https://www.skhynix.com/eng/pr/pres sReleaseView.do?seq=2086

>>60037301
>2Znm
What did they mean by this?

>>60037301
>384-bit I/Os
topkeke

>>60037301
>AYYMD IS FINISHED & BANKRUPT
>Intel /nvidia prices increase thrice

>With a forthcoming high-end graphics card of 384-bit I/Os, this DRAM processes up to 768GB(Gigabytes) of graphics data per second.

>More memory bandwidth than AYYMD HOUSEFIRES 512GB/s max HBM2 2 modules

AYYMDPOORFAGS CONFIRMED ON SUICIDE WATCH

>>60037301
>new memory is released, is faster than older memory

HOLY PENIS WHAT A SCOOP

>>60037301
>wait for pascal
>wait for vega
>wait for gddr6
REEEEEEEEEE

File: UNILAD-tampon-protest-3420134572.jpg (64KB, 726x378px) Image search: [Google]

64KB, 726x378px

>>60037301
you do know hbm3 is already in the works and is post to provide 64gb of vram and 1tb+ of bandwidth right?

also this is old news. hbm3 and gddr6 info was released back in august of 2016. everyone knew hynix was going to release gddr6 in 2017 with the first cards utilizing it in 2018.

https://arstechnica.com/gadgets/2016/08/hbm3-details-price-bandwidth/
>HBM3: Cheaper, up to 64GB on-package, and terabytes-per-second bandwidth

>>60037301
i guess nvidia btfo too since they're using hbm2 in their high end pascal workstation cards. wow everyone btfo and finished~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Oh good I can't wait for another 8-10 GDDR modules on my card sucking up 40% of the card's power budget

>>60037509
512-bit 16GHz (lel) GDDR6 = 1024 GB/s
4-stack 1024-bit 2GHz HBM2 = 1024 GB/s

Fury X = 4096-bit bus @ 500MHz = 512GB/s on HBM1

>>60037712
>Both Hynix and Samsung are working on HBM3 (Samsung is calling its version "Extreme HBM"), although neither has committed to a firm release date. AMD's current GPU roadmap points to its Navi architecture—the follow up to Vega, which itself follows Polaris—using "NextGen memory," and is due for release in 2018.
>Both Hynix and Samsung
>Hynix ALSO WORKING ON GDDR6
HYNIX BTFO AND FINISHEDDDDDD HYNIX! HOW WILL HYNIX EVER RECOVER?

>still 32bit per module

Lol

>>60037841
This. The Pascal GP102 already reaches faster speeds than HBM2. GDDR6 will probably reach 1 TB/s on a 384bit bus.

>>60038384
>by 10GB/s
>while using 4 times the power
Uh, yeah.

>>60038416
AMD uses 4 times the power? It is not that bad, but yes, I think Vega will consume at least 50W more.

>>60038441
I know you're intentionally being retarded, but HBM will always consume much less power than off-die memory

File: AMD-HBM-Die-Stacked-Memory.jpg (54KB, 800x450px) Image search: [Google]

54KB, 800x450px

>>60038441
HBM2 vs GDDR5X.
One stack of HBM2 has more bandwidth 5 modules of GDDR5X, it also uses a shitload less power.
Power you could be using for the GPU.

>256bit GDDR6 card will have 512GB/s of memory bandwidth, same as HBM2 2 stack meme but much cheaper due to not dealing with immature through silicon vias & interposer cost

TOP KEK, AYYMDPOORFAGS

>>60038473
HBM should also have lower latency.
On top of that, it uses way less power and not waste PCB space so you can get similar stuff to a Nano.

GDDR* is only okay for midrange-lowend stuff

SK Hynix also makes HBM so I guess they're btfo-ing themselves?

>>60038565
Nvidia also uses HBM2 in P100 (mezzazine and PCIe) so I guess they also got BTFO

File: 1491513758321.png (44KB, 198x255px) Image search: [Google]

44KB, 198x255px

>>60037617

Wait™

GDDR6 follows the same path as GDDR5/X, which is to upclock the memory in return for higher bandwidth. But this also means it consumes far more power.

HBM will always be the better implementation of the two. Die stacks + larger memory channels at lower speeds = better than non-die stack + smaller memory channels and niggahertz speeds.

>>60037443
underrated

>>60043325
No, seriously, what did they mean by this? I'm curious.