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: 777 KB, 948x747, 1502197904509.png [View same] [iqdb] [saucenao] [google] [report]
62049213 No.62049213 [Reply] [Original] [archived.moe] [rbt]

Where were you when AMD EPYC took over China?

Tencent, JD.com and Baidu.
This comes a few days after Intel's CEO left the American Manufacturing Council.

>> No.62049217

>>62049213
Oops, forgot link.

http://ir.amd.com/mobile.view?c=74093&v=203&d=1&id=2295144

>> No.62049246

DELET. REEEEEE

>> No.62049250

>>62049213
I doubt JUSTnich leaving Trump's council has anything to do with that.

Regardless AMD taking over China could be pretty cool.

>> No.62049255

>>62049213
DELID THIS

>> No.62049263

>>62049217
>AMD EPYC™ Momentum Grows with Datacenter Commitments from Tencent and JD.com, New Product Details from Sugon and Lenovo


AMD will be rolling cash next year.

>> No.62049282

>>62049217
>In August 2017, Tencent's market value rose to USD$409.50 billion. It also became the second Asian company after Alibaba Group to surpass USD$400 billion market cap.[22]

>JD.com is the world's leading company in high tech and AI delivery through drones and robots, and possesses the largest drone delivery system, infrastructure and capability in the world. It has also recently started testing robot delivery services and building drone delivery airports.[8][9][10][11][12][13][14]
>260B CNY revenue


NICE

>> No.62049298

Good news? incoming tanking of stock

>> No.62049398

I wonder if this ban is still in effect.
>The US government has blocked Intel from shipping high-end Xeon processors to China's supercomputer builders – and other American chip giants are banned, too.
https://www.theregister.co.uk/2015/04/10/us_intel_china_ban/

>> No.62049405

>>62049398
AMD doesn't give a fuck, probably why the US Gov. isn't paying attention to it.

>> No.62049410

>>62049217
FUCK
FUUUUUUUUUUCK
GHOST OF OTELLINI HELP ME

>> No.62049422

Pajeets GTFO, now based chinks are our best friends!

>> No.62049462

>>62049398
>Fabs are in deutchland
>Chips go to china
>Only a few top earning jew executives are in america

US aint got shit on AMD

>> No.62049532

https://www.digitaltrends.com/computing/amd-studios-creates-custom-production-pipelines/
Meanwhile in murrica.

>> No.62049561

Baidu was no surprise, they were a foundation partner for EPYC, but JD is out of nowhere. As is Tencent.

>> No.62049581

>>62049561
Expect Amazon to follow soon.

>> No.62049593

HEIL EBYN! :DDDD

>> No.62049614
File: 1.49 MB, 2048x1536, spurdoEBYN.jpg [View same] [iqdb] [saucenao] [google] [report]
62049614

>>62049593

>> No.62049621

>>62049217
Oy vey. We need to shut this down immediately.

>> No.62049622

>>62049581
Amazon is the one that's up in contention, they have more reason than anyone to seek density on their servers, and EPYC offers that, the question is, just how dependant upon Intel are their system, for a company like that a change to a very different arch like EPYC could be problematic.
1&1 is another foundation partner, they're one of Europe's largest broadband companies and data hosts.
Dropbox is already on board, another massive customer.
Microsoft are going EPYC for project Olympus.

EPYC isn't even on the shelves yet and it already controls 90% of the industry.

>> No.62049654

>>62049622
>change to a very different arch like EPYC could be problematic.
explain please. it's just another x86, is it not?

>> No.62049674

>>62049622
>totally different
It's still x86-64.
>>62049654
Yes, just with more NUMAness.

>> No.62049726

>“China Internet and e-commerce companies need more compute cores and higher memory bandwidth. We saw AMD EPYC processors have up to 32 cores, providing competitive advantage over current 2P server systems, and the eight memory channels enable greater memory bandwidth, which are believed to better match domestic customers’ requirements,” said Andrew Wang, technology leader of hardware system department at JD.com. “AMD EPYC will help JD.com improve the total cost of ownership (TCO) of our server systems. JD.com will collaborate with AMD on Big Data, AI and Cloud Services based on AMD EPYC in the future.”

>JD.com will collaborate with AMD on Big Data, AI and Cloud Services based on AMD EPYC in the future.”
>JD.com will collaborate with AMD on Big Data, AI

Shieeeet, so, 4-5 years for AMD to turn out a card that pisses over nVidia in AI? I mean, with a customer like that to receive input from, they could do it

>> No.62049754

>>62049726
Yes, AMD already patented their own edition of meme cores.
And yes, anything Vega-based can operate large datasets, something nvidia can't do.

>> No.62049779

This thread is the proof that amdfags are the biggest shitposters and subhuman trash on this board. Always playing the victim and whining about shitposting when thier company is attacked while shitting up every thread.

>> No.62049791
File: 43 KB, 1002x1480, justfuckmywife8-min.jpg [View same] [iqdb] [saucenao] [google] [report]
62049791

>>62049779

>> No.62049807

>>62049779
>AMD housefires etc etc
>shitting up the threads

suuure.

>> No.62049840
File: 31 KB, 395x558, delicia.jpg [View same] [iqdb] [saucenao] [google] [report]
62049840

>>62049779

>> No.62049856

>that swastika

Jim Keller really wanted to make the message known.

>> No.62049901

What the fuck.
Its only zenver1 and it's already busy murdering Intel in the datacenter.
What will happen once zenver2 lands?

>> No.62049915

>>62049901
Holocaust 2.0

>> No.62049937

>>62049915
But Holocaust 2.0 is already happening.
WHAT WILL HAPPEN NEXT?

>> No.62049951

>>62049901
Project Starship, 48C/96T 7nm with PCI-E 4.0

>> No.62049963

>>62049951
NO
IT CANT BE HAPPENING

>> No.62049964

>>62049398
AMD is a multinational. US can't do shit to them.

>> No.62049990

Things looking up for ayymd vega is even catching up with Pascal.

>> No.62050008

>>62049990
>catching up
NVidia needs to figure out their own HBCC first.

>> No.62050041
File: 2.59 MB, 480x271, numanuma.gif [View same] [iqdb] [saucenao] [google] [report]
62050041

>>62049674

>> No.62050057

>>62050008
Not really. HBM and stacked memory wasn't even needed.
Before HBM, AMD's cards had excess memory bandwidth anyway. They weren't limited in that regard when it came to GPU performance.

So obviously, what they do is increase memory bandwidth. And not things like ROP processors.

>> No.62050074

>>62050057
HBM's uses may use lie elsewhere, onboard memory for APUs for example.

>> No.62050095

>>62049405
AMD isn't shipping high end processors to china

They're shipping lots of weaker computing units and putting them together in china, thus, bypassing the US ban entirely

>> No.62050108

>>62050057
HBCC, not the memory itself.
HBCC is the logic that operates Vega's all new page-based memory subsystem.
As of now Vega10 is the only GPU in the world to be able to handle really large, CPU farms territory, datasets.

>> No.62050125

>>62049901
Possibly 7nm 5GHz 8-core drawing 60-80 watts.
6-core 4.somethingGHz <45w laptop CPUs.

Intel will be absolutely obliterated.

>> No.62050169

>>62050125
I want dat 7nm laptop.

>> No.62050171

>>62050125
12 core 95W and 24C 180W;

>> No.62050173

>>62050074
This is the end game.

Mcm Vega *cough* I mean Navi on 7nm 4 cores + 16 core apu ryzen gen 2 at 5ghz apu and 3ghz GPU and 32gb of hbm2+ in 2020 here we cum

Discreet motherboards CPUs and gpus will be a thing of the past.

AMD literally have the tech to do it now but they are gonna milk it based infinity fabric

>> No.62050196

>>62050173
Doing it now requires engineering of physical layer wide enough to support MCM GPU.
That and writing firmware/drivers is going to be ASS.

>> No.62050224

>>62050169
We all do.

>> No.62050248

>>62050173
>AMD literally have the tech to do it now but they are gonna milk it based infinity fabric
AMD is already testing the new custom apu for CERN. 16C/32T CPU+GPU 2000 core HBM.

>> No.62050265

>>62050248
What?
Where?

>> No.62050308

>>62050265
Shareholders meeting
16 ryzen x86 Core, 6-wide
288-bit CPU Memory Controller (4×72-bit, 64-bit + 8-bit ECC)
102.4 GB/s via DDR4-3200 (ECC Off)
85.3 GB/s via DDR4-2666 (ECC On)
102.4 GB/s between CPU and GPU via GMI
~2000-core Polaris GPU
2048-bit GPU Memory Controller
4 GB HBM SGRAM Memory (2 chips at 2GB)
512 GB/s GPU Bandwidth

>> No.62050346

>>62050308
Polaris and not Vega?
That's kinda silly.

>> No.62050359

Wait seronx is that you?
Gotta take the pills man.

>> No.62050371

>>62049398
But it doesn't apply to datacenters, right?

>> No.62050384

>>62050371
No.
Intel more or less lost chinks anyway.

>> No.62050391

>>62050346
Vegas was not ready last year.

>> No.62050405

>>62049213
this is something we knew since epyc launched

>> No.62050408

>>62049213
Oh I get it. the Infinity Fabric arrows are shaped lke a Swastika, because it is burning Intel to the ground

>> No.62050414

>>62050095
It's just a glued together cpu.
Nothing high tech

>> No.62050424

>>62050405
That was Baidu.
Now two more XBAWKS HUEG chink companies are adopting EPYC.

>> No.62050510

>>62050424
we already knew amd took over china since the ban in intel
why do you think mcm took over so quickly on amd? lol

>> No.62050525

>>62050510
yes, it clearly has to only do with a trade embargo on by no means has anything to do with the fact that EPYC demolishes Intel in every measurable aspect at every price point.

>> No.62050573

>>62050525
yeah that was the problem that epyc demolishes intel
not the fact that china as a whole threw lots of money on amd
http://www.anandtech.com/show/10268/china-calling-amd-forms-joint-venture-for-x86-server-socs-in-china
nooooo...

>> No.62050627 [DELETED] 
File: 112 KB, 691x771, 1497565895125.jpg [View same] [iqdb] [saucenao] [google] [report]
62050627

>>62049213
OY VEY

>> No.62050659
File: 33 KB, 650x525, 67413maxjops.png [View same] [iqdb] [saucenao] [google] [report]
62050659

>>62050414
Yeah, Intel said, so we can ship those 32 core EPYCs to China no problem they won't gain advantage over American Intel.

>> No.62050682

>>62049622
Like most companies Amazon server rollouts are in waves, replace a few hundred kilowatts with new servers and see how it goes.
Scale out latter if it's worth it

>> No.62050691

>>62049726
Google already took a piss on them.

>>62047937

>> No.62050705

>>62050408
Skylake-X is the Intel Solution Portable Oven.

For personal use.

>> No.62050883

I knew AMD was commies and terrorist, Intel supports American engineering

>> No.62050915
File: 105 KB, 1280x720, (You) .jpg [View same] [iqdb] [saucenao] [google] [report]
62050915

>>62050883
http://blogs.intel.com/policy/2017/08/14/intel-ceo-leaves-manufacturing-council/

>> No.62050928

>>62050883
Intel supports feminist frequency.

>> No.62050962

>>62050928
A proud American movement?

>> No.62051000

>>62049532
But I thought Vega was DOA and was worse than Fermi. Raja Poo (xD) was supposed to be fired. wtf, did /g/ lie to me?

>> No.62051017

>>62051000
>firing the lead architect of every good Radeon starting with R300
Are kids out of their fucking minds?

>> No.62051021

>>62051000
>surprised 5 years later that AMD sells trashbin silicon to gamers

>> No.62051255

AMD can't sell their crap to real companies so it has to cry to the Chinks

>> No.62051281

>>62051255
Even Intel considers these companies part of the the "Big 7" you fucking retarded shit.
What hardware company would ignore a $400 billion cap client? Are you fucking mental?

>> No.62051300

>>62051281
>taking the b8 so hard
Thank God it's September soon.

>> No.62051308

>>62049532
>AMD Studios, much to the producers’ surprise, offered to put AMD software engineers to work writing a custom player for the production.

oh fuck you AMD, stop playing with last century toys, give me better games!

>> No.62051317

>>62049398

>jews
>abiding by this

I guarantee intel has been selling just at a higher smuggler price.

>> No.62051328

>>62051308
They've already partnered with Bethesda and Ubishit to (hopefully) unJUST their engines.

>> No.62051361

>>62051328
>bethesda
>unJUSTing their engine
hahaha

>> No.62051373

>>62051328
>Bethesda
huh, do you think they will finally make the unfuck of gamebryo that todd uses since morrowind happen? that would be the day.

>> No.62051375

>>62051255
Nobody buys intel.

>> No.62051379
File: 109 KB, 2250x2324, 1502534850116.png [View same] [iqdb] [saucenao] [google] [report]
62051379

>>62049213
DELID

>> No.62051394

>>62051328
to be fair, ubisoft been pretty good recently with PC ports, WD2 ran like a dream on my 390, too bad their games don't get better

>> No.62051413

>>62050196
Does anyone really have any faith in AMD to do drivers for an MCM chip? Really?

>> No.62051423

>>62051413
More money = more code monkeys.
The more code monkeys you have the easier it's to make a good driver.

>> No.62051430

>>62051413
they maybe late but they deliver pretty good drivers for the last 3 years

>> No.62051452

>>62051300
And the board will be flooded with skylake x owner.

>> No.62051463

>>62051452
>implying somebody will be insane enough to buy the thing

>> No.62051468

>>62051017
>Firing the guy who hasn't done shit since he left Apple, and only cost AMD market share.

Yeah, nothing to be gained from getting rid of that dead weight.

>> No.62051473
File: 182 KB, 755x1024, lisa_su.jpg [View same] [iqdb] [saucenao] [google] [report]
62051473

>>62049405
> mfw mcm

>> No.62051483

>>62051328
>brand new gamebryo
>same old gamebryo but now with hair that doesn't look like dogshit

I guess that's fine too. It made waifu mods better

>> No.62051490

>>62051468
Jensen please.
You goons have to clone HBCC in one R&D cycle.
Time's up.

>> No.62051509

>>62051483
I think it's mainly about AMD moneyhatting Bethesda studios to implement Vulkan in more games

>> No.62051516

>>62051430
No. Crimson made them better. It wasn't until the tail end of the hawaii life span did AMD get it's shit together on the driver front.

>> No.62051525 [DELETED] 

AYYMD HOUSEFIRES 180W

>> No.62051527

>>62051509
Implementing Vulkan implies bretty heavy engine rearchitecturing.
That's where AMD and their tools come in, including their bretty awesome GPU profiler.

>> No.62051530

>>62051468
>polaris - gained 10% amrketshare back
>vega - created niche in itself SSG, project 47, RED and EPIC CEO on stage, buttbudy with bethesda and ubisoft

but whatever you say.

>> No.62051531

>>62051490
Raja couldn't even clone tiled rendering in three architectures.

But don't get hot under the collar.

>> No.62051537

>>62051525
go back to wccf

>> No.62051548

>>62051516
omega was pretty good.

>> No.62051549

>>62051527
So awesome, no one else uses it.

>> No.62051550

>>62051531
Raja made it ages ago, in Imageon.
Hilarious, ain't it, Huang-san?

>> No.62051554

>>62051549
you don't know what you're talking about

>> No.62051560
File: 16 KB, 253x199, 1496544183097.jpg [View same] [iqdb] [saucenao] [google] [report]
62051560

>>62050008
HBCC is nothing more than a DMA controller.
Nvidia already has one and its called pinned memory.

I have several AMD systems but they really can get out of control w/ the marketing :
https://en.wikipedia.org/wiki/CUDA_Pinned_memory

>> No.62051562

>>62049213
You know, it makes me wonder, there are some business guys who say "winning in business by letting someone else suffer doesnt usually work well for your own business in the long term, if possible a feasible compromise or a fair win is much more desirable"

and I am getting why this is so in the long run. Intel may have gotten so much bad flack from their business practices that it may have actually affected some of the buying decisions right now.

>> No.62051567

>>62051548
It was just the transition from Catalyst to Crimson. Nothing special.

>> No.62051572

>>62051509
>AMD moneyhatting Bethesda studios
haha, i'm pretty sure if they try they can buy whole AMD, not just RTG
you have no idea what kind of margins game development brings if it hits, no to mention zenimax at their back

>> No.62051581
File: 298 KB, 515x438, 1491255558070.png [View same] [iqdb] [saucenao] [google] [report]
62051581

>>62050108
https://en.wikipedia.org/wiki/CUDA_Pinned_memory

Fell for the marketing meme.

>> No.62051594

>>62051527
>Elder Scrolls 6
>Massive engine restructuring and implementation of Vulkan allows for a massive increase in performance and allows for hundreds of mods to be loaded without slowing the game down to a crawl.
Excite.

>> No.62051596

>>62051572
Oh, I see. Agreements were made out of the kindness of Bethesda's heart. My bad.

>> No.62051600

>>62049779
>>62050414
>>62050510
>>62050573
>>62051255
Weak damage control, guys.

>> No.62051612

>>62051549
anon, they simply ported gpu profiler that used on consoles to pc, so EVERYONE uses it. It's slightly cut down version, but still better than nothing.

>> No.62051613
File: 33 KB, 1060x722, 1501250367381.png [View same] [iqdb] [saucenao] [google] [report]
62051613

>>62051255

>> No.62051617

>>62051525
Reminder to report shitposters.

>> No.62051618

>>62051550
Must be why he can't implement it in Fiji, Polaris, and Vega... he's such an esteemed expert.

>>62051554
What a devastating retort. "I know you are but what am I" is up next in your arsenal of bullshit?

>> No.62051641

>>62051328
>AMD make a better product than intel.
>Bethesda learn something new.
This is a parallel universe isn't?

>> No.62051649
File: 7 KB, 200x195, 1496295403167.jpg [View same] [iqdb] [saucenao] [google] [report]
62051649

>>62051527
> mfw nvidya performs better in vulkan
> mfw nvidya already has a more open and widely supported GPU profiler that's better.

>> No.62051657

>>62051649
>nvidya performs better in vulkan
Excuse me what

>> No.62051667

>>62051618
you seem pretty devastated desu

>> No.62051670

>>62049398
> China sets up a shell company in Romania
> buys processors for "ISP local infrastructure""
> ships processors to China from there

Silly America, this is how you got Titanium out of the USSR for the SR-71 and stuff.

>> No.62051678

>>62051649
doom has 200 fps limit, so we don't know which is faster

>> No.62051688
File: 505 KB, 1920x1080, 580 1060 game breakdown.png [View same] [iqdb] [saucenao] [google] [report]
62051688

>>62051678
Given equivalent hardware, though ..

>> No.62051690

>>62051600
literally stating that amd already had china on its pocket with or without epyc being good is damage control

reverse shillng on its action

>> No.62051694

>>62051657
does it not?

>> No.62051712

>>62051688
but these are midrange cards

>> No.62051721

>>62051667
Kohai, you have much to learn.

>> No.62051725

>>62051618
polaris dont have primitive culling? AHAHAHAHHAHAHAHAHAHAHAHAHAHA
what an idiot stop posting anon you are embrarrasing your sorry ass

>> No.62051735

>>62051721
I am devastated.

>> No.62051736
File: 34 KB, 676x666, 1492334740830.png [View same] [iqdb] [saucenao] [google] [report]
62051736

>>62051712
here is broken vega

>> No.62051740

>>62051527
It doesnt. The renderer can be almost completely independent of any of the framework that actually runs the logic. What's important is the result

>> No.62051741

>>62051581
>>62051560
Pinned memory requires explicit implementation.
HBCC is transparent.
tl;dr hardware > software.

>> No.62051742

>>62051725
I think you need to sit down and read it again.

>> No.62051760

>>62051735
If I were spewing blatant lies that are easily refuted, I would be too.

>> No.62051774

>>62051760
What lies?

>> No.62051800

>>62051774
Now we feign ignorance. What will the AMD shill do next?

>> No.62051804
File: 262 KB, 550x502, AMD-EHP-32coreAPUgreenland.png [View same] [iqdb] [saucenao] [google] [report]
62051804

>>62050248
>>62050265
>>62050308

>> No.62051805
File: 34 KB, 345x309, 1493414238936.jpg [View same] [iqdb] [saucenao] [google] [report]
62051805

>>62051649
>>62051678
http://www.phoronix.com/scan.php?page=article&item=intel-radv-nvidia-13&num=2

I hope you understand that Nvidia is also backing and contributes to the Khronos group :
https://www.khronos.org/

Yes, Nvidia performs better. It's called better drivers + open consortium software stack.
Watch some videos from the Khronos group. One of the top contributors is Nvidia staff.

> mfw fanbois are unaware how shit actually gets done in industry.
> mfw you hardly see AMD reps involved in Khronos talks

>> No.62051824

>>62051463
i don't know, skylake x + overclocked air cooled vega 64 liquid edition sound like a comfy combo for winter break.

>> No.62051832

>>62051800
You seem to be confused. Did you not claim the gpu profiler wasn't used by anyone else? It's not something someone who knows what he's talking about would state.

>> No.62051839

>>62049213
Why didn't they just make one die?

>> No.62051847

>>62051805
>linux
>relevant
>games
find a word that doesn't fit.

>> No.62051849

>>62051560
AMD has it to for professional cards. Hbcc is unique because it takes advantage of hbm's reduced latency for

>> No.62051853
File: 120 KB, 456x337, 1494431549238.jpg [View same] [iqdb] [saucenao] [google] [report]
62051853

>>62051741
Pinned memory requires drivers and software just like HBCC. Hardware doesn't run on magic pixie dust. It runs on drivers and software.

DMA stands for Direct Memory Access brainlet.
That's what the hardware is. That's what HBCC is. Nvidia has DMA HARDWARE that can access main memory directly which is exactly what HBCC is.

HBCC is not transparent. A user must define how much main memory they want (pinned) for DMA to the GPU the same way you do it with PINNED MEMORY on an Nvidia card.

When is summer going to end? Tired of fanbois/shills/brainlets commenting on shit confidently that they have no understanding of.

>> No.62051855

>>62051805
thats why they are keep trying to create abstraction layers on vulkan? its because they are perfoming better?
or because fp16 is about to come back on gaming as a relevant feature and they have fucked up pascal big time on it?

>> No.62051863

>>62051805
Steak through the heart of these idiots in denial.

http://www.redgamingtech.com/the-vulkan-interview-exclusive-interview-with-the-khronos-group/

>> No.62051884

>>62051832
Where does that burden of proof lie again? Who else is using it?

>IB4 consoles

>> No.62051893
File: 25 KB, 641x530, 1494912778105.jpg [View same] [iqdb] [saucenao] [google] [report]
62051893

>>62051847
> mfw brainlets think Vulkan is just about games
> mfw brainlets have no clue who even develops Vulkan
> mfw brainlets aren't capable of googling vulkan performance in windews and seeing Nvidia beating AMD there
> mfw a brainlet thinks AMD leads vulkan development or support
> mfw summer fags

>> No.62051895

>>62051884
>b-but this doesn't count because I say so
my work here is done
thanks senpai

>> No.62051900

>>62051853
a user doesnt need to define anything on hbcc hence why its agnostic

>> No.62051902

>>62051805
didn't AMD pretty much gave them mantle?
what else would you want them to contribute?

>> No.62051912

>>62051895
Oh boy... You sure showed me... Thanks again for proving that no one but Zenimax are using it.

>> No.62051948

>>62051902
Vulkan is straight up mantle, even the Nvidia guy on the council admitted it. Isbjust a buttblasted nvidiot who's in la la land

>> No.62051968

Well fuck you all and your meme GPUs, the thread is about EPYC.

>> No.62051994
File: 66 KB, 720x707, 1491601973605.jpg [View same] [iqdb] [saucenao] [google] [report]
62051994

>>62051849
Repeat after me : DMA (Direct Memory Access)

>>62051855
Vulkan is a very low level language. Yes, to spur development for it, you have to recognize your development community and put in effort to abstract low level functionality to a level that developers are comfortable with. And yes, it performs better. Take 15min from shitposting on this thread and google it.

FP16 isn't a new phenomenon. It is implemented and exists in several cards. While RADEON marketed FP16 (HARD) w/ vega, since they're busy writing custom ROMs for miners, they have yet to enable it.

Pascal beats vega. I'm not a fanboy. I have AMD/Intel/Nvidia hardware, run windows/linux/macosx. Call a spade a spade and stop being a simpleton shill.

https://www.youtube.com/watch?v=SYAWIdQS5Uk

Radeon makes compelling hardware (on paper). When it comes to drivers/software/developer tools (where the rubber meets the road) they fall off a fucking cliff which is why OpenGL functions better than vulkan on their own dam cards.

>> No.62052012
File: 198 KB, 725x355, hbcc.png [View same] [iqdb] [saucenao] [google] [report]
62052012

>>62051900
Where the fuck are you shills coming from?
http://www.guru3d.com/articles-pages/amd-radeon-rx-vega-64-8gb-review,31.html

Do you even understand how DMA works?

>> No.62052024

>>62051994
we both know what abstraction layer for nvidia means
having a low level console like api was something every dev wanted hense why mantle got loved and spawned both dx12 and vulkan
its not like they dont know how to develop for it its that they dont want to spend money on optimising shit

>> No.62052028

>>62051994
>DMA (Direct Memory Access)
why is everyone up AMD butt for having large datasets if nvidia can do it?

>> No.62052075

>>62051994
>let's put a high level abstraction layer on top of the low-level API
That kinda defies the whole point.
But we'll that's NVIDIA.
>>62052028
NVIDIA can do it buuuuuuut you need to modify your software to do exactly that.

>> No.62052126

>>62051375
I hope they just ditch their in-house dogshit engine and use lumberyard or so.

>> No.62052135

>>62050196
So, basically chiplets like on consoles?

>> No.62052139

Fuck off GPU brainlets.
This is CPU thread.

>> No.62052149

>>62050095
Quite smart of them, I wonder what percentage of sales are chinese sales now.

>> No.62052158

>>62051000
Well, to be fair, RTG also didn't have to throw RX Vega out the front door on half-finished alpha drivers, either.

>> No.62052219

>>62052028
Because Radeon group has clever marketing and ways of behaving like they're doing something new and exciting. Meanwhile, they're rewording/re-marketing/repackaging things that already exists...

Just about every feature you see Radeon boosting about exists on Nvidia cards already. The question is which one. Professional or consumer vidya. To that, someone asks : Well, does Radeon offer more to the consumer than nvidia? The reply is : No and not unless they have to.
Nvidia has had pinned memory and DMA for years. DMA is not transparent nor can it ever be because you're dealing with a shared memory resource. There are tons of drivers/software and it requires user specification (how much do you want allocated). This feature is even more pronounced on professional series cards just as it is w/ AMD. AMD pulls the same shit as Nvidia : They gimp the fuck out of consumer cards and put all of the interesting features in their pro-line cards. DMA (HBCC) is gimped on Vega cards. The real significant uses of HBCC (the ones that got demoed and the ones that drove 8K video production) are restricted to their workstation/pro cards... Just like nvidia.
RDMA (remote dma) is disabled on consumer cards by both Nvidia/AMD. All of those exciting Radeon demos were on professional/workstation cards... Yet brainlets think it will be accessible on Vega RX.

Young summer fags and their shilling. Both companies pull this shit w.r.t to pro/workstation vs consumer. I have no clue why people love shilling for Radeon when they pull the same crap as Nvidia. My hunch is that most people have no clue wtf they're talking about.

Do yourself a favor and ask AMD if they're going to expose FP16 to compute flows....
For gaming flows, you're going to be disappointed by how small a bump in performance it produces... FP16 might be x2 faster than FP32 on paper, but there is a shit town of hardware/firmware in the pipeline that matters w.r.t real-world performance.

>> No.62052241

>>62051968
>>62052139
give it up, i don't even know whether they are shilling or smart ass.

>> No.62052264

>>62052241
I don't really know why GPUs cause so much shitflinging.
These are specific tools at best and toys at worst.
Nothing more, nothing less.
Boring as shit.

>> No.62052269

>>62051560
A dma controller is implemented in hardware. Pinned memory, firepro GMA, and hbcc are memory management schemes that take advantage of dma. And like the other anon said it's transparent and acts in the same way as paging in an OS.

>> No.62052309

>>62051902
>>62051948
Yes they did give them the seed code for Vulkan and yet there obviously has been years and years of development beyond that point to mature it.

Maturation is what gets wide spread adoption... If mantle was mature enough for developers, it wouldn't take years and years of extended development.

Why I highlighted Nvidia's involved at Khronos was to define whose contributing to polishing/maturing it. The Radeon group has issues w/ polished drivers/software/dev tools. It's the major thing they struggle with.

>> No.62052328

>>62052012
HURR PAGING ON WINDOWS ISN'T TRANSPARENT BECAUSE YOU HAVE TO ALLOCATE IT

Are you retarded?

>> No.62052392

>>62052219
>Because Radeon group has clever marketing
hahahahahahahaha

>> No.62052402

>>62052309
>years and years of development beyond that point to mature it.
Except there wasn't. The Nvidia chairman called mantle the basis for vulkan. And yes, it was years in development by EA DICE until AMD partnered up with Johan anderrson. Johan had stated everyone but AMD rejected the idea of a low level API and it wasn't until after they brought it to market that others tool an interest
Hell, the xbox1 used a neutered version of dx11 until just before mantle was released to the public. You're literally making up pure fantasy history

>> No.62052406

>>62052392
>tfw too intelligent to buy nvidia

>> No.62052424

>>62052269
Anon obviously had no understanding of what hardware enabled HBCC which is why he went on shilling as though it was something ground breaking.

> Pinned memory, firepro GMA, and hbcc are memory management schemes
> that take advantage of dma.
> And like the other anon said it's transparent and acts in the same way as paging in an OS

Thus nullifying any extended praising of one implementation vs the other. All are gimped on consumer cards from Nvidia to Radeon group. Pro line cards : the shit they demoed have these features fully enabled in all of their glory. So, there is no reason to be giving extended lip service to HBCC while ignoring that it's existed for years as pinned memory on Nvidia (Consumer cards).. Btw, do yourself a favor and look up the respective latency of the various memory management schemes. Yet again, you see Radeon's software/firmware/drivers rear its ugly head and shit on performance. So, as far as I'm concerned, HBCC is yet to be detailed and proven on consumer cards. It's just another bullet item marketing uses to wow brainlets that is a gimped handmedown from pro cards

>> No.62052444
File: 452 KB, 1500x1001, 1488507741816.jpg [View same] [iqdb] [saucenao] [google] [report]
62052444

>>62052139
This is cosplay thread now.

>> No.62052457
File: 2.51 MB, 500x500, 1492298680486.gif [View same] [iqdb] [saucenao] [google] [report]
62052457

>>62052328
No, you are as evidenced by your idiotic remarks about Pinned memory being non-transparent. It's the same thing shit for brains. So, stop using stupid words that have no meaning. You allocated pinned memory and you allocate HBCC memory. It's the same dam thing dummy.

>> No.62052511

>>62052444
>FE Fates
Eh.
Gross.

>> No.62052543

HURR DDURR MALLOC IS LITERALLY THE SAME THING AS THE JVM'S MEMORY MANAGEMENT

>> No.62052569

>>62052264
Because they are specific fools at best and goys at worst.
Nothing more, nothing less.

>> No.62052626
File: 27 KB, 560x315, 1179.jpg [View same] [iqdb] [saucenao] [google] [report]
62052626

>>62052402
I'm literally not making up a single thing. Vulkan has been through how many years of development from how many industry partners? You comment about frameworks for customized hardware (embedded closed systems) as if you can copy/pasta it to general purpose computing... You can't which is why the Khronos group was given the task to make this a reality with tons of industry partners and contributors over multiple years. Name dropping of talking heads and all-stars doesn't change the fact of what happens in the trenches to get these platforms to a level of maturation to be an open standard. Xbox1 isn't open nor was mantle :
> The Mantle API is only available as part of AMD Catalyst, only for Microsoft Windows, but not for other operating systems such as Linux.

So, cut the bullshit. People who do real software development know exactly how consortium work as well as seed code. If mantle was open, polished, and ready for primetime, it would have went to market that way years ago.

Radeon can't even write functioning drivers for their flagship vidya cards and you want me to believe they could have made a broad industry standard language that work across all vidya/operating systems/platforms? Nigga please

>> No.62052801

>>62052626
AMD literally donated Mantle to Khronos. It resulted in glnext intiative and eventually Vulkan.
Please stop making shit up.

>> No.62052864
File: 184 KB, 800x980, 1491066625155.jpg [View same] [iqdb] [saucenao] [google] [report]
62052864

>>62052543
> mfw I finally get a brainlet to admit its not hardware
Now we can have get into drivers/software where Radeon group tends to be shit.

So, how does HBCC software perform :
http://www.guru3d.com/articles-pages/amd-radeon-rx-vega-64-8gb-review,31.html

> HURRR DDURR you can use JVM's Memory management for Real-time latency sensitive operations
> muh fuckn memory scaling for vidya games
> Muh pcie communication latency
> Muh Main system memory latency
> Muh pass through coming/going through DMA controller
> Muh round trip latency

> Muh custom memory management software (HBCC) that performs like shit because I have poo in the loo software developers who don't know how to fully exploit hardware.
> Muh jabs at malloc that performs better
> Muh marketing spin on a shit tier malloc wrapper that performs like ass

Get BTFO'd Object oriented transparency brainlet that forgets that those things cost in performance.

>> No.62052926
File: 42 KB, 600x600, 1495484669610.jpg [View same] [iqdb] [saucenao] [google] [report]
62052926

>>62052801
Mantle literally was not open and only worked on specific architectures and thus had no ability to become an open standard which is why AMD (hoping to steer) an open standard pushed it to Khronos hoping it would give them an advantage in steering the industry just like all consortium groups that have a dominant and founding member(s).

Please stop acting like a wet behind the ears star eyed brainlet who doesn't know how the industry functions or how down companies try to make their way back in via seeding consortiums w/ unpolished code they're familiar with that gives them a leg up over their competitors if it becomes a standard.

You can literally feel the pain the Khronos group has gone through to try to make Mantle into something other than an intern level optimization framework.

>> No.62053004

>>62052926
>Mantle literally was not open
AMD offered Nvidia to use it. Nvidia refused. then AMD tested it with Battlefield 4 and gave it to Khronos.

>> No.62054028

>>62052219
How buttblasted do you have to be to write such a wall of text? lmao

>> No.62054247

>>62053004
Wrong AMD shill. They never offered Nvidia to use mantle. It was strictly in house development. They said AFTER it was complete they would let Intel and Nvidia use it. But they gave it to Khronos instead who modified it for Vulkan.

>> No.62054409

>>62054028
High GWAM + clear train of thought
Takes less time typing that wall of text than shitposting

>> No.62054495
File: 87 KB, 409x409, 1495500738136.jpg [View same] [iqdb] [saucenao] [google] [report]
62054495

>>62053004
Even if this occurred, which it didn't they'd only do so in hopes that Nvidia's superior software team polished it just like Khronos has been doing for years. Google pulls the same shit w/ tensorflow and any company led (open framework). If you have half a brain, you'll see that they do so to gain advantage over and sway the market towards something they control.

This open bullshit is a meme. No one opens anything that is highly valuable. AMD locks down their consumer cards just like Nvidia. They do demos with pro-line cards, use marketing speak to get you hyped about the feature being on consumer cards and then they take a shit on the drivers/software stake if not outright disable it.

Vulkan is nowhere near primetime and that's after 4 years of cross industry contribution. This just goes to show you what the state of mantle was.

What Radeon can do right now is be clear on what features will be on the Radeon RX vega cards and when they'll actually write the fuckn drivers to be able to use it.

Once they manage to write proper software/drivers for their cards, then talk to me about Vulkan support. They're chasing too many things too far out w/ too little a budget.

>> No.62054499

>>62054247
>in house development
>literally Johan Andersson at DICE made it
>in house development

>> No.62054540

>>62054495
>This open bullshit is a meme. No one opens anything that is highly valuable
now this is some new level of retardness.

>> No.62054602

>>62054540
That's the nVidia™ way.

>> No.62054608
File: 2.30 MB, 1822x1224, 1499234569267.png [View same] [iqdb] [saucenao] [google] [report]
62054608

>>62054540
Consortiums and open platforms are a meme in order to get highly talented people to contribute to a platform for free or to lure people into an ecosystem that has tiered pricing.

The only retards are the brainlets who contribute to a billion dollar company's platform for free.

> Muh free platform
You are the product as is your data

>> No.62054648

>>62054608
Dumb frogposter.

>> No.62054660

>>62054602
exactly, AMD can not afford to do that with 20% market share. it has to be "the good boy".
>>62054608
>Consortiums and open platforms are a meme in order to get highly talented people to contribute to a platform for free or to lure people into an ecosystem
and this is bad? right.
>You are the product as is your data
like I care as long as they deliver something good and cross platform you dumb frogposter.

>> No.62054706

>>62054660
>it has to be "the good boy".
Almost everyone is a "good boy".
It's just nVidia.
Hell nVidia are such insufferable faggots that Sony of all people do not want to work with them after ebin PS3 expirience.

>> No.62054934

>>62052864
Are you really linking some muh gaymes benchmark as an example of HBCC being shit? Really?

>> No.62054995

>>62054706
>hurr durr nvidia is responisble for the shitty PS3
Sony wanted PS3 to use 2 cell chips and it was too late when they saw it's fucking retarded. They literally went at the last minute to nvidia to salvage the console. It's one of the reasons PS3 was delayed almost a year. If anyone fucked sony, it's IBM as they gave microsoft much better cpu despite IBM&Sony working togehter on the cell.

>> No.62055042

>>62054995
didn't sony refused ibm's chip somewhat similar xbox 360's chip design? then microsoft jumped on it?

>> No.62055057

>Thread about Server CPU
>B-but Radeon.
This board is beyond autistic.

>> No.62055085

>>62055042
No, Sony used CELL for CPU plus some off-the-shelf 7 series GPU from NVIDIA.
MS used PPC-based tri-core CPU (Xenon) combined with custom ATi GPU (Xenos, the first GPU with unified shaders).

>> No.62055088

>>62055042
No

>> No.62055103

>>62055057
Well you gotta shitpost about something.
You can't shitpost about EPYC, it's good, and market adoption rate is silly high given its completely new platform.

>> No.62055108

>>62055085
>>62055088
I know what sony used. I vaguely remember that sony refused the Xenon design, and went for Cell.

>> No.62055114

>>62055103
>Well you gotta shitpost about something.
Like I said, beyond retarded.

>> No.62055118

>>62055108
That definitely wasn't the case, hell, Sony was the driving force of CELL development.

>> No.62055129

>>62055114
That's /g/.
It's a shithole.
Basically /v/ with CPU/GPU/any other form of consumer tech.

>> No.62055158

>>62055118
I was remembering wrong, microsoft benefitted from the development of cell, without paying for R&D.
https://www.theguardian.com/technology/blog/2009/jan/01/sony-xbox-ibm

>> No.62055171
File: 769 KB, 948x747, epos.png [View same] [iqdb] [saucenao] [google] [report]
62055171

>>62049213
Holy shit they can't get away with this

>> No.62055196

>>62055171
They can.
People in datacenter really really hate Intel for price hiking, segmentation and some other itty bitty thingies.

>> No.62055225

That leaves only Google, Tencent and Amazon from the 7+1, right?

>> No.62055274
File: 417 KB, 653x635, 1497938426909.jpg [View same] [iqdb] [saucenao] [google] [report]
62055274

>>62054934
Are you implying they have this shit available to compute workflows on consumer grade cards? Are you that dumb? If so, do you really think this is anything different than pinned memory?

So yeah, I am pointing out (on consumer cards) for which developers have written in support that its shit. It's shit because the hardware pipeline, firmware, drivers, and software have been over promised w.r.t real world performance.

As the typical idiot posting here knows fuck all about comp. arch, PCI-E latency is significant as is memory access time. Now multiply this by two and account for the interrupt/latency of getting the data into the portion of the pipe it belongs in on the return side of the transaction.

Meanwhile, you're stalled out on the video card waiting for memory. Memory starved. There's a reason why you simply add more onboard memory to the GPU if you have memory issues.

For compute flows, the same principal applies. PCIE latency and DMA access are brutal compared to GPU local memory access. Meanwhile brainlets think HBCC is the second coming...

>> No.62055293
File: 145 KB, 1500x1054, ourgoys.jpg [View same] [iqdb] [saucenao] [google] [report]
62055293

>>62055196
I guess we know whose who now

>> No.62055295

Dear retarded frogposter,
Fuck off from this fucking thread.
It's about CPUs, not GPUs.

>> No.62055410

>>62051688
>RX 580
>base boost 1340
>"very good" overclock is 1500Mhz

>GTX 1060
>base boost 1708
>worst stable overclocks start around 2.1Ghz

I guess it's fair. I'm just happy Nvidia stepped up and put atleast a 192bit bus back on the x60 series. The 960 was such a piece of trash that was only possible because the competitor card was a literally 3 years old rebranded card.

>> No.62055555

>>62055410
>RX 580
>base boost 1340
>"very good" overclock is 1500Mhz

>GTX 1060
>base boost 1708
>worst stable overclocks start around 2.1Ghz

>muh amd cant overclock myth
go fuck yourself you dumb favela nigger, completely false

>> No.62055604
File: 108 KB, 480x480, 1496542148113.png [View same] [iqdb] [saucenao] [google] [report]
62055604

>>62055295
What about CPUs? This is /g, do you have enough of a technical understanding to have an intelligent conversation beyond posting headlines? AMD made a good CPU and is winning marketshare. This isn't reddit where you spend 300 posts jacking a company's dick. Sometimes the convo forks. Were discussing comp arch. and code. It's all on the same motherboard and interacts. Get w/ the conversation of fall off brainlet

>> No.62055656

>>62049213
Who cares? Servers will run on ARM chips soon enough. This is a lost cause.

>> No.62055688

>>62055555
QUINTUPLES CONFIRM
REVIEWERS ARE TO DUMB TO OVERCLOCK AMD

>> No.62055828

>>62055656
>arm
>real computing power
pick one

>> No.62055862

>>62049462
AMD doesn't own have fabrication facilities. And Global Foundries' 14nm Fab 8 is in Malta, New York.

>> No.62055870

>>62055555
I'm not saying it can't, it just can't as well as Nvidia. Face it, AMD silicon isn't as efficient at competetive clocks, the 580 is literally just a 480 binned for high clocks and it heats up and draws more power out of the box than the 1060 ever did, while delivering similiar performance.

>> No.62056025

>>62055870
>Face it, AMD silicon isn't as efficient at competitive clocks,

Sitting on a 290x that can get a near 30% boost, stop lying on the internet, especially to people that can fact check your bullshit

>> No.62056038

AYYMD HOUSEFIRES 180W

>> No.62056161
File: 299 KB, 2625x2619, 🤔.jpg [View same] [iqdb] [saucenao] [google] [report]
62056161

>>62056025
>500W to get 780 performance
AMD truly has the deepest lore

>> No.62056561

>>62051805
AMD invented vulkan and nvidia doesn't even give a fuck about opencl

>> No.62056762
File: 41 KB, 500x1130, 480.png [View same] [iqdb] [saucenao] [google] [report]
62056762

>>62056161
>780 performance
lol i would have upgraded already if it was that shit, does it feel bad that a 7970 is now better than a 780?

>> No.62057036

>>62051550
What are you talking about?
Raja took 3 architecture to finally implement it. Not even well according to the terrible driver support and literal hardware level issues
What the fuck has Raja been doing this whole time? Jen at least has billions at his disposal to do whatever he wants. Money is the ultimate nitrous injection in the semiconductor industry

>> No.62057085

>>62056025
If I'm driving at 5 mph who cares about about a 30% boost cause I still can't hit fucking 60 on the freeway

>> No.62057351
File: 73 KB, 700x565, 1502831978496.jpg [View same] [iqdb] [saucenao] [google] [report]
62057351

>>62049779
>>62050883
BASED FAVELA MONKEY

>> No.62058212

>>62057036
>Money is the ultimate nitrous injection in the semiconductor industry
You're literally in a thread where the biggest guy in semicon is getting his ass kicked by a company not even 1/10th the size and completely broke and in debt to boot.

>> No.62059469

>>62057036
>Jen at least has billions at his disposal to do whatever he wants. Money is the ultimate nitrous injection in the semiconductor industry
And yet money did not help SGI.
And nVidia is rapidly heading that way.

>> No.62059482

>>62049217
THIS COULD...
THIS COULD MEAN MY DREAMS TRUE?
RYZEN THINKPADS? RYZEN SURFACEBOOKS? RYZEN MACBOOKS?
In any case, Intel is practically dead and it's a corpse.

>> No.62059489
File: 108 KB, 631x630, 20150624_AMD_kokoro_chan.jpg [View same] [iqdb] [saucenao] [google] [report]
62059489

holy shit the incels are butthurt in here lmao

>> No.62059496

>>62059482
> RYZEN THINKPADS?
Where have you ever been?
Lenovo registred A275 and A475 ThinkPads (5 index are explicitly AMD models).

>> No.62059546

>>62049937
Holocaust the ashes of Intel and devour their kike souls.

>> No.62059615

>>62059496
OH MY FUCKING GOD
THE HYPE, MAN, THE HYPE

>> No.62059654

>>62059615
RR will deliver a swift kill on Intel's mobile markets.
Superior, more efficient cores and graphics that's like 10 years ahead of Intel's.

>> No.62059679

>>62059654
I'm not a macfag, but I hope that Apple also goes Red this or next year.
At least they will demostrate that they aren't that lost.
Hopefully Microsoft also will switch.

>> No.62059711

>>62059679
They will go red once AMD can finally implement TB3.
Which is next year.
Expect non-trashcan Mac Pro's with hopefully actual CPUs (EBYN/Jewripper).

>> No.62059745

>>62059711
>the mac pro is gonna be epyc again
Hopefully mememagic will make Tim Cook say this.

>> No.62059763
File: 105 KB, 255x345, jew killer.png [View same] [iqdb] [saucenao] [google] [report]
62059763

Post Nazi husbandos.

>> No.62059772

>>62059745
It would be a nice return to form.
Current trashcan killed whatever was left of Mac Pro.

>> No.62059795
File: 36 KB, 665x385, timcookappleceogay-665x385.jpg [View same] [iqdb] [saucenao] [google] [report]
62059795

>>62059745
>see pic
yeah, he's gonna stay with intel

>> No.62060823
File: 1.25 MB, 1002x1480, t.JUSTnich_0.png [View same] [iqdb] [saucenao] [google] [report]
62060823

>>62049213

>> No.62062147
File: 96 KB, 638x710, 1500260377912.jpg [View same] [iqdb] [saucenao] [google] [report]
62062147

>>62059763
DELEEEEEEEEEEEEET

>> No.62063302

>>62052511
You're gay.

>> No.62063506

>>62057085
see >>62056762
its like you deny facts, still playing 1440p 60 fps in the latest gaymes, stay mad favelanigger

>> No.62063561

>>62054995
It's the american way to cuck a chink.
Also IBM probably knew G80 would make Cell useless anyway.

>> No.62063580

>>62049213
>This comes a few days after Intel's CEO left the American Manufacturing Council.

what are the ramifications to this?

>> No.62063596
File: 365 KB, 500x275, 1500318173973.gif [View same] [iqdb] [saucenao] [google] [report]
62063596

>>62059763

>tfw you will never be a contract Keller

>> No.62063624
File: 27 KB, 1440x1080, maxresdefault.jpg [View same] [iqdb] [saucenao] [google] [report]
62063624

>>62059763
OY GEVALT

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