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 ]
[ERROR] No.66612606 [Reply] [Original] [archived.moe] [rbt]

Why is this thing so big?

And why isn't AyyyyMD focusing on single core performance and creating more efficient CPU's instead of super-gluing 4 dies together?

>> No.66612626

>>66612606
>Why is this thing so big?
Because it has so many cores.

>And why isn't AyyyyMD focusing on single core performance and creating more efficient CPU's instead of super-gluing 4 dies together?
They're just about neck and neck with Intel on single core now with Ryzen 2. You been living under a rock, or are you getting half a rupee for this thread?

>> No.66612631

>>66612606
>single core performance
Because at 4 GHz computers a single thread is already insanely fast. And most workloads are multithreaded these days. Example
- Word
- UI rendering
- disk access
- browser in background
- playing music
- network stack

All that happening at once. So more cores is would be better then faster cores.

>> No.66612657

>>66612626
>They're just about neck and neck with Intel on single core now with Ryzen 2. You been living under a rock, or are you getting half a rupee for this thread?

Are you retarded? The 8700k easily hits 5Ghz with an OK cooler and it murders Gayzen 2.

You idiots fell for the Muli-Core™ meme.

>> No.66612659

>>66612631
>b-but muh gaymen!!
Intel eternally BTFO.

>> No.66612672

>>66612606
>Why is this thing so big?
I/O. Underneath the heat spreader is mostly empty space. Those I/O pins take up space on the other side though.

>> No.66612675

>>66612606
because Intel has made no real advances. They can design a smaller, cheaper glue chip and still be competitive.

>> No.66612689

>>66612606
Ahaha the intelet is absolutely seeeething

>> No.66612692

>>66612675
I dunno man, my 7900x is a fucking beast. And I'm not even overclocking.

If they do drop the 7980XE to $800 or some shit I'll definitely upgrade.

>> No.66612698

>>66612606
>And why isn't AyyyyMD focusing on single core performance and creating more efficient CPU's instead of super-gluing 4 dies together?
How's the monolithic design working out for Intel?

>> No.66612719

>>66612692
the 7900x is probably the least competitive cpu since pentium 4

>> No.66612751

>>66612692
>paying 800 dollars for a tiny upgrade
>just using it for gayming anyway

The absolute STATE of /g/

>> No.66612757

>>66612659
as if gaming was a real-world task.

>> No.66612790

>>66612606
> MCMs are pathetic glue

I guess somebody's a little behind the curve with what's going on with Icelake and Intel's Plan B for Xeons next year...

>> No.66612794

>>66612698
>How's the monolithic design working out for Intel?

I've heard that by the end of 2018 they will actually have 10nm 2c laptop chips with the iGPUs enabled!

>> No.66612878

>>66612606
it's cool, you're just a faggot.

>> No.66612897

>>66612606
AMD engineering is too stupid to think about that
Just slap more cores lmao

>> No.66612915

>>66612606
Explain to me why this is going to be such an advantage over an I5 6600K?

>> No.66612943

Single core performance is hard. Ever since Sandy Bridge, Intel has only been making small incremental improvements in SC. AMD's moar cores approach has carved them a very nice niche in the server and workstation market. Only in gaming is Intel still at the top, but not even by a huge margin.

>> No.66613012

>>66612657
>>multi-core meme
>flashback when ryzen 1000 got released
>battlefield 1
>i7-7700k bottlenecked 1080 Ti
>1080 Ti bottlenecked R5 1700X
Games are moving over to DX12 and Vulkan. Wherever those two engines are intel barely stands its ground. Enjoy your DX11 and OpenGL single core hogs with your now 6C/12T housefire and power bill generator

>> No.66613021

>>66613012
Aside from games, though. My i5 seems like it's balancing evenly across all 4 cores and I'm doing what I do every night. Watch 1080p streams and shitpost. Why is Ryzen a revolution or something?

>> No.66613046

>>66613021
>but why is a superior example better when my needs are mediocre?
Performance might be irrelevant for you, but that's not the same for everyone.

>> No.66613056

>>66613021
Its power efficiency is better than of intel, and along with those extra cores & threads it's very competitive in workstation loads.
Today's programs use threads too, I myself have an i5 with an aftermarket cooler and it's still keeping up well

>> No.66613062

>>66612943
Well, there's only so many compromises you can make to security before the performance benefit offset diminishes.

>> No.66613063

>>66613046
Is it, in fact, superior though? What's the advantage? Better single core performance for games?
>>66613056
Thanks for the info, I appreciate it.

>> No.66613298

>>66613012
>Games moving to dx12 and vulkan
I heard this 2 years ago and most are still dx11

>> No.66613578

The processor real estate is very small. It's 213mm2, meaning 600ish per 450mm wafer (this is Ryzen 1 also). The interposer is what occupies so much space; in practice it isn't necessary for it to be so big as it is passive, but it allows more pins to be allocated (4094 of them) which allow TR to accommodate the memory channels, PCIe lanes, etc.. Really only half of them need to be occupied as the same interposer is used in EPYC/SP3, the others are NC.

>>66613021
From a technology standpoint, it's scalability. The industry is moving towards scalable solutions, increased parallelism whenever possible (HPC, content creation, and even gaming and PLM/CAD). Infinity Fabric is going to be key to building a NUMA, highly scalable, high bandwidth both core to core and also core to DRAM and core to coprocessor (likely through xGMI or PCIe) infrastructure.

Gaming and the diy market is not very big although still profitable especially since gamers are so gullible from a technological standpoint; AMD understands that they don't need to be the apex of the gaming market to command a significant share back and that their design scalability is one of the major advantages towards high impact market penetration using what is practically a single design. So no, AMD at the moment does not have the "gaming" advantage, but with 7nm (feature sizes on par with 10nm), Zen 2 will very likely be well ahead in every single performance metric; it would likely flip the script completely on Intel.

>> No.66613768

>>66612606
Because bingbus was a mistake. Also because thermodynamics and shiet.

>> No.66613804

>>66612631
Multicore is even more advantageous on Linux, where the OS is dozens of small processes communicating, the kernel is multithreaded, and many applications go full multiprocess. My FX-8350 is crap at single threaded, but I'm only now starting to feel the pinch on performance because those eight cores share the load so well.

>> No.66613820

>>66612606
What's the performance on that thing when it comes to homework? How decently does it run YouTube at 480? What's the strength of the wifi signal?

>> No.66613821

Intel used to push niggahurtz advancements fast until a few years ago when it stagnated, unless you can double them again multicore is better

>> No.66613847

All.the glue betweeen the cores

>> No.66613867

>>66613821
The real problem is that x86 just isn't capable of going superduperscalar (4+ IPC) like VLIW or other more modern architectures. That would send performance through the roof and keep TDP sane. The Mill CPU or a new architecture Microsoft is working on seem to be the way forward.

>> No.66613868

>>66612692
>he unironically bought a ****x

>> No.66613874

>>66612606
>super-gluing 4 dies together
Large servers have been using multi-CPU motherboards for decades and nobody batted an eye, but put a couple dies under the same heat spreader and people have a fucking stroke for some reason

>>66612657
>confusing single-thread performance with overclocking headroom
anon those aren't the same thing and if you need an explanation for why, then you don't belong here

>> No.66613903

>>66612606
Large IHS improves heat transfer. ie. 7740X outperforms the 7700K and almost identical to the 8700K in clock speed despite inferior process.

>> No.66613968

>>66612626
People still don't understand the IPC is on par now. Even Zen 1 IPC was higher than Haswell. The only thing that lacks is clock, but it will increase further on 7nm. 4.3 GHz are already possible, 700 MHz to go.

>> No.66614205

>>66613012
No one gives a fuck about gaymes you homo

You probably still live with you mommy

>> No.66614228

>>66612719
I got it when Threadripper wasn't even out, famalam. You are gay for even saying this.

10core/20thread is enough for what I do with 64GB RAM and a Vega Frontier Edition and I'm solid for a few years.

>> No.66614234

>>66612751
I use it for media production you fucking autist. Kys please.

>> No.66614235

Lol handlets btfo

>> No.66614258

>>66612915
The 6600 is like four years old. Are you joking?

>> No.66614818

>>66614228
>Not getting 2990X this year

>> No.66614842

>>66612626
>They're just about neck and neck with Intel on single core now with Ryzen 2

The absolute state of AMDelusion

>> No.66614912

>>66612657
fuck off back to plebbit, onions boy.

>> No.66615016

>>66612606

>> No.66615039

>>66614842
Not that guy, but go ahead and prove that statement wrong, anon. You won't be able to, because he's right.
What little advantage Intel has in single core performance now come from delidding and overclocking the 8700k to the point of being a housefire.

>> No.66615100

>>66614234
Ahaha sure you do.

>> No.66615476

intel has shitty consumer practices, expensive

amd has good consumer practices, not expensive

not a tough choice for me, i'm rocking my fx8300 just fine

>> No.66615514

>>66615039
>Not that guy, but go ahead and prove that statement wrong, anon. You won't be able to, because he's right.
You cannot prove something doesn't exist. In this case, AMD's performance doesn't exist.

>> No.66615520

>>66615514
>t. couldn't find any benchmarks showing a huge gap in single-core performance between 2nd gen Ryzen and Coffee Lake

>> No.66615568

>>66612657
>with an OK cooler
Did you mean 0K (zero Kelvin)?

>> No.66615584

>>66614842
..

>> No.66615593

>>66612657

0/10

Go back to /v/

>> No.66615602

>>66614228
>he bought a 7900x prob 1 month before threadripper released
>he bought a vega fe
the memes write themselves

>> No.66616263

>>66612657
back to /v/eddit, little man.

>> No.66616332

>>66615568
Kek, underrated

>> No.66616891

>>66612606
All I want is something that can match an 4th gen i5 quad core at 3.4 ghz. Now trick is I want it to have decent graphics and tie with a 15.6 laptop and only require a 25watt or lower adapter with 8 hours battery life.
Then you will kill Intel.

>> No.66617128

>>66612606
>single core performance
>efficent

>> No.66617226

>>66612606
Xeons can't compete anon, get over it.

>> No.66617470

>>66613768
could you not talk about stuff you clearly don't know anything about? thanks

>> No.66617572

>>66617470
First half of his post is accurate though. Monolithic garbage is ancient and forever obsoleted.

>> No.66617707

>>66612719
>what is 7740X
>>66612692
Well no shit lol. If you got it before TR there's no point in selling it, you did not make an argument for anything.
>>66612657
>Gayzen 2.
It did not even come out faglord, only a refresh.
You got 8700K only thanks to AMD. You need delid to hit anything close to stable 5GHz without AIO or monstrous cooler.

>> No.66617747

>>66617470
>faglord does not know shit about tooshpaste
>faglord does not know shit about non-monolythic dies potential advantages
https://www.youtube.com/watch?v=G3kGSbWFig4

>> No.66617771

>>66612606
arm did the same thing, can't make it faster, so lets glue 8 together and see what happens.

>> No.66617999

>>66617572
why are you equating bingbus with monolithic designs? bingbus is still more than fine on relatively low core counts

>> No.66618111

>>66615039
Eh, Ryzen has a huge disadvantage in avx2, but even that is a relatively niche use case. In the tests the Stilt used, once you factor out avx2+ performance there's only a 6% difference between pinnacle ridge and coffee lake.

(ER) = extremeties removed
(>=256b) = avx2/512 benchmarks removed.

>> No.66618783

>>66617747
Anyone else see this?
>15:24
coincidence?

>> No.66618788

>>66612606
this is bait thread
dumbasses

>> No.66618810

>>66612692
OK, I don't believe anyone would actually buy this.
Post proof.

>> No.66618859

>>66615584
>174/168=1.0357...
3.5% IPC advantage.
The only thing keeping Intel afloat is that Ryzen doesn't overclock at all.
This may change with 7nm.
I'd be shitting my pants if I was Intel right now.

>> No.66618888

>>66613867
VLIW would absolutely rape cache space so I don't see it happening for general purpose CPUs

>> No.66618973

>>66615584
>2600X still beating 8700K at same clocks on MT, despite lower IPC.
Intel's Hyperthreading is garbage.

>> No.66619648

>>66612606
>super-gluing 4 dies together
I bet intel wishes they did this. Yields on the massive 28 core dies they use must be absolutely abysmal.

>> No.66619697

>>66612657
>shits about ipc
>brings OVERCLOCK to the table
>INTEL IS BUTTER BECAUSE IT CUN UVERCLUCK

fuck this idiots

>> No.66619770

>>66618888
Nice quads, but Ryzen's got absolutely massive caches. The L3 literally takes up about a 3rd of the die space on each CCX. You could double the size of a Ryzen die and still have pretty great yields on 14/12nm processes. Zen was at 80% top-bin (1800x) yield 6 months into production IIRC. Throwing more SRAM at the problem won't be an issue.

>> No.66619775

>>66619648
I actually went and did the math on this: assuming equal defect density and based off of AMD's 80% stated full fat yields, AMD would be able to get about 200 fully functional dies from each wafer. Intel would get 36 at the same yields.

Assuming all AMD dies are put to Epyc, thats about 50 or so full 32c CPUs (at about $4k each) per wafer, netting about $200k total. To match this, intel would have to sell each of the 28c dies for a minimum of $5700 USD.

Keep in mind, this doesnt take into account the fact AMD can use the other not-fully-working dies for other things depending on what's fucked. These fuckhuge 28c dies? Meant for Xeons only, so if they come off the line bad, into the bin they go.

>> No.66619801

>>66619775
>into the bin
heh

>> No.66619862

>>66612657
>You idiots fell for the Muli-Core™ meme.
So did intel.

>> No.66620435

>>66617999
>relatively low core counts
t. 2011

>> No.66620767

>>66619775

So is intel selling their cpu's really cheap? Or is AMD jewing us the fuck out?

>> No.66621603

>>66612657
₹0.05 has been deposited into your account

>> No.66621674

>>66613012
Why are AMD niggers so hypocritical?

>> No.66622779

>>66620767
>intel...cheap
lolno
>amd...jewing
lolno

>> No.66622823

>>66612606
This, Intel is better because they are all about performance per mm of silicon. That is true technological innovation. Anyone could just take a bunch of CPUs combine them, and shit out some 20 core low frequency shit.

>> No.66623066

>>66620767
I guess Intel is still making a profit, even if tiny, with their processors, so if AMD sells an equally capable product at the same price or slightly lower, that's a yuge margin for AMD, so probably a bit of the two. At this rate, EBYN will fuck Intel in the ass, unless they develop something similar to infinity fabric. Intel had something kinda like that, but can't remember its name. It also wasn't so mature.

>> No.66623781

>>66612657

>> No.66623800

>>66615568
lmao

>> No.66623825

>>66617999
t. corelet

>> No.66624666

>>66613867
>like VLIW or other more modern architectures

yeah, late '90s called and want their Itanium back.

>> No.66624765

How's about single core performance in older games? What's the use in upgrading if the new processor can't deliver smooth gameplay experience. or like FX, older games ran like shit compared to Phenom II. Older games here refers to 1996 - 2k4 titles. Single core only. Modern hardware should be able to deliver 100+ fps on older games w/everything jacked to the max and @ high resolutions w/butter smooth experience without breaking a sweat. If not, then something is wrong. Not everyone plays latest gen games. Nor does everyone use strictly multi core programs. I still use older programs cause they "just work". Why fuck up a good thing?.

>> No.66624811

>>66612606
Because you’re retarded! Don’t you know that no single threaded application needs high single core preformance? If it does and it’s still single core than the developers are shit!

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