ZEN.
Images are sometimes not shown due to bandwidth/network limitations. Refreshing the page usually helps.
You are currently reading a thread in /g/ - Technology
You are currently reading a thread in /g/ - Technology
![13275824_10100152902684159_721751408_n[1].jpg](https://i.imgur.com/H66W7SO.jpg "13275824_10100152902684159_721751408_n[1].jpg")
It's here.
>wccftech
Got some die shots and specs.
http://wccftech.com/amd-zen-cpu-performance-double-fx-8350/
![AMD-Feature-Featured-APU-CPU-635x357[1].jpg](https://i.imgur.com/0PmuEfx.jpg "AMD-Feature-Featured-APU-CPU-635x357[1].jpg")
>>54704999
Zen modules are now quad-cored, with full allocated resources given to each core. Not 1FP per two cores like with Bulldozer.
Zen is 14nm FinFET.
Posting wccftech articles should be an instan permaban.
none of this means anything until benchmarks come out.
>>54705032
So you're saying that Zen die shot is just CGI and totally not corroborating all the info the public have on Zen so far?
>>54705066
Indeed, but it's still awesome to see the new info.
>HBM APUs
AMD, MAKE IT HAPPEN
>>54705072
No, you fucking retard.
I'm saying that wccftech is clickbait bullshit combed from forum posts.
There is no comparison between Zen and Vishera. The original slide read Orochi, which is the name of all 8 core Bdver family dies. It was later revise to say Excavator, the bars are just showing a general IPC uplift.
In no way did AMD ever indicate that Summit Ridge would have twice the throughput of an FX 8350. That is entirely a wccftech invention.
You need to be banned for being this dumb.
There already is a thread about this anyway, learn to use the catalog.
>>54704999
>wccftech
Anon please. I excited for Zen, but fuck anything from that site.
>Die shot
How does that prove anything? You can't see shit without an electron microscope and even it would be hard to say what you are looking at.
>>54705134
They're accurate at least 10% of the time
>>54704999
>wccftech
It's a currynigger rumor shitsite, why do you take it as truth ?
>>54705026
Imao, modules again, and now with double amount of pseudo "cores" on each module.
>your face when you are being cucked by AMD for the second time
D E A D
E
A
D
>>54705129
In addition, currytech plagiarized Guru3d in one of their Star Wars articles.
>>54705348
Four cores are pushed together, and the cluster can't be cut down any smaller. Call it what you want.
All this means, negatively, is that there will be no dual-core Zen chips.
>>54705405
about time we are starting to get rid of dual cores
>>54705405
I don't care for as long as they did not declare their "improved single core IPC" being actually a single "module" IPC, which consist of four shitty cores. And then that explains why 4-module or so-called "16-cores" Zen chip only performs as a 4c/8t intel haswell chip.
You understand what that means, right? You will have shit-tier performance on anything that is not optimized for at least 4 threads to make a full load on a single module.
AMD did it gain.
>>54705405
Any chance of 12 core consumer chips?
>>54705533
that's an awful lot of factual evidence you have there mr pajeet
>>54705533
"no"
>Double performance of 4GHz Octacore
>with 60% IPC increase
So it's running at 5 GHz?
How can Intel even compete?
>>54705533
Shut the fuck up with your FUD you cretin.
A single Zen core has a 40% IPC uplift over a single Excavator core.
Zen cores are not small, and they don't share any resources apart from an L3.
http://dresdenboy.blogspot.de/2016/02/new-amd-zen-core-details-emerged.html
>>54705553
Zero.
Summit Ridge is only an 8 core die, and AM4 only supports dual channel memory. There are no consumer MCMs coming.
>>54705405
They can't just laser off the parts that don't work? They will probably still offer dual cores just like they do with FM2+, maybe they'll even bring back the tri-core with their SMT for 6 threads on the super cheap
>>54705558
Haha, I really hope that I'm wrong there. But, fuck, if I'm right than we will just get another fucking Vishera, but now with double amount or "cores" on the modules.
>>54705590
But AMD will probably still have from 10-16 core chips you can buy on whatever the new server platform from them is
>>54705533
>oh noes my old single threaded games will not run as greatly as i want too
this is 2006
>>54705605
Yeah for enterprise they have 16 core, and 32 core Opterons, but those come at enterprise prices.
>>54705605
I want 16 core CPU on my desktop, for no other reason but to screen my logical processor window.
>>54705590
Yeah, no surprise. Because Excavator was 1c/2t and then Zen is 1c/4t, so that's not an improvement by any means, just double amount of "cores". Imao.
>>54704999
Those aren't Zen chips.
>>54705639
So buy used xeons that run at 2.6ghz on a dual socket board
About how much is this gonna run me?
>>54705713
AMD's probably going to charge like $200 for the 4 core 8 thread chips
>>54705896
Sold.
>>54704999
nice concept-art
Im in the market for a new CPU. Curently using an overclocked phenom ii lel. What do you guys suggest? It will be mainly used for gayman
>>54705927
wait for zen then buy
>>54705124
Would the HBM memory be on die?
>>54706138
>HBM
>on die
>on
>die
Of course not.
>>54706164
Yea I figured. How would it be added to system? On RAM like modules.
>>54706195
You'd buy a new motherboard that already has the HBM soldered to the socket. HBM doesn't exist as modules (AFAIK). You'd pick motherboards based on RAM capacity.
>>54706195
HBM goes on the package, connected via interposer.
Its the only way HBM functions.
>>54706251
Nothing in this post is even remotely accurate.
>>54706258
Source:
wccftech.com/amd-zen-hbm-motherboard-ram
>>54706251
Oh cool
>>54706258
Package?
>>54706435
>being this retarded
>>54706501
Yep. HBM only signals over short distances. The longer the distance the more current and tweaking needed to maintain integrity. HBM only works on an interposer unlike HMC which has a separate configuration for long wire lengths.
An APU with HBM would look similar to AMD's Fiji GPU.
>>54706534
Yea I understand. Do you think this is gonna add a lot to the price? How about performance?
>>54706583
HBM yields are still pretty low (especially HBM2) so it'll be a little bit on the pricey side.
Performance will depend on the use case, the CPU won't really benefit from it, the GPUs performance however will skyrocket since it won't be bandwidth-starved by shitty DDR3 speeds.
>>54706583
If Raven Ridge has HBM, and it is still is an if because nothing about it is confirmed, I could see the top binned SKUs priced around $250.
As for performance, the 8 CU in Kaveri and Carrizo are still incredibly bottlenecked. They lose nearly half of their performance just from lacking adequate bandwidth, if they had HBM on package then performance would increase accordingly.
A theoretical APU with 4th gen GCN CU and HBM on package could be competitive with the GTX 950/960.
>>54706534
I want that keychain
>>54706583
It would be pricey, but it would mean you don't even need a GPU for pretty solid gaming at 1080p, so it'll be around the cost of a CPU+mid range GPU, probably with a little premium on top
>>54706683
The mere thought alone of such possibilities makes me hard as diamondillium. What an age to be alive.
Honestly though, AMD would be fools to not push out a HBM enabled APU as soon as possible, or atleast taunt the public with the development of one.
Anyone else notice how AMD's stocks very nearly hit 4.2 today, from 3.88?
>>54705896
Can I put my GTX980 on an AMD mobo?
Who's the cuck if this setup works? AMD? nVidia? Or is it the peace marriage build the world needs the most?
>>54707056
AMD LMAO
It literally hit 4.20
>>54707071
Doesn't matter what combination you do
>>54707027
They need HBM on all future APUs, theres no other way to get IGP performance where it needs to be.
I'm hoping that the 2017 launch is enough time for AMD to have built up a substantial volume of the die stacks, even if they only have 2GB on package it would be a big deal. 2GB of first gen HBM would provide up to 256GB/s memory bandwidth. Thats enough bandwidth for something like a Radeon 7870, maybe even more given the new arch, and color compression.
Kaveri/Carrizo with 2400mhz DDR3/4 has just 38.4GB/s.
>>54706656
>>54706683
>>54706932
Hopefully Zen doesnt make us disappointed.
![74f9ac176ff637774c34239973a15989[1].png](https://i.imgur.com/6ro2o5b.png "74f9ac176ff637774c34239973a15989[1].png")
>>54707075
You sure senpai?
>>54707124
If it's even only a 40% IPC uplift, if AMD still offers more cores/threads for less than intel it'll be a win
>>54707143
Summit Ridge is 8 cores.
The new Broadwell-E will have up to 10 cores.
AMD isn't the moar coars company any longer.
>>54707168
Broadlel-E will be like $1500 or more apparently.
I'd rather just go for a Haslel Xeon, they'd be cheaper.
>>54707120
Theoretically, what kind of performance could one estimate out of HBM(1), 8 CU, and DCC.
I would add 14nm to that pile too, but we've yet to see real world tests of that, only what AMD have told us at select events.
Pick any of the above 2, and we have something we can draw on from the current marketspace. But all 3?
>>54707168
Current intel 8 cores cost like 1k man
>>54707177
Regardless, that is what Summit Ridge is competing against.
AMD's 8 core parts aren't competing against intel's highly clocked quad cores.
>>54707180
Pic related shows about how Kaveri would perform if it had adequate bandwidth. Radeon 7750 with 8 GCN1 CU paired with DDR3 vs the same card with GDDR5.
Kaveri has newer arch than this, but the performance per clock wouldn't be radically different.
The 3rd gen GCN in Carrizo would improve performance a little further.
I can't say anything about the newest 4th gen GCN as we haven't seen it yet, but its the largest generational upgrade in the line. The per clock performance uplift is supposed to be pretty big.
If I had to guess, I'd say 8 4th gen CU is comparable to 12 1st gen CU. Thats a complete shot in the dark though.
>>54707071
Use a Fury X and 980 on an Intel motherboard.
Single thread performance is the only thing that matters. Just look at this benchmark:
http://cpu.userbenchmark.com/Compare/Intel-Core-i7-5960X-vs-Intel-Core-i3-6320/2580vs3535
>>54707710
I'm retarded: the post
>>54704999
Trips. It's happening.
Zen will be what Bulldozer was supposed to be, but adjusted for a 4 year gap.
>>54705533
>>54705599
Jim Keller specifically spoke against the half-resources and low core performance approach
He's been interviewed on video, he talks about adding ARM components for efficiency, dropping to 14nm, improving silicon quality and yields, giving cores more resources and individual power WITHOUT going back down from 8, etc.
>>54707710
I love being able to run file compressions on ultra while playing crysis at locked 60fps.
Posted using my FX-8320
Not buying until 12 or more cores.
>>54710185
Shit man. I'm fine with an 8 core with SMT at $500. I'd buy the shit out of it if it's faster than Haswell.
>>54704999
>double 8350
the i5-6600k manages to do that right now
so glad I'm upgrading this summer and not waiting
>>54704999
>I want to believe
>>54708597
>Crysis 60fps
>8320
unless that shit is at 5.0ghz you're lying
post a video
>>54710506
if you're talking single thread workloads, yes.
but imagine 6600k performance, only 8 cores and hyperthreading. With applications using multithreading more than ever, it could be great. Even intel is thinking of making a 6 core consumer chip now.
>>54710506
What do you even have? I was eyeing the 5820k, but decided to wait for more cores mostly for building and ffmpeg.
I'd like to know the feel of make -j16.
>>54710561
6 Core? Why do you need 4 Cores for? I don't have 2 Cores!
*runs to muh shekels*
>>54712696
kek
>>54705405
>no dual core
But why? I can't think of anything my mother does that will even saturate a single core of her g3260.
>>54707071
It makes zero sense to lock your already failing CPU chipset to your shrunken GPU one. Its a fucking PCI-E slot. It works with PCI-E devices. You can plug an intel PCI-E SSD into an AMD motherboard.
Go back to /v/. Or just go out and get an education, dingus.
