[ 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: 57 KB, 500x500, AdoredTV.jpg [View same] [iqdb] [saucenao] [google] [report]
69262626 No.69262626 [Reply] [Original] [archived.moe] [rbt]



>> No.69262713
File: 250 KB, 534x488, ZEN.jpg [View same] [iqdb] [saucenao] [google] [report]


>> No.69262951

At this point they could have Lisa Su make fart noises on stage for 30 minutes while slides with Zen 2 appear and it'd still be better than what Intel just put out there.

>> No.69262956


>no sound

you massive faggot


Vega II is going to be a soft launch. The die is too god damn big and is really not worth it to the gaming audience. Maybe will be useful for HEDT markets. The other soft launch would likely be the 3850X, which is the anniversary chip (if leaks prove to be true), because said anniversary isn't till May and announcing it at CES to launch it in May would thereby make it a soft launch. Finally, they probably need a little bit more time to ramp up supply of the platinum sample Zen2 16c/32t dies to maybe find those extremely unique dies that can clock up to 4.8 or even 4.9GHz all core turbo while still keeping cool and not clocking high in power usage.

>> No.69263034
File: 361 KB, 2000x1700, master suicide list.jpg [View same] [iqdb] [saucenao] [google] [report]


>> No.69263036
File: 173 KB, 1000x1000, 5657rugyu67.jpg [View same] [iqdb] [saucenao] [google] [report]

It's a fucking JIF, you dumb fuck.

>> No.69263052

Here's with sound. Lisa Su and AMD France liked the video on twitter btw

>> No.69263055

>inb4 he's wrong but because AMD chips are even better

>> No.69263061

These aren't news.

>> No.69263066


There's a webm of it out there you glue huffing retard.

>> No.69263070

.webms are too big for /g/, you dumb fuck.

>> No.69263107
File: 223 KB, 969x983, tim JUST.jpg [View same] [iqdb] [saucenao] [google] [report]

WCCFTech is garbage fake news don't click link please

>> No.69263111
File: 55 KB, 678x495, 54a6s5465as4df645sad65t7496a854.jpg [View same] [iqdb] [saucenao] [google] [report]


You posted a fucking link.

>> No.69263121

>more Vega bullshit

Just fuck off already and give us Navi

>> No.69263129

I don't have a link for .webm, you dumb fuck.

>> No.69263141

I'd pay for Lisa's fart noises on a vinyl.

>> No.69263146
File: 14 KB, 250x250, iaint.jpg [View same] [iqdb] [saucenao] [google] [report]


>> No.69263148
File: 8 KB, 211x193, 1485473031601.jpg [View same] [iqdb] [saucenao] [google] [report]


That sounds like a personal problem to me you god damn troglodyte.

>> No.69263156

How powerful will the Vega II GPUs be?

>> No.69263181

Over 9000 watts.

>> No.69263193
File: 324 KB, 882x758, 1502790639694.jpg [View same] [iqdb] [saucenao] [google] [report]


>> No.69263203

AMD will talk about 7nm Ryzen 3000 series dekstop (Zen2)

entire tech press BTFO because they said it was NO WAY that AMD would even mention Zen2 3000 ryzen desktop series

>> No.69263215

It's just a single, maybe pair, of $1k Vega20-based SKUs.

>> No.69263221

sorry i dont trust currytech vega ii is a hoax navi is real

>> No.69263250


>> No.69263268

CPUs will be shown and discussed, Vega II/Navi will be shown and discussed, Zen/Zen+ will get a pretty hefty price cut throughout the entire line (like AMD always does before releasing next gen) including even THREADRIPPER and THREADRIPPER+.

>> No.69263283
File: 37 KB, 474x415, 7654746754754776556.jpg [View same] [iqdb] [saucenao] [google] [report]

clear display of antisemitism, 9900K is God's chosen central processing unit for gaming!

>> No.69263292

A daily reminder latest non-reference 2080 Ti all have triple 8-pin setups on board. Radeon did four 8-pins only once and even back then it wasn't AMD themselves, but TUL's PowerColor's extremely over-engineered Devil 13 monstrosity.

>> No.69263296
File: 33 KB, 691x750, 1519151364375.png [View same] [iqdb] [saucenao] [google] [report]


>> No.69263315

There wasn't even that for the last 4~5 Lakes. Barely ~2%. And ALL of it was immediately lost with Meltdown.

>> No.69263368

this. The only "performance improvements" were because Intel continually raised base clocks with 100Mhz and said new generation was 5% faster even though it was no IPC gains, only clock gains (going from 3,2Ghz base to 3,3Ghz base etc)

>> No.69263387

They also improved avx and marketed it as abstract 15% performance improvement

>> No.69263390

Also faster memory, etc.

>> No.69263392
File: 188 KB, 400x600, 1545672612521.jpg [View same] [iqdb] [saucenao] [google] [report]

This but unironically

>> No.69263393

>platinum sample Zen2 16c/32t dies to maybe find those extremely unique dies that can clock up to 4.8 or even 4.9GHz all core turbo while--
TSMC's 7nm already has roughly 92% of PERFECT yields, you reality-denying Intbecile.

>> No.69263408

TSMC never discloses yield data.
Where did you get that again?

>> No.69263410

>92% of PERFECT yields


>> No.69263418

>improved avx
How is introducing and FORCING of the AVX 512 down people's throats, which is BARELY used by anyone and is essentially a vaporware, is an "improvement" any? It's the literal fucking PhysX of instruction sets.

>> No.69263430

>patch security holes
>clock your shit even higher to compensate
The state of "jewish" technology.

>> No.69263432

What I meant is they make up some special snowflake workload and optimize it, then market it as 15% improvement for everyone even though nobody gets that. You are overreacting.

>> No.69263433
File: 64 KB, 500x331, MAKOMINAKO.png [View same] [iqdb] [saucenao] [google] [report]


>> No.69263440

AVX512 is really good, acktually.
It just needs to be more widespread enough.

>> No.69263448

>AVX512 is really good
In 1.5 piece of a useless software-and-his-dog, out of thousands of programs and applications out there.

>> No.69263453

>92% of PERFECT yields


>> No.69263454

That's not the actual yield data for N7.
Get me an actual internal TSMC docs for N7 yields.

>> No.69263457
File: 52 KB, 500x970, rx90nitro_performance-per-watt_1920-1080.png [View same] [iqdb] [saucenao] [google] [report]

Unlike with AMD gpus Nvidia performance at least scales with power consumption.

>> No.69263466

It's not widespread enough to be widely used.
It's coming with Icelake and Zen3 for client and mobile, so the mss will increase rapidly.

>> No.69263476

Mobile when? My 2009 Sandy Bridge laptop needs to fucking rest already.

>> No.69263505

Mobile Zen2?
Early-mid 2020.

>> No.69263513

Ironically enough, even gen1 Zen, with that 14nm GloFo shit, had yields of roughly 65% in perfectly binned stones. When they moved to 12nm with Zen+, that drastically increased to 87% of perfect bins. And that's just GloFo garbage, kek. Then TSMC came in and now utterly BTFOs Inturd out of water, since Intelaviv's own fabs have barely ~47% with their 14+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++nm garbage, while 10nm is SO ABYSMAL, SUB-30%, that they had to STOP AND GUT IT WHILE REVERTING BACK TO 14nm and EVEN TO 16nm. This is the main reason why Farting Sound will be 14nm, even though INITIALLY they planned to make it 10nm. Intrash is fucked to no end.

>> No.69263514

how far away is ddr5 for the normal consumer?

>> No.69263525

>It's not widespread enough to be--
PhysX. Case and point.

>> No.69263529

Wait, is the shroud thing true?

>> No.69263531


>> No.69263536


>> No.69263542

AVX512 isn't locked to one vendor.
AMD will implement it in Zen3, probably including VNNI.

>> No.69263556

He should do asmr videos.

>> No.69263558

half of 10nm so July 1017

>> No.69263562

Zen 3 won't be anywhere earlier than 2022, though. Because it's now clear as a day that after Zen 2 we will have Zen 2+.

>> No.69263573


>> No.69263576
File: 32 KB, 474x472, 586u54u6h54698549.jpg [View same] [iqdb] [saucenao] [google] [report]


>> No.69263578

Zen3 is 2020.
There's no Zen2+ or anything of sorts.
Just more tocks.

>> No.69263593

not that i'm aware of. but he did ramble about how AMD is a poor people brand and you only buy AMD when you can't afford Intel/Nvidia during one of his livestreams (i think it was after threadripper had launched but not sure). and as for ninja i don't know either. i think the person who created the image has a personal vendetta against some of the listed items. most are true though

>> No.69263595

Either they're going to rename 3 to 2+, or they're going to do 2+ first and then delay actual 3 a little bit further, probably because of yet another massive sandbagging due to much better results with 3 than what they've expected initially (always remember than with gen1 Zen their own expectations were greatly exceeded).

>> No.69263615

No, Milan is Zen3.
There's no Zen2 refreshes.

>> No.69263619

I remember the threadripper thing. I was after he accidentally destroyed his new threadripper build and he was mad as fuck for a few weeks after that.

>> No.69263623

PhysX is open source now too, yet no one wants to touch that pile of filthy trash even with a 15-feet pole, because thew damage was already done and tarnished reputation will never be recovered. Soon same will be with DefectX 12.

>> No.69263640

>There's no Zen2 refreshes
On the roadmap so far it seems like there isn't, but Zen """""3""""" is supposed to be exactly just that - a 7nm once again with added bonuses from matured EUV processing. A literal refresh, just numbered 3 instead of 2+. And that might change in the near future.

>> No.69263650

No one wants to touch PhysX because everyone who wanted GPGPU physics already made their own custom implementation.
PhysX is just obsolete.

>> No.69263664

Eh, If Vega II is ~20-30% higher clocked than Vega 64, it'll be 20-30% faster than Vega 64, since Vega is front-end bottlenecked at 4 triangles per clock and thus raising the clocks raises the bottleneck and there was already plenty of shader performance trapped behind that bottleneck on Vega 56/64.

It's not revolutionary, but it's not nothing either.

>> No.69263666

It's a new core.
Very interesting one at that.
Every numbered core is a tock.

>> No.69263674

shut the fuck up amdrones it's going to be another refresh you get nothing

>> No.69263676

Assuming 56/64 CU's like Vega I, they'll be the percentage faster that they clock higher than Vega 56/64, due to the per-clock front end geometry bottleneck.

>> No.69263682

>It's a new core
7nm VS 7nm+ on EUV is now "new core"?

>> No.69263684
File: 30 KB, 216x250, 1498256245652.jpg [View same] [iqdb] [saucenao] [google] [report]


>> No.69263694
File: 13 KB, 400x400, 1539237186694.png [View same] [iqdb] [saucenao] [google] [report]


>> No.69263701

How the everloving fuck are nodes even remotely related to microarchitectures?

>> No.69263712


>> No.69263719

gtx 1080 power for $300

>> No.69263734
File: 10 KB, 325x325, 1546724641915.jpg [View same] [iqdb] [saucenao] [google] [report]


>> No.69263747



>> No.69263757

1060 for 100
1070 for 120
1070 Ti for 150
1080 for 200
1080 Ti for 250
2070 for 300
2080 Ti for 400
2x 2070 for 600
2x 2080 Ti = wait for NaVi

>> No.69263758

Nope. 1 day and 3 hours.

>> No.69263771


Nice try AMD shill

>> No.69263772


>> No.69263778

Go back to your /hw/ cesspool, 2ch.hk Ukrtrash.

>> No.69263779

If you care about ecelebs you should kill yourself, but here's what a quick "shroud amd" search brought me (and therefore brought you):

From what I can gather, he fucked up his threadripper build or something.

>> No.69263790


>> No.69263794

i wish but i doubt it

>> No.69263795

A cocaчepы тo тyт пpи чeм?

>> No.69263819


this link says otherwise


>> No.69263827
File: 10 KB, 251x201, 1538313408948.jpg [View same] [iqdb] [saucenao] [google] [report]

>even better

>> No.69263830

They're mostly brainwashed Inturd shillers and noVideo milking lolcows, so they hate strongly on AMD-anything. /g/ is different, as it loves AMD very much.

>> No.69263903

Never underestimate the jew conditioning.

>> No.69264038


>> No.69264156

I want to believe

>> No.69264165


2021 at the earliest, 2022 most likely

>> No.69264182

2021 for servers, 2022 for client.

>> No.69264279
File: 14 KB, 631x310, 1543201660279.png [View same] [iqdb] [saucenao] [google] [report]

>> No.69264481

It's used in academia and it's the only complete general purpose physics simulator until bullet 3 is done. Most games use custom single purpose sims with separate cpu hard body and gpu soft body if they need both

>> No.69264583

>it's the only complete general purpose physics si--

>> No.69264708

I dont think did general soft body, just a separate cloth (and similar) solver and their main component was hard body only. Physx before nvidia even started extending it had mesh deformation for stuff like crumple physics, non euclidean materials or whatever the gel like sit is called, etc.

The drawback of being so general purpose is that performance takes a hit. One of the Unity engine devs said he was able to create a cloth component that was like 2x faster because it handles just cloth. But physx supports almost everything out of box, they just didnt need it all

>> No.69264800

>I don't think
Yeah, exactly. You don't.

>> No.69265082


>> No.69265138
File: 40 KB, 400x388, 0007865.jpg [View same] [iqdb] [saucenao] [google] [report]

>higher is better

>> No.69265160

The amdrone meme is real.

>> No.69265389
File: 69 KB, 1327x1222, 1543241975159.png [View same] [iqdb] [saucenao] [google] [report]

U seem MA

>> No.69265523

AVX512 is way too fucking complicated. The extra 16-byte and 32-byte instructions are nice, but the masks for 64 byte vectors (512-bits) is insanity.

>> No.69265997

Funny enough, apparently in the Zen 2 AMD greatly improved AVX calculations by merely brute forcing the fuck out of AVX 256 two times over, essentially rendering 512 utterly useless and obsolete since everybody uses 256 and almost no one (besides just bribed paid Inturd shillers) uses 512.

>> No.69266218
File: 22 KB, 400x400, 1507574846759.jpg [View same] [iqdb] [saucenao] [google] [report]


>> No.69266328
File: 252 KB, 1024x576, 1546961725525.png [View same] [iqdb] [saucenao] [google] [report]


>> No.69266388

>liquid cooling
>103C stock
Jesus fucking christ.
Jews making their own ovens, what a time to be alive.

>> No.69266440

What a flop

>> No.69266537
File: 2.09 MB, 1323x207, 4w6457577.png [View same] [iqdb] [saucenao] [google] [report]

The most funniest thing in that pic is in the stock temps comparison of 2700X and 8700K, all while 2700X being an 8 core CPU and 8700K is a 6 core CPU. And 2700X costs quite a bit less than 8700K. And then there's also this factor in the play:

>> No.69266843

>2009 Sandy Bridge

>> No.69266893

Eta for AMD's keynote?

>> No.69266952

>92% yield

u w0t m8

>> No.69266953
File: 567 KB, 1479x651, CREATURA.jpg [View same] [iqdb] [saucenao] [google] [report]

See >>69263772.

>> No.69267006

Ah, thanks

>> No.69267026

Why is it that wccftech is "curry tech non-news fake bullshit" when it talks positively about Intel or Nvidia but it's the paragon of facts when it says something good about AMD?

Mental gymnastics and wishful thinking or double standards?

>> No.69267029

I like how amdrones can only speak in Red Team talking points. Nothing suspicious at all

>> No.69267061
File: 1.51 MB, 774x255, DELID CIA.png [View same] [iqdb] [saucenao] [google] [report]


>> No.69267065


Zen = origination design <<< 14nm
Zen+ = tock <<< 12nm
Zen2 = tick + new node <<< 7nm
Zen3 = ticktock + mature node & EUVL <<< 7nm EUVL
Zen4 = ticktock + new node & EUVL <<< 5nm EUVL
Zen5 = ticktock + mature node & EUVL <<< 3nm EUVL

AMD confirmed that thanks in part to the massive success of Zen1 in Enterprise and consumer desktop space, they had the funds to create multiple design teams whose sole job is to leap frog the predecessor architectures. This means that Zen1 & Zen2 are created by the primary team. Zen2 is basically all the lowest hanging fruit and design optimization that couldn't make it into Zen1 + a bunch of new things that were discovered through application and game design that improves core performance. Larger L3, bigger op cache, better front-end, 256-bit AVX pipelines *4, etc.

Zen1 and Zen2 had Team Blue & Green collaborating to success. Team Blue has moved onto Zen3 and Team Green has moved onto Zen4. Q4 2019 to Q2 2020 we'll start hearing about Zen3 and by Q4 2020 we should get a product launch. Then by Q4 2021 to Q1 2022, we should get Zen4 and by 2023-2024, we should close out with Zen5.

All these timelines are subject to change based on market conditions and development of each node:

7nm EUVL, 5nm EUVL, 3nm EUVL

>> No.69267091


Because if the whole site is run by pajeets they obviously have inside info on AMD.

>> No.69267109

>Paki mudslime

>> No.69267155

>but it's the paragon of facts when it says something good about AMD?
Because WCCFtech is clickbaiting garbage that will get something right once a day. All information in their 'article' has been circulating for months and nothing noteworthy has been added. WCCFtech has a habit of circulating every rumour around, praise themselves for getting 'it' right and linking it all back to each other for more clicks.

>"they are incredibly confident they can beat everyone (including Intel and NV) to market with 7nm products"
Every spokesperson at AMD has been talking about their time to market with a superiour 7nm product. Before the chiplets thing was confimed the only thing they talked about is how confident they were from the early samples they got from the fabs.

>"AMD will be soft launching (read: announcing) at least two 7nm product lineups."
wow. Its not like we didn't know every high end chip designer was waiting for the new node to drop. (read: intel isn't high end)

>> No.69267182

Surely the better comparison would be the 1700 vs the 2700x. Either that or the entire R7 1xxx and 2xxx lineup stacked against each other.

Nobody bought the 1700x when the 1700 existed, or to a much smaller extent, the 1800x existed.

What site did you get that from, I'm interested at looking at stuff from sandy bridge/snoozedozer era and seeing the timeline from there.

>> No.69267194
File: 80 KB, 503x503, common sense.jpg [View same] [iqdb] [saucenao] [google] [report]


>> No.69267225

We only have confirmation of 7nm so far correct? The current time line stops at Zen 3. Hitting 5nm and 3nm on silicon is going to be ridiculous honestly, We're already pushing the boundaries as it is. Even if new materials are found physics will still fuck things up.

>> No.69267246
File: 1.02 MB, 1098x1170, dumb.png [View same] [iqdb] [saucenao] [google] [report]


>> No.69267424


>> No.69267492
File: 390 KB, 1200x1553, 1543448512572.jpg [View same] [iqdb] [saucenao] [google] [report]


>> No.69267530
File: 2.10 MB, 1329x207, 6r7ty76787898.png [View same] [iqdb] [saucenao] [google] [report]

>Nobody bought the 1700X when the 1700 exis--

>> No.69267541

The reason AVX512 exists was because they were expecting to get 10nm working which would have reduced power so drastically that it wouldn't bottleneck the rest of the processor as much. 512 is usefull but in current intel processors only if your workload actually uses enough of them. else its better to generate without the 512 instr.

>The current time line stops at Zen 3
I doubt it. Lisa Su is probaly already looking over the horizon what they need to do post zen3. Process node tech luckely isn't their worry. TSMC will dish out the cards on what design changes are going to facilitate their fight with the elements. First thing they'll do is split the design for consumers (if they even care about those anymore) and datacenter/server guys. Core scaling will limit amd to 200-300 cores before you can't supply the frequency and scaling required to make a substantial improvement. making much larger and specific cores together with breaking compatibility will be the first thing to do. more and more software has become open source and anything 'old' will get emulated with more ease than ever. see AV1, Nextcloud, Corporate linux. Everyone will run gentoo and finally recognise the true power of my riced setup.

>> No.69267561

What is the difference between Vega Navi? Pretty ignorant to AMD gpus.

>> No.69267633

>What is the difference between Vega Navi?
Vega was never intended to be consumer GPU initially, as it was made with heavy buttcoin mining in mind. After the fad crashed and bubble burst, AMD got coprnered and now they're making stricyl consumer Vega with heavy GAYMEN in mind, which is Vega II: Eclectic Bungalow. Navi was always intended to be much farther and purely a consumer GPU, now Vega II is that, so Navi is going to be even more powerful.

>> No.69267653

>What is the difference between Vega and Navi?
Vega was never intended to be consumer GPU initially, as it was made with heavy buttcoin mining in mind. After the fad crashed and bubble burst, AMD got cornered and now they're making strictly consumer Vega with heavy GAYMEN in mind, which is "Vega II: Eclectic Bungalow". Navi was always intended to be much farther and purely a consumer GPU, an alternative to Vega, but now Vega II is that, so Navi is going to be even more powerful. Market situation forced them to change their initial plans.

>> No.69267658

So basically go for Navi over Vega if I care to upgrade my gpu the next year.

>> No.69267691

Intel is kvetching

>> No.69267695

No, go Vega II. Navi is waaay far into the future.

>> No.69267819
File: 458 KB, 758x711, a2eb82693819cff14a4f3d66ffdb523e57823b8706e32603d98de2c02b9471dd.png [View same] [iqdb] [saucenao] [google] [report]

>> No.69268035

Not going to lie I thought the difference would be bigger. Twice the numbers during the first 6 months is still quite the large gap to be fair. People were buying the 1700 up until the supply starts to dry out and the 1700x had a price drop. So yes, nobody bought the 1700x when the 1700 existed until the 1700 started to stop existing in shops.

>> No.69268073

A retard-a-you.

>> No.69268195
File: 153 KB, 1080x1350, 1546967386851.jpg [View same] [iqdb] [saucenao] [google] [report]

18 hours to go boys

>> No.69268225

oh fuck fuckfuckufjkcufkcuckcfuijk

it's Viceroy research all over again.

AMD stock getting shorted big time inc

>> No.69268250

This whole b-die thing is a meme if I'm not a benchmark weenie, right? There's a $70 price difference between 3200mhz cas 14 and 16.
I read that the less-than-full rated speed problems are being resolved so I should just get cl16 right?

trying to plan a preliminary build around this 3700x so i can keep an eye out for deals

>> No.69268461

hey my fx 6350 was running fine at 115°C

>> No.69268507

Any leak that starts confident then gets vague and wordy is fake this is fake

>> No.69268517


>> No.69268544

They switch the naming scheme after zen 2. What would be "zen 2+" is "zen 3"

>> No.69268624
File: 46 KB, 803x552, hmm.jpg [View same] [iqdb] [saucenao] [google] [report]

I'd pay money for that

>> No.69268687

AdoredTV's leaks are confident and he is confident.

Blame the bitch tech press for creating this mess.

>> No.69268721

>no twitch stream

its almost as if amd doesnt want the real gamer to know about their products

>> No.69268790

3200mhz cl16 is fine

>> No.69270333

As much as I want to grab a Zen 2 CPU, I just bought my 2700x and I'm hoping it will last me a few years. I want to get the mature 7nm CPU when it comes out. That'd be the Ryzen 4000 (unless they skip and go to 5000) or something.

>> No.69270575

What game do you think they'll fake benchmarks for?

>> No.69270592

How is this exclusive? Everyone is saying this.

>> No.69270593

AMD is well-known for sandbagging, so that may be true.

>> No.69270594


>> No.69270642
File: 1.10 MB, 1293x2048, 1546240027184.jpg [View same] [iqdb] [saucenao] [google] [report]


>> No.69270754


Samsung is ramping up its 5nm risk production this year. It's a matter of when, not if. Granted the complexities of desktop & enterprise CPU/GPUs are radically significant to the complexities of RISC ARM designs, still. If Samsung has it working, then TSMC for the more complex systems only another year or two behind at most.

>hitting 5nm and 3nm is going to be ridiculous

Nah. Managing quantum effects even at that small is still within capabilities. Once we get down to 1-2nm, then it becomes much more challenging. But the thing is that by the time we get to say the tock of the 3nm node, chiplets will have been a thing in market for some time. It's quite possible that AMD might spend up to 2 full design generations with chiplets on 7nm. So Zen2 and Zen3 to really iron out the design and configuration challenges to maximize performance.

Then we move to 5nm and they spend another 2 design generations on that, with Zen4 and Zen5. Then we get to 3nm and either we get a Zen6 and Next-gen CPU architecture or we get a next-gen architecture v1 and 2, all chiplet based. Moving forward, chiplets will drive the performance curve to get around the challenges of Moore's Law.

Consider that a 8c/16t chiplet of the Zen2 EPYC die is ~72mm^2. That is roughly half the size of a Zen1 die. At 3nm (napkin math), an 8c/16t chiplet will be ~36mm^2. This is with pure linear scaling of the die shrink and doesn't account for EUVL, which brings a 15% reduction in area to the printed design (aka transistors move closer, performance increases, and yields go up as a result with more dies printed per wafer). Which means that in the space of 2 Zen2 chiplets @ 7nm, you can fit 4 Zen5 chiplets @ 3nm.

Also bear in mind that the AM4 CPU package is pretty large, so you could at 3nm theoretically be able to do 4 chiplets & an i/o die in its space for a total of 32c/64t in consumer space; then with TR/EPYC scale that up to 128c/256t easily.


>> No.69270786

They've literally never sand bagged. With Zen the 40% IPC was their declared target when they hired Keller and were about to start the project. They ended up doing better.

You're not sandbagging by saying you're going to aim for an 80% on an exam and then get a 90%. Sandbagging would be actually performing lower on tests and then blowing everyone away on the finals, or lying and saying you did badly and then again acing the final. Doing that would have gotten AMD sued by shareholders because it's misrepresenting their product, and that aside the unpredictability and untrustworthiness of sandbagging would turn away investors.

>> No.69270865







GAAs will lead to some major improvements in performance of the silicon. Die-stacking, when it inevitably gets out of R&D for high performance silicon will lead to another huge boom in performance, and CNTs will at some point in the next 20 years finally make their way out of R&D and into large scale manufacturing projects including CPU/GPU space. There are two big challenges to CNTs that if solved would basically turn the CPU/GPU market on its head.

1. Mass production
2. Superconductivity at room temperature.

If these two are properly solved in the next 20 years, we'd be able to make a CPU with hundreds of cores and thousands of threads clocked to 6-7-8GHz and a GPU with tens of thousands of shaders with terabytes of RAM associated clocked to 3-4GHz with the same level of power and heat as a GTX 2080Ti today. It would blow the door of computational performance right off its hinges, through the fucking wall, and out into the stratosphere above. It would be as revolutionary to computing as how revolutionary computing was from using an abascus to the first x86 CPU.

At the minimum, we're looking at another 10-12 years of steady increase in perf; and that's enough time for a potential breakthrough in manufacturing design.

>> No.69270902

O rite guis howre doin? :3

>> No.69270935

>He's only going to have 8c 16t

>> No.69271118

Gimme gf

>> No.69271160
File: 52 KB, 400x400, wccftech.jpg [View same] [iqdb] [saucenao] [google] [report]


>> No.69271162

8c/16t is the new king of corelets

>> No.69271198

So many keks to be had if the new leaks are accurate

>> No.69271263


>> No.69272627

In case of AMD and ESPECIALLY APUs, it's not a meme.

>> No.69272641
File: 2.23 MB, 320x384, 1539299904198.gif [View same] [iqdb] [saucenao] [google] [report]


>> No.69272646


I started on a 66mhz 386 enhanced Pentium and have been back and forth on platforms as each offered cost benefit advantages at the time. I have been on a 2500k since launch and it should tell you how long it's been since Intel did something big. The 3700x is going to be my new 2500k and will keep me happy with 1080 or 1440 should I choose it later. By the time I actually need a new processor were going to be hitting 64+ cores on consumer desktops and acceptable 4k performance should be there too. Sure I could blow 3k on a godly rig but I can never justify it.

>> No.69272683

You are literally me, except that I've started with 266MHz Pentium II Klamath and that I'm currently on 2600K. I'll be completely skipping 4K and going straight to 8K, though.

>> No.69272936

>386 clones

>> No.69272958

The rumor is they revamped the front end so let's hope the compute isn't starved like with OG Vega.

>> No.69273218

The upside is that Stock they will be midway between the 2080 and 2080ti

Downside is they are gonna be 800-900$ due to the insane costs of HBM2+ and a monolithic die on the brand new 7nm node

The second upside is that with a waterblock they will OC well into the 2000mhz range and when hard OC'ed with proper voltage tuning and water loop that can pull the heat off them they will beat a 2080ti

They will also be better in SLI than the 2080ti's because due to AMD drivers and HBM memory we have seen vega and fury cards scale insanely well in dual GPU configs with extremely smooth frame timings and high minimums. You dont get the same frame drops and stutters on them that you get from GDDR memory gpu's

The issue with dual Vega-II's in SLI would be the limits of PCIe 3.0 but AMD is putting PCIe 4.0 on their latest chipset so bandwidth wont be an issue anymore.

If you are build an ultimate high end gaming rig this year Two custom water block'ed vega-II's in SLI will stomp the shit out of dual 2080ti's

The only real application for this setup would be either a 4K 144hz panel or driving one of the new OLED 8K TV's

>> No.69273471


I do want to add that practically everyone forgot about the original Vega whitepaper with the whole NGG pipeline promise that would alleviate the geometry bottleneck by having it handle 11 polygons per clock instead of the 4 with the current limitations. They never enabled it even though there are references inside AMDVLK, AMD's Vulkan driver. There's also the question of primitive shaders as well and whether that will be done also. Both of these features would need to be enabled transparently for anyone to use it, I would argue for AMD to have any chance at using it.


If all they will do is announce 7nm Vega+ for Vega II like what they did with Zen+, it is going to be massively disappointing to everyone. I really doubt that whatever tweaks and the die shrink would do is even close enough to let Vega reach the 2080 or above performance. I think they really need all their whitepaper features like primitive shaders and NGG in working order to even have a chance against the current Turing cards.

>> No.69273596

In terms of raw specs we know based on the instinct cards that with weak power delivery and shitty reference cooler the vega 7nm chip can hit 1800mhz.

We also know from vega cards that performance increases dramatically when HBM gets clocked up which means they were bandwidth starved.

The New HBM2+ standard has 1.6gbits of throughput meanins a 1600mhz memory clock over the paltry 900mhz that vega had on its HBM

Combine that with a raw core speed of 1800 mhz boosting to 1900mhz and you have beaten a 2080 soundly.

Also we know that the voltage curve on this 7nm node is much flatter than the 14nm node that means the limiter for vega's core clock isnt voltage, its heat.

So if a heat sensitive vega GPU can hit 1800mhz core with a crappy reference fan cooler and voltage is not the limiting factor then when you slap a custom water block on it, fix the voltage stepping away from default and tuned correctly to your specific silicon you should see core pushing into the 2000-2100mhz range and possibly even higher if you can control the heat.

If the core is fully fed by the HBM at 1600mhz and you can push the core speed to 2100mhz that will match if not slightly beat a 2080ti in benchmarks.

>> No.69273714


HBM2+ was just annouced no one has modules ready to sell. Hynix and Samsung only have 1200 Mhz modules at best right now. And your core clock is MI60's max clock. I guess if you overclock you can get the core clock boost but HBM won't get high enough to 1600 MHz most likely 1400 to 1500 MHz max. With all that you can get to above 2080 with that but not the 2080 Ti. I contend it will be a repeat of Vega I and Nvidia top cards again if AMD have not done something to address the most serious issues with the original Vega.

>> No.69273729

It wasn't made for memecoins, GCN just lends itself better to computing tasks than nvidia cards do on the same price range.

>> No.69273769
File: 963 KB, 2560x1440, 1544253505864.jpg [View same] [iqdb] [saucenao] [google] [report]

>> No.69274034

weebass amd shill

>> No.69274070
File: 90 KB, 549x413, 1545772870574.png [View same] [iqdb] [saucenao] [google] [report]


>> No.69274304

Better to be a weeb, than to be a brainless milking lolcow of a marketing victim cuckold.

>> No.69274454

can't wait for amd to announce $150 midrange navi with gddr6 alongside vega 2 with hbm2 as to end card.

>> No.69274468

jelly nvidiot has no chinese cartoons

>> No.69274766

>Vega II is going to be a soft launch
They will probably announce that you can buy VEGA20 a week from now if you're willing to pay a LOT of money. It won't be cheap.

>How powerful
Very powerful if you're the target audience and you're probably not. VEGA20 is clearly some kind of workstation/datacenter card. The Linux kernel driver for VEGA20 is done already (which is why I think it's close to release) and there's a lot of things that make you go huh such as dual video decoding and encoding. There's quite a lot of things like that which makes it very clear that this ain't no consumer card.

>AVX512 is really good
It's a joke and it's not a standard. You need to know the specific CPU you're targeting to take advantage of it because it's a pile of extensions to AVX256 and which of those you get varies wildly between Intel CPUs.

>> No.69274885

You don't need "b-die" RAM and all it means is "best RAM produced by Samsung". The term doesn't even mean anything since it varies from month to month. And you really don't need it. A kit that's 3000 MHz with CAS latency of 15 is great and those are cheap. Your motherboard probably can't run higher than that anyway.

Also, beware that Ryzen's can't into four sticks of high-speed memory. No idea why but a 2400G for example will do 3000 MHz just fine with two sticks but can't get higher than 2666 with four. Doesn't matter if it's b-die or not.

>> No.69274944

There will be no 12 core or 16 core version of Ryzen for AM4. 8 core is the highest and the price structure is going to remain the same. Do not expect anywhere near 5.0GHz

There will be no Navi announcements; those GPUs have not even made the rounds with AIB partners at this stage

Stop believing in random self-centered Youtubers

t - someone who works for a system integrator

>> No.69274971

>weebass amd shill
I'm guessing you are entire correct due to the limited possible answers to the following simple question: How did she get those promotional AMD items which are not sold to the public?

>> No.69275102
File: 180 KB, 1037x1033, 1539280510711.png [View same] [iqdb] [saucenao] [google] [report]

>There will be no 12 core or 16 core version of Ryzen for AM4. 8 core is the highest and the price structure is going to remain the same. Do not expect anywhere near 5.0GHz

Only 5 more hours bud and we'll see who's right:
Person with track record of providing accurate CPU leaks and who has had his claims corroborated by several others
System integrator employee

>> No.69275151

>who has had his claims corroborated by several others
No he hasn't
That ES "leaked" benchmark is not a genuine CPU. And Navi has not even been sampled yet by any AIB vendor. Almost everything Adored has claimed is not true or very misleading.

See you in 5 hours

>> No.69275359
File: 375 KB, 360x360, AMD at CES.webm [View same] [iqdb] [saucenao] [google] [report]


>> No.69275520

>what is undervolting

>> No.69275535

>It's a joke
It's not a joke, as it actually works and brings massive benefits in computing, but it's proprietary as fuck and black-boxed, so it'll never gets tract in mainstream sphere.

>> No.69275542


>> No.69275555
File: 1.75 MB, 732x312, INTURDED.png [View same] [iqdb] [saucenao] [google] [report]


>> No.69275558

based and redpilled
cringe and bluepilled

>> No.69275631

>>69275555 (checked)
>Imagine being intel

>> No.69275807

Intel haven't polished the 8c16t yet
AMD is striking them already with 12c24t/16c32t already
FEELS LIKE WWII all over again.

>> No.69275970

>Inturd haven't been polished yet
Cue in the "Zionist Golden Inturd" meme.

>> No.69276095

>looks worse than the gif

>> No.69276137
File: 81 KB, 444x444, 1502143993622.jpg [View same] [iqdb] [saucenao] [google] [report]


>> No.69276180
File: 1.80 MB, 514x916, 74r756rh675.png [View same] [iqdb] [saucenao] [google] [report]

It's JIF, motherfucker.

>> No.69276276

Hello Jim.

>> No.69276376
File: 499 KB, 1228x362, tick-tock-pao.png [View same] [iqdb] [saucenao] [google] [report]

Intel actually killed (pic) in response to Ryzen
and it will never go back for a long time especially when Rome arrives as AdoredTV predicted on what it will become

>> No.69276422
File: 107 KB, 1280x720, 1546914119149.jpg [View same] [iqdb] [saucenao] [google] [report]


Well, there we have it.

Steve from Hardware Unboxed just relased a 12 minute video where he murders Nvidia and then takes a dump on their tombstone.

There's no fucking way he would burn the bridge with Nvidia unless he has insider information about AMD Navi (and knows how good it will be).

>> No.69276427

He hired Tim. Maybe he's just retarded.

>> No.69276485

He's the literal DSP of hardware channels.

>> No.69276665


>> No.69276691

Reddit banned his videos? really?

>> No.69276786

Some [nVidia shill] subreddits have banned links to his videos, yes.

>> No.69276800
File: 187 KB, 1030x1030, nz8KH3ksEoNbzudX.jpg [View same] [iqdb] [saucenao] [google] [report]


>> No.69276806

You guys should stay over there

>> No.69276823

Go back and fellate jewvidia there.

>> No.69276835

I don't support Nvidia, nor go to reddit.

>> No.69276857

I don't think we just saw the same video there friendo. He just got cucked by Nvidia and is salty. That's it.

>> No.69276881
File: 10 KB, 192x246, 1542380403591.jpg [View same] [iqdb] [saucenao] [google] [report]


>> No.69276887

I don't care about reddit mate, but if """someone""" banned his videos there it may mean that """someone""" is in the payroll

>> No.69277143


>> No.69277370

>NGG pipeline
Under Raja's 'leadership' the software team never even *started* the work of trying to implement those features for gaming in a way that would be transparent to game devs. Hell, the gaming driver stack for Vega didn't get started at all until insanely late in the game, such that Vega launched on pre-alpha drivers based off Fury drivers with half the feature stack missing.

>If all they will do is announce 7nm Vega+ for Vega II like what they did with Zen+, it is going to be massively disappointing to everyone. I really doubt that whatever tweaks and the die shrink would do is even close enough to let Vega reach the 2080 or above performance.

How much faster is a 2080 than Vega 64? You should assume Vega II will be roughly the same percentage faster than Vega 64 that it clocks higher, so that's roughly 20-30%. I have no idea off hand where a Vega 64 + 25% lands relative to a 2080.

>We also know from vega cards that performance increases dramatically when HBM gets clocked up which means they were bandwidth starved.
Eh, the bandwidth is a secondary issue to the front end geometry bottleneck of 4 triangles per clock. Unless AMD revises GCN's front end, you will only get performance increases proportional to clock increases.

>> No.69278614

sheeit, i also still have a 2500k and am switching to a 3700x.
the plan is to buy a nice 4k monitor and downscale games to 1080 or 1440p until we can get 4k/60fps capable gpus for $500

>> No.69278970
File: 1.92 MB, 3200x1800, Unzip 'em.png [View same] [iqdb] [saucenao] [google] [report]


>> No.69279493


>> No.69279664


>> No.69281082


Name (leave empty)
Comment (leave empty)
Password [?]Password used for file deletion.