High performance xtors have more insulation, they're thicker, and take more passes to produce. Everything from the source and drain wells is different. HP parts do not magically mean higher yields. At all.

>> No.68788760

>>68788537
intel's what?

>> No.68788763

>>68788707
HBM isn't made with the same generic DRAM that your desktop DIMM is, guy. They don't take complete DRAM dies then put TSVs into them. HBM DRAM is made on its own wafers.
How the dies are stacked and packaged is what reduces the number of working stacks and keeps production limited.

>> No.68788779

>>68788078
Might as well just wait for zen 3+

>> No.68788788
File: 60 KB, 524x331, 1378190867137.jpg [View same] [iqdb] [saucenao] [google] [report]
68788788

>>68787521

>> No.68788798

>>68787817
Believe in I/O die. Zen+ WX TR had too many dies and not enough memory controllers.

>> No.68788812

>>68788693
>make 64 core server cpu
>only 8 cores for mainstream
You know that's bullshit. It's gonna be 16 cores, two 8 core chiplets.

>> No.68788836
File: 89 KB, 534x299, received_338481236710224.png [View same] [iqdb] [saucenao] [google] [report]
68788836

>>68787959
STOP FUCKING KVETCHING AND JUST BUY IT YOU STUPID GOY

>> No.68788849

>>68788677
Look up the width of an atom some time.

>> No.68788861

>>68788763
Right, apologies. I didn't mean to imply that in my post. What I MEANT was how much of the DRAM production lines were dedicated to HBM instead (since they're usually produced on the same lines).

>> No.68788862

>>68787521
>16c @ 5.1Ghz
lol sure thing cuck

>> No.68788866
File: 59 KB, 769x697, 1532405925165.jpg [View same] [iqdb] [saucenao] [google] [report]
68788866

>>68787521

>That 3700x

>> No.68788890

>>68787521
These boosts are before XFR, right?

>> No.68788909
File: 295 KB, 419x419, 1537765997528.png [View same] [iqdb] [saucenao] [google] [report]
68788909

>>68788021
I got a 1700 and a NH-D15 for that very reason, but holy fuck I want that 3850x now fuuuuuuuck.

>> No.68788914

>Still no TSX
>Still no LPDDR support
Well shit, I hope someone will tell them eventually

>> No.68788924

>>68788582
STOOOOOP

>> No.68788930
File: 16 KB, 380x544, it's time.jpg [View same] [iqdb] [saucenao] [google] [report]
68788930

>>68787521
>16c
>5.1ghz boost
If they release a 32c Threadripper with 5ghz boost I'm going to shit myself laughing.

>> No.68788942

>>68788914
>LPDDR
For literally what purpose. DDR is already slow as it is.

>> No.68788944

Highly doubt this is true and pulled from his ass.

>> No.68788947

>>68788914
Dude, you can use low power DIMMs in anything you want. The IMC isn't going to shit itself because you want to run 1.2v DDR4 instead of 1.35v.
TSX is a meme used for nothing but some PS3 emulator.

>> No.68788950

>>68788930
I mean, they could just go whole ham and slap out a 64 core threadripper since, they're both effectively going to be the same

>> No.68788963
File: 17 KB, 528x404, 1542567212255.png [View same] [iqdb] [saucenao] [google] [report]
68788963

AMD drones are more happy about the Price than clocks and TDP.
Means this leak is fake as fuck.

>> No.68788981

>>68788485
10nm by 2020?

>> No.68788983

>>68788963
your post means you are fucking retarded

>> No.68788986

>>68788942
Because the vast market lays in laptops which AMD completely ignores. When anandtech asked them
>When we asked why AMD is not supporting LPDDR memory on Ryzen Mobile, the response was related to performance. If we look at almost every single Intel 15W notebook on the market, despite the processors supporting both LPDDR3 and DDR4, they all use LPDDR3 by default. If we ask the OEMs, the answer lies in the power consumption during lower power modes such as idle: LPDDR memory can achieve much, much lower power modes than standard DDR. We put this to AMD and they looked confused, saying that literally none of their OEM partners had ever brought this up in conversation as a requirement for a future platform.
AMD managers are as clueless as they can get

>> No.68788990

>>68788983
he was just pretending

>> No.68788995

>>68788947
LPDDR uses different protocol, it's not ddr3l

>> No.68789001

>>68788981
10NM SUPERPOWER

>> No.68789006

>>68789001
Incel is poo now?

>> No.68789008

>>68787521
Pricing makes no sense. They ought to be way cheaper.

>> No.68789025

>>68788986
Those kinds of laptops aren't getting 6 core CPUs anyway, maybe they'll add the support if they decide to branch out into memebooks.

>> No.68789033

>>68789008
They will be. No reason to charge so little if they're going to murder Intel either way.

>> No.68789042

AMD doesn't have thunderbolt so they're basically useless for people doing real work

>> No.68789062
File: 3.42 MB, 4366x2638, 1543758753426.png [View same] [iqdb] [saucenao] [google] [report]
68789062

>>68789006
Indians love Intel.

>> No.68789065

>>68789042
*maintenance of the deprecated systems work

>> No.68789066

i swear u guys are gonna jinx zen 2 like vega. It'll be a perf jump sure, but I dont think it'll be as massive as people think. I got an 8700K because i needed the clock speed and single core perf as there are STILL SOME SHIT SOFTWARE THAT RELIES ON 2 CORES such as some CAD/Business apps. AMD fooled me bigly so im not gonna buy into the hype yet.

>> No.68789069

>>68789042
""""""people""""""

>> No.68789081

>>68789025
So what? I'm fine with 4 cores. I don't want intel garbage in my laptop.

>> No.68789084

>>68787521
Why is there no 8 core part that clocks to the moon? Why is there no 16 core with an iGPU? The prices will be either way lower or way higher.

>> No.68789094

>>68789042
>intel shills getting so desperate now that not even muh singlecore is a valid excuse anymore that they move the goalposts onto meme shit like thunderbolt as the sole reason for continuing to remain cucked by Intel

>> No.68789096

>>68789081
4 cores is enough -intel

>> No.68789100

>>68789096
4 cores? What do you need 2 cores for?

>> No.68789102
File: 97 KB, 433x419, 1521721293117.jpg [View same] [iqdb] [saucenao] [google] [report]
68789102

>>68789062
>intel fans are all poos and chicoms

>> No.68789115

>>68789100
4 cores is enough on a laptop. 6 cores is just battery hogging

>> No.68789123

>>68789066
You underestimate the handicap GF process put on Zen. Even if Zen 2 was simply fixed Zen the new process would do wonders.

>> No.68789125

>>68788849
Lmao just shrink the size of a proton, neutron, and electron

>> No.68789127

>>68787521
>almost 200 replies
>not even 60 posters
It sure as fuck smells like plebit/amd here.

>> No.68789128

>>68789062
American micro devices

>> No.68789135

>>68789081
>So what? I'm fine with 4 cores.
Do you see any 4 core CPU in their lineup?

>> No.68789140

>>68789084
Why in the fuck would you be paying $500 for a CPU and not getting a discrete GPU?
There's also probably die space constraints on the higher core parts

>> No.68789141

>>68789135
Do you see any mobile CPU in their lineup?
I'm getting tired of this conversation

>> No.68789145

>>68789125
They're already fucking point-like, that's not where the size of the atom comes from.

>> No.68789157

>>68789141
>Do you see any mobile CPU in their lineup?
That's what I have been saying since the beginning. They don't support LPDDR because they are not intended for ultrabooks.

>> No.68789161

>>68787521
It'll be gimped AF by shenanigans like IF and lack of optimizations.

>> No.68789170

>>68787521
>99 US shekel
God i wish ram prices cheap enough by the time this released.

>> No.68789173

>>68789157
2700U and 2500U were advertised for ultrabooks but still didn't support LPDDR. LPDDR support is a feature of memory controller and neither intel nor amd design different memory controllers for desktop and mobile chips.

>> No.68789175

>>68788707
>4 8GB stacks are on the MI60
Nvidia put 32GB on their newest 6000 as well. It's still just 4 stacks at 1GB per slice for 4096 wires.
If we magically assume HBM next year can quadruple density and achieve 32GB per stack, that's still 8000 wires which need to be routed just to provide 256GB. It's beyond an "engineering challenge", the sheer complexity of the signal routing just isn't viable.

HBM is a boutique solution for very specific applications, where RAM amount isn't necessarily a factor but bandwidth requirements are very, very large. The only things it really could be used for are GPU VRAM or possibly off-die L4 caches.
We're all still waiting for an AMD APU with HBM2 to feed the graphics, and most people expect it would only be a 2GB chip at most. Give me that and I'll start to think about integrated HBM for consumer CPUs

>> No.68789186

>>68787521
When will they double the core count again? With 7nm EUV, 5nm or 3nm GaaFet? Will it be with Zen 3, 4 or 5?

>> No.68789200

>>68789173
Their iGPUs would be fucking crippled though. But yeah, a strange oversight.

>> No.68789202

>>68789170
>ram prices
>ever coming down
ever since that one fire the companies that make ram realised no one was going to stop them from selling ram at disgustingly high prices

>> No.68789227

>>68789202
I hope they all get sued to oblivion. Literally holding humanity back those fuckers.

>> No.68789241

>>68787521
>5GHz boost
if true it's going to be the modern day holocaust

>> No.68789252

>>68789227
Never goy, ram prices will literally never reach pre 2013 levels again

>> No.68789257

>>68789200
LPDDR3 has like 85% of DDR4 performance. The gains lie in idle power consumption - it's like 10% of DDR4.

>> No.68789258

3700X + 1070 @1440p
Yeah or nah?

>> No.68789261

>>68789202
I thought chink going to crash the market?

>> No.68789266

>>68789258
>1070
Nah, but if you must, then OK.

>> No.68789271
File: 6 KB, 222x228, gaming.jpg [View same] [iqdb] [saucenao] [google] [report]
68789271

>>68787959

>> No.68789287

>>68789266
I already own it. Upgrading CPU.

>> No.68789302

>>68788812
This! All the dies with >8 cores will be used in ≥36 core epycs, while PC consumers get the most faulty silicon with half or more cores dead. This begs the question if we'll see ≥8 core dies coupled with navi die in ryzen.

>> No.68789303

>>68789084
>Why is there no 8 core part that clocks to the moon?
Good quality 6 core dies are binned into 3700 and 3700X
> Why is there no 16 core with an iGPU?
Each core die is 8 cores.There is room on the package for 2 dies. You can either have a core die and a dummy die, a core die and a GPU die, or two core dies.
>The prices will be either way lower or way higher.
Why?

>> No.68789320
File: 10 KB, 250x250, 1516703566590s.jpg [View same] [iqdb] [saucenao] [google] [report]
68789320

>>68789062
Only mildly surprised to find my cuntrymen are also intcel losers.

>> No.68789322

>>68789062
Are pajeets seriously still buying Core 2 Duo?

>> No.68789325

>>68787521
>3600 series
damn that one looks really sweet

>> No.68789332

>>68789123
Nah i looked those up it should be an improvement across the board for power, efficiency, etc. im just not so sure yet that it'll eclipse intel by a whole lot if at all in ipc/clock speed Imo if they meet near parity which is pretty much confirmed at this point because going further than 4.7 results in diminishing gains. I spent alot on this fucking cpu in keepin it cooled down and stable but I cant really say even if OPs pic is 100% true that Im dissatisfied in the perf. My next cpu would prob be AMD but 8700K is probably good for another 3-4 years or longer because of how things are becoming more GPU bound idk. I used to rock team red but last year i got this chip for $310 and Vega is well... vega. I got a used 1080ti for 550

>> No.68789350
File: 655 KB, 1920x1200, 1502261452790.png [View same] [iqdb] [saucenao] [google] [report]
68789350

>>68787822
>Zen 2 Athlon
>Same chiplet design
>Die one has 1/8 cores alive
>Die two has 1/8 cores alive

Cute

>> No.68789392
File: 98 KB, 1024x652, 2005_MX-5_Miatafront-7-8-beauty-1024x652[1].jpg [View same] [iqdb] [saucenao] [google] [report]
68789392

>>68789350
>NB Miat

Cute

>> No.68789425

These 7nm cpus are really starting to get me hyped. How much would 2133 ram hold me back if I transferred it to a new build? From what I understand, AMD is more dependent on ram speed.

>> No.68789446

>>68787822
no you dork

>> No.68789450

>>68788862
First stage of grief: denial

>> No.68789453

>>68789425
Yeah, Ryzen loves fast RAM. 3200+ would give you a significant jump in performance.

>> No.68789478

>>68787843
>high-end
I like your sense of humor

>> No.68789480

>>68788862
it's actually two 8c dies, and we know fore 100% sure that 88c can go to 5.1Ghz

>> No.68789482
File: 3.47 MB, 480x270, 9f2.gif [View same] [iqdb] [saucenao] [google] [report]
68789482

>>68787521

>> No.68789524
File: 104 KB, 1280x850, UwU.jpg [View same] [iqdb] [saucenao] [google] [report]
68789524

>>68789392

>> No.68789548

>>68788947
though given how advantageous tsx is in that emulator I don't see it not being relevant for other things in the near future

>> No.68789552

If this is true AMD are going for Intels jugular (and taking a small slice of Nvidia's low to mid tier pie whilst they are at it).
Vendors won't be able to resist putting AMD parts into their prebuilts even with Intel bribes.
It's going to be annudah shoah.

>> No.68789564

>>68789127
and how would you know

>> No.68789607

>>68789161
>t. literal brainlet

>> No.68789641

>>68789186
actual node-shrink
maybe
not sure if 5nm is a half node or not

>> No.68789678

>>68789175
>that's still 8000 wires which need to be routed just to provide 256GB. It's beyond an "engineering challenge", the sheer complexity of the signal routing just isn't viable.
Welp. Completely forgot that the 1024 pins were per-die, and not per-controller. My bad.
>HBM is a boutique solution for very specific applications, where RAM amount isn't necessarily a factor but bandwidth requirements are very, very large. The only things it really could be used for are GPU VRAM or possibly off-die L4 caches.
>We're all still waiting for an AMD APU with HBM2 to feed the graphics, and most people expect it would only be a 2GB chip at most. Give me that and I'll start to think about integrated HBM for consumer CPUs
Fair enough. Right now their biggest 2 applications are definitely L4 and APU memory. There have been papers researching them as L4 that showed HBM1 as a viable candidate (40ns latency, huge bandwidth) with 30% gains in memory-bound processes on a simulated CPU. I still think HBM L4 has a CURRENT place in HEDT/server platforms, though; most of the bottlenecks in current performance are due to memory bandwidth and latency. We can crunch numbers absurdly fast, we just have difficulty feeding those number crunchers.

>> No.68789687

>>68789186
>yfw this isn't 7nm's final form
Hopeful for the future, will probably upgrade from 7nm to 5nm + new architecture desu.

>> No.68789705

>16c on AM4
I still don't think this is viable. There's only enough bandwidth in 2 DDR4 memory channels to feed 8 cores doing generalized workloads (almost anything but rendering), and only JUST at that. Unless they're somehow magically amping X570/B550 up to 4 channels, the 3850X is going to be almost as bandwidth starved as the 2990WX.

>> No.68789714

These leaks may not be far off the truth. Look at what EPYC 64 is doing. True scaling from Naples 32 cores and then some. It's actually double the speed on their demo but could be even more on final production silicon due to process and design improvements.
None of these seem unreasonable on TSMC's 7nm node. AMD are not going to CES this year just to bluster.
Intel, prepare your anus.

>> No.68789726

>>68789332
It's so funny how you talk about your system so much and nobody gives a shit. Your "8700 will be good for another 3-4 years... Right?" is even funnier. Yes it will be fine. Now fuck off.

>> No.68789733
File: 35 KB, 1560x230, Screenshot from 2018-12-04 18-55-15.png [View same] [iqdb] [saucenao] [google] [report]
68789733

More confirmation from the HardOCP thread.

>> No.68789736

>>68789482
The FINGLONGER!

>> No.68789743

>>68787521
>Ryzen 5 3600X 8C/16T 4.8 GHZ
INTEL ON SUICIDE

>> No.68789758
File: 33 KB, 258x245, 41241.jpg [View same] [iqdb] [saucenao] [google] [report]
68789758

>>68789062

>> No.68789760

>>68789733
and he would know why?

>> No.68789769

>>68788513
Yeah, the pricing alone is unbelievable.

>> No.68789785

>>68789322
Just did a currency conversion, a core 2 duo in indian money is like 9 dollars.
Idk if that's as cheap as it sounds considering earnings in india but maybe it is

>> No.68789793

>>68789760
Because he is El Chingón.

>> No.68789798

>>68789733
If Kyle's buying it, it's usually true. HardOCP generally gets good damn leaks.

>> No.68789820

>>68789705
The video said the 16c processors may be limited to the b550 x570 (?) chipsets because of the tdp while the rest would remain compatible.
>>68789733
I think he's reliable. Plus if the information is an actual leak, that means it is circulating out there somewhere. Other people may have it and not be leaking it. Also, the person who leaked it may have made slight, intentional mistakes so it couldn't be traced.

>> No.68789830
File: 7 KB, 225x225, gengar.jpg [View same] [iqdb] [saucenao] [google] [report]
68789830

>>68789760

>why would Kyle from HardOCP have any idea whats going on in the consumer hardware industry

>> No.68789873

>>68789760
Longtime insider knowledge.

>> No.68789881
File: 1.48 MB, 350x263, thumbs up from jesus.gif [View same] [iqdb] [saucenao] [google] [report]
68789881

If these leaks are even halfway accurate this is big. If I can upgrade my 4690 to an 8 core 16 thread r5 that clocks up to nearly 5Ghz I will be over the moon. 2019 will be the worst year in intel's history.

>> No.68789898
File: 98 KB, 1272x127, Ryzen 5.jpg [View same] [iqdb] [saucenao] [google] [report]
68789898

>>68789881
>8c/16t 4.3 GHZ
>20 Vega Compute Unit
>$199
>possibility of overclocking to 5Ghz (but most likely 4.7 Ghz)

>> No.68789911

>>68789830
never heard of the guy

>> No.68789912

>>68789820
>The video said the 16c processors may be limited to the b550 x570 (?) chipsets because of the tdp while the rest would remain compatible.
Yeah, I'm saying that unless they make B550/X570 somehow compatible with both dual and quad channel memory controllers, the 16c part is going to be stuck with 2 memory channels. DDR4 can currently feed ABOUT four cores per memory channel when it's OC'd past 2133MHz. Unless Zen 2's memory controller can get DDR4 to 4GHz+ reliably, the 3850x is going to be about as good as the Threadripper 3950x in rendering but much worse in every other capacity, including gaming.

>> No.68789913

>>68788100
That won't work since you won't hit the thermal limit. It's the arch that defines clock on Ryzen.

>> No.68789914

>>68789733
I bet what Jim's wrong about is the IO die cutting or the way he's suggesting having all those controllers. Surely AMD has some trick up their sleeve.

>> No.68789928

>>68789332
>He thinks it's just a clock speed bump
Oh you poor delusional fool.

>> No.68789934

I'm going to come in with a bit of a dampener: if Zen 2 hits 4.6GHz at the highest end stock (just as Zen+ hits 4.4GHz stock), and a bit more IPC, I'll be happy. The current chips are very good. They don't need much more to at the very least match the competition.

>> No.68789938

Someone tell me how the fuck they're going to feed 16 cores with only dual channel memory. I'm dumb, but think this is a problem.

>> No.68789943

>>68787521
>12 cores
>5GHz
Holy fuck, I'm so glad I fell for the AyyMD shills and got a good x470 I can upgrade with in a year or two

>> No.68789957

>>68788259
Its a 10-15% increase in speed over 2xxx. Not that far fetched really.

>> No.68789980
File: 162 KB, 1000x1000, intel poo in da loo.png [View same] [iqdb] [saucenao] [google] [report]
68789980

>>68787521
intel on suicide watch

>> No.68789996
File: 92 KB, 1173x879, nuclear-weapon-bomb-castle-romeo-1.jpg [View same] [iqdb] [saucenao] [google] [report]
68789996

Say the word. SAY IT!

>> No.68790001

>>68789957
It's going to be equivalent to a 2700X at 5.15Ghz with Intel's AVX2 performance. And on top of that, we can overclock.

Zen2 is the absolute monster we all wanted.

>> No.68790005

>>68789912
Oops, I responded a little too quick. Is it reasonable from a cost or compatibility standpoint that they would double the memory channels? Or would this cut into their threadripper market?

>> No.68790012

>>68788675

Kekd

>> No.68790078
File: 75 KB, 1080x787, 1516239130508.jpg [View same] [iqdb] [saucenao] [google] [report]
68790078

>>68787521
VELLY GOORD, AMD MAKE GLEAT PLOCESSOLR

>> No.68790085

>Ryzen 3 3300
>6C/12T
>only $99

This seems way too good to be true. It's practically highway robbery but seeing how the exact same company made halfway decent 6C/12T processors more widely available for sub-$300 I wouldn't be surprised if this is actually isn't total bullshit.

>> No.68790118

>>68789733
Okay, this is epic.

>> No.68790143
File: 79 KB, 1000x1000, 1541534505904.jpg [View same] [iqdb] [saucenao] [google] [report]
68790143

I don't want to believe this, mostly because I don't want to get my hopes up. Will Amada really give us 16c/32t for $450?

>> No.68790151
File: 424 KB, 480x270, 1542278432448.gif [View same] [iqdb] [saucenao] [google] [report]
68790151

>>68787822

Fucking lol. Quadcore is dead. 4c/4t and 4c/8t is dead (if above leaks prove to be true). 6c/12t is the new baseline. Not even 6c/6t exists in this timeline.

Intel:
https://www.amazon.com/Intel-i9-9900K-Desktop-Processor-Unlocked/dp/B005404P9I?ref_=w_bl_hsx_s_pc_web_2529650011
>8c/16t
>95W

AMD's answer:
>R5 3600 3.6GHz base w/ 4.4GHz boost @ 65W for 179.99

That's 67.2% price difference (less) for basically 95% performance while being 30W less TDP

Intel:
https://www.newegg.com/global/au-en/Product/Product.aspx?Item=N82E16819117811
>12c/24t
>140W

AMD's answer:
>R7 3700 3.8GHz base w/ 4.6GHz boost @ 95W for 299.99

That's 23.68% higher clocks with 35W less TDP with parity or superior performance for 70.29% price difference (less)

Intel:
https://www.amazon.com/dp/B0775X4FFV/ref=psdc_229189_t2_B079CBH65X
>16c/32t
>165W

AMD's answer:
R7 3800X 3.9GHz base w/ 4.7GHz boost @ 125W for 449.99

That's 26.3% higher clocks with 40W less TDP with parity or superior performance for 75% price difference (less)


This SO catastrophically undercuts Intel's market & margins, if it ends up being true, their stock prices will be like a fucking brick out of a plane at 36,000 feet. Intel would h ave to reduce their margins by 70.8% PER chip anon. PER FUCKING CHIP SEGMENT. To match AMD (if Zen2 matches performance or exceeds it) minimum or maximum by 50% if it comes to be within 95% of the i9 offerings in every segment.

Intel's 16c/32t chip costs EIGHTEEN HUNDRED DOLLARS sans tax. A 50% drop is $900, a 70% drop is $1,274.39 mark down. This would be selling chips at a loss WORSE than Microsoft selling the 360 at a loss due to RROD and Sony selling PS3 at a loss because of cell COMBINED.

Hoooly shit.

>> No.68790179
File: 20 KB, 284x285, 1506992834466.jpg [View same] [iqdb] [saucenao] [google] [report]
68790179

>>68787521
HAHAHAHAHAHAHAHAHAHAHAHAHAHA
>3600X
>8C/16T
>4.8Ghz
>$229
Imagine how the people who bought a 9900k are going to feel when these are announced. The anti-semitism is going to be delicious.

>> No.68790203

>>68790005
From a cost standpoint, no problem. Adding 2 more memory controllers is absurdly cheap on the chiplet design. Compatibility is where it stumbles. AM4 was designed for 2 memory channels, and it can only have 2. Wiring motherboards for 4 AND 2 is, to my knowledge, an impossible or VERY difficult task, and would massively increase prices for B550/X570 boards.

>> No.68790218
File: 87 KB, 800x600, 1543628143979.jpg [View same] [iqdb] [saucenao] [google] [report]
68790218

>>68789958
>8c/16t APU
Maybe if it falls in price a smidge and has confirmed GNU/Linux compatibility I'll pick one up. I like the idea of it, and then I'd have a spare 2600x to shove into a server or sell to one of my friends desperate to upgrade.

>> No.68790282

>>68790203

They can technically launch in the same board ### segment with quad channel support. There's nothing that would stop them from doing that. For majority of people and OEMs, dual channel for Zen2 is satisfactory in performance; whilst the upper tiers would benefit from quad channel support in AM4 spec.

It doesn't break socket compatability to do this. Zen and Zen2 conclusively also gain a lot of performance from proper memory timings than they do channeling or speed. That's not to say that additional RAM channels feeding the CPU isn't good. But something like 3200MHz CL14 with very tight timings can easily match 3800MHz performance. You only see another major jump forward when you move up to 4 and 4.2GHz DDR4 memory speeds with standard timings.

Same way, all AMD has to do with Zen2 for dual channel configurations is recommend something like 3400MHz CL14 with extremely tight timings (which will mean you'll have to spend a bit more on memory), but in turn that would give you as nearly matching performance as 4000MHz CL16 quad channel.

>> No.68790301

>>68789938

It isn't if you understand how the memory controller on the ccx works.

>> No.68790348

Forget Moore's law. We now have MOAR CORES LAW!

>> No.68790369

Someone have that Rome with the H on it?

>> No.68790377
File: 838 KB, 580x742, Epyctler.png [View same] [iqdb] [saucenao] [google] [report]
68790377

>>68790369
this one?

>> No.68790394
File: 700 KB, 360x203, Now I've lost it - 4th.gif [View same] [iqdb] [saucenao] [google] [report]
68790394

>>68790377
Bless you anon. For some reason I lost it.

>> No.68790405
File: 816 KB, 1070x601, epyc suicide.png [View same] [iqdb] [saucenao] [google] [report]
68790405

>>68790394
you're welcome
I have this one too

>> No.68790420
File: 357 KB, 1920x600, 1515246455913.png [View same] [iqdb] [saucenao] [google] [report]
68790420

>>68787521

>> No.68790429

>>68788079
>>68788124
Holy shit are you me! I thought I was the only one that was a turbo autist about this sort of thing.

>> No.68790455

>>68787738

Then you didn't watch the video because his first source is the same fucking chart as the 5ghz Epyc, he even shows that part of the image near the end.

>> No.68790526

>>68788358

Yeah remember when HBM failed to do shit against Nvidia? Same will happen with 7nm.

>> No.68790563

Y'all over hype shit then make excuses when it doesn't meet expectations.
8 cores at 5.0ghz
Until I see real benchmarks from 3rd parties, I'm calling bullshit.

>> No.68790590

Guys, this isn't a leak, it's a prediction.

>> No.68790595

>>68790563
This, I look forward to all the back peddling when zen 2 flops.

>> No.68790605

>>68787521
DELID THIS!

>> No.68790619
File: 649 KB, 902x1294, Plz no bully.png [View same] [iqdb] [saucenao] [google] [report]
68790619

So what happens when Intel caves in and uses 7nm for their Skylake cores?
They can already clock 5.1ghz on 8 cores with 14nm.
Would that mean a 6.0ghz clock would be possible?

>> No.68790630

>>68790595
>peddling
non-english merchant spotted

>> No.68790637

I just bought a 9900k, fuck

>> No.68790644

Calm down guys, it's fake as fuck.

>> No.68790651

>>68790563
>>68790595
Don't worry, Keller will make a you a new architecture due 2022, just wait.

>> No.68790670

>>68787521
>8C / 16T is 4.0GHz, 4.8GHz boost
>12C / 24T can be 4.2GHz, 5.0GHz boost
Why? I don't need more cores, just clock it up. Now I don't know which one should I take.

>> No.68790675

>>68790619

>skylake on 7nm?

Nothing, its Skylake on 7nm. All its problems with Spectre & Meltdown will come along.

>5.1GHz on 8c w/ 14nm

Yeah, for a 140W TDP, god damn nuclear reactor on glass.

>6GHz possible

Sure, you'll only need a $6,000 industrial chiller plugged into your CPU to pull it off.

>> No.68790692

>>68789980
*poo in joo

>> No.68790698

>>68788456
>1080p
It's almost 2019 stop buying low resolution monitors you idiots. The faster idiots stop buying 1080p monitors the faster monitor manufacturers will stop selling them.

>> No.68790723

>>68790619
Anon, the only reason they keep clocking so high is because Intel is on the most mature 14nm process on the planet making hundreds of thousands of chips. Just by pure chance, those chips will eventually clock higher than usual due to the sheer amount of them.

>> No.68790732

>>68787817
>Remember the 2990WX is slower than a 2700X in like 90% of tasks
That's a lie. https://www.phoronix.com/scan.php?page=article&item=amd-linux-2990wx&num=1

>, and loses to the 9900k in 99%.
Thats another lie. The 9900k gets BTFO by the 2950x in HEDT tasks, much less the 2990wx.
https://www.phoronix.com/scan.php?page=article&item=intel-core-9900k-linux&num=7

>> No.68790735

>AMDrones actually ate up the rumor from their favorite AMD shill
>they think that X370 boards can handle the 12 and 16 core Ryzens
(lol

>> No.68790750

>>68790619
>and uses 7nm
Too late, if they're using TSMC to make their new CPUs they're using 7nm+ or 5nm

Intel billions + Poo VP + Pooduri + Jim Killer would end AMD

>> No.68790752

>>68790735
my taichi sure as fuck can. I saw this coming 2 years ago.

>> No.68790775

>>68790735
>(lol
kys

>> No.68790787

>>68790670

>Why?

Because 7nm is a full node shrink over 14nm, 2x density, 2x perf, 0.5x power. Zen2 is linearly scaled Zen in application coupled with chiplet design philosophy. AMD doesn't have 300M chip yield capability like Intel with all its world over fabs anymore, they have to make it so that each and every single wafer can return 90%+ yields, with good dies that can be used segment over, partial dies can be used as well, and defectives dies as well. Previously partial dies were thrown in the trash and defective dies too. That's basically you paying anywhere from 5,000-12,000 dollars for a wafer and then throwing out around 3,000 dollars on immature node and ~1,000 on a mature node, worth of chips because of defective dies or partials. So that's a 25-50% loss per wafer.

Chiplet guarantees that when you put in 12,000 for a 7nm wafer, you get 11,000 back (immature node) and when you put in 5,000 on a 7nm's mature node, you get 5,000 worth of product back. Additionally, when you look at the leaked segments, the R3 3600 = 6c/12t? That's an 8c/16t chiplet with a defect affecting one of the cores and since asymmetrical cores aka 7c/14t or 9c/18t would be janky as fuck, that extra core is fused off. This gives you a 6c/12t part. Then, for any die that has a major defect, that becomes your dummy die that's thrown on the Ryzen 3xxx package.

6c/12t die + dead die + I/O die; sell for 100 bucks. The dead die is useless to you, but the presence of a dead die means that when everything is soldered (heat will spread out into the dead die and dissipate easier).

R5 8c/16t? You'd think these are single chiplets, but they're actually 2x 8c/16t chiplets with major defects affecting the other four cores. So in fact they're 4c/8t + 4c/8t + I/O die for your 8c/16t.

Next the R7s are your 6c/12t R3 dies but binned for best speed, which gives you a config of 6c/12t + 6c/12t + I/O die = $329.99 3700X.

Lastly, true 8c/16t chiplets = R9s; & max low power + solid clocks = EYPC.

>> No.68790798

I wanna believe it's almost another shoah happenings. Tfw I'm still holding my 20 core dual xeon rig. Im glad the age of corelet's are over.

>> No.68790805

>>68790752
Even the most expensive X470 boards can only handle 10-cores.

>> No.68790806

DELETE THIS ANTISEMITIC THREAD RIGHT NOW!

>> No.68790818

>>68790644
can confirm. work at AMD

>> No.68790836

>>68787521
This looks too good to be true. Even without any IPC improvement this would be amazing, since we know IPC is going to be better as well this just looks unbelievably good.

Well, even if it's fake my 4790K is still holding up remarkably well for my use cases, so I'll be waiting to see if this shit is for real. If it is, I'll very likely be upgrading.
>>68790670
>Why?
Better CPUs get better silicon, of course. The lower-end SKUs aren't going to be binned as well as the higher-end stuff.

>> No.68790848

>>68790787
So it doesn't sound like you can overclock the difference away, the larger core count has the best physical hardware. In first generation 1600X had larger default clocks than 1700X so it's weird how things have turned around. I guess I'll have to go 12 core next.

>> No.68790851
File: 324 KB, 813x1402, intelseethingpajeet.png [View same] [iqdb] [saucenao] [google] [report]
68790851

>>68787959
ipajeet

>> No.68790862

>>68790805
you dont know what you're talking about. new processes also means less power needs. Not to mention the tachi is actually bonkers overkill for anything today. VRMs on it can handle 300 watts

>> No.68790901

>>68787817
That's because literally half the cores on a 2990WX don't have direct access to RAM. Zen 2 CPUs don't work the same way at all.

>> No.68790903

>>68790851
*indiantel

>> No.68790914

>>68790420
kek

>> No.68790923

>>68787521
Finally. I'm poor (on western standard, but a middle class here) and I can afford rendering PC for cheap. Based AMD.

>> No.68790950
File: 80 KB, 500x500, lrn2fkngrmrfgt.jpg [View same] [iqdb] [saucenao] [google] [report]
68790950

>>68787959
Reminder of what? AMD being objectively superior is just about every way?

Thanks, I needed that reminder.

>> No.68790954

>>68790563
It will be 5Ghz on a couple of threads at most just like Intel does.

>> No.68790963

>>68790735
>Dumbfuck incel shills not understanding that Ryzen 9 is going to have quad channel memory support, and that he next Threadripper will have octa channel support on the 64c parts because AMD intends to fully press its advantage by increasing core counts at the same time it increases IPC with uarch revisions and increases clocks (or lowers TDP) with node shrinks.
lmoa

>> No.68790994

>>68787521
I'm new to this. Please explain what is going on here. From these threads I've seen AMD provides better returns for the money you chip in while Intel basically charges double for slight improvement in performance. Is that correct? In the OP's pic, what how do the latest Intel desktop processors compare in terms of performance and money? I am trying to build a PC and pretty much going with AMD. I want to play latest games in decent resolution and FPS. Can you guys share some of your configurations? I understand the processor, GPU, RAM, motherboard to an extent, hard disk and monitor are the only things to worry about?

>> No.68790997
File: 96 KB, 395x290, 1539306840388.png [View same] [iqdb] [saucenao] [google] [report]
68790997

I-I don't need it... I-I can wait for Zen3(+) with my 1600.

>> No.68791006

If their yields are too high we may end up in another situation where 2700's were being sold as 2600's.

>> No.68791009

>>68790994
suggest something with a budget of 1500$ (excluding the monitor, keyboard and mouse)

>> No.68791013

>>68788003
everything over 1080p is placebo for games

>> No.68791021

>>68791009
Go to the PC building general you dork.

>> No.68791026

>>68790848

I'm not really sure by what you mean by "so it doesn't sound like you can OC the difference away..."

Take the leaks with a grain of salt, and assume around a 10-15% disparity in final numbers of base & all-core boost. That's the recommended spec. Can you clock higher than that? Yeah, probably, another 1-150MHz is realistic to all-core boost. Which may bring it to parity with leaked numbers. Stability will be lottery, because each die is different no matter what. This is an unavoidable fact of silicon design and manufacturing.

All that said, Clockspeed isn't everything and believing in it is unironically dumb. That's like questioning why a 1.8GHz jaguar console CPU can BTFO a 4GHz Pentium 4. Numerically the P4 should be superior, even when the console core is considerably more capable. That's because of IPC, better cache design, interchip latencies, hit/miss in L2/L3, bandwidth to memory, etc; and instructions per clock capabilitiy. ALL THESE are factors in CPU performance.

The point of Zen, honestly, is to stop giving a shit about high clock dick waving, and focus more on very strong single-threaded performance AT base clock, and then tackling N-core scaling. Then if you have power and voltage left over, focus on high all-core boosts. This is economical, and a net positive.

Intel, because of their monopoly and billions in R&D, and the fact that they can print 300M chips to satisfy market demand, by the sheer density of SO MANY FUCKING DIES, can produce 4, 6, and 8c chips that can OC to 5GHz and they can then sell it to the market for whatever price they want.

>first gen 1600X had better clocks than 1700X

Yeah, cause 1700X = +2c/4t. Power/perf/watt are all important and there's a finite amount of all three. Its a balance. AMD wants to tackle N-scaling, while competing with Intel as much as possible on clocks. All that said, 3700 will be the sweet spot, X if you wanna experiment with OC. 3800X if you wanna ride it for 7 years & 50X for anniversary.

>> No.68791027

>>68791013
t. reslet

>> No.68791053

>>68790637
Imagine buying from Intel in this day and age. What a sad cuck.

>> No.68791073

>>68790997
Buy Zen2 and then Zen4 for 5nm desu.

>> No.68791092

>>68790997

Yeah, desu, I'd be dreading too. The core/thread + I/O thing is legit. If it shows up in PC Ryzen 3xxx market it pretty much confirms its going to be in the PS5 & Xbox Scarlet. CPU core + I/O + GPU (console).

This means that by 2020-2021, you want to either be on Zen2 or Zen3 minimum, because all games moving forward will be designed for this spec. It will be optimized for this spec and so on. When games are ported over, they'll as a result, run better on Zen than they will on Intel (though Intel will pull through by brute forcing the difference via raw clOcks). PS5 & Xbox Scarlet will have a life expectancy of 5-7 years, then another ugprade cycle; likely around the time Zen5 drops. At point which you'd probably see 16c/32 ala 8c/16 + 8c/16 + I/O + GPU in console spec.

In the coming years, you won't need CPUs that clock to eleventy-jiggahertz on a 1-2 cores to play games at 60fps because ports are ass. Your 3.5GHz Zen2 + I/O die config will satisfy pretty much all cases, even if the port is ass.

>> No.68791121
File: 125 KB, 960x878, 1502535551592.jpg [View same] [iqdb] [saucenao] [google] [report]
68791121

HOW COULD THIS HAPPEN TO ME

>>
Name (leave empty)
Comment (leave empty)
Name
E-mail
Subject
Comment
Password [?]Password used for file deletion.
Captcha
Action