Nvidia Pascal

Nvidia’s next-generation graphics processors are code-named “Pascal”. Based on the company’s roadmap that it demonstrated back in March, Pascal GPUs are due sometime in 2016. Next-generation graphics chips from Nvidia will support stacked high-bandwidth dynamic random access memory (DRAM) (SK Hynix’s high-bandwidth memory (HBM) or Micron’s hybrid memory cube (HMC)), unified memory addressing for CPU and GPU, NVLink interconnection for high-performance computing platforms as well as new graphics, compute and multimedia features that will be a part of DirectX 12, OpenGL 5.0 and other forthcoming application programming interfaces.


In 2016 displays with ultra-high-definition (UHD) resolutions like 4K (3840*2160, 4096*2160) or 5K (5120*2160) will get much more popular than they are today. Therefore, dramatically improved graphics processing horsepower of Pascal GPUs as well as extreme bandwidth provided by stacked HMC or HBM DRAM devices (we are talking about 1TB/s – 2TB/s bandwidth here) will be appreciated by the market.


What is Pascal?
Just as the Volta GPU was expected to, Pascal will have “3D stacked memory”—memory chips stacked one on top of the other and joined by circuitry, resulting in graphics cards that can access up to 1 terabyte per second (TBs) of bandwidth. The 1TB/s threshold will be achieved by stacking memory on top of the GPU itself, with a silica substrate between them, and then cutting a hole through the silicon to connect each layer. Nvidia’s current GPUs can access only 288GB/s; upcoming Pascal graphics cards, then, will be about 3.5 times faster.
According to Nvidia’s CEO, Jen Hsun-Huang, 1TB/s is fast enough to move “all of the data from a full Blu-Ray disc through the chip in 1/50th of a second.” It should be noted, though, that manufacturing GPUs with a terabyte of DRAM on them would make for some rather expensive graphics cards. Nvidia says that how much memory the Pascal cards will actually contain will come down to cost. Also, one must consider how much bandwidth are graphics card buyers willing to shell out for?


http://www.kitguru.net/components/graph...
http://www.kitguru.net/components/graph...



Leta 2016 pride Pascal in leta 2017 Volta. Če prav razumem bo Pascal za splošno rabo se pravi za nas Volta pa bolj za super računalnike?

NightHawk je 31. jan 2015 ob 13:18 izjavil:

RejZoR je 31. jan 2015 ob 11:02 izjavil:

Torej prvo vprašanje pred vsemi, will it be gimped?

Verjetno ga bodo ob izidu res gimpali na kakih 30% svojih polnih zmogljivosti, da bo direkten konkurent takratnemu AMD flagshipu.

Bodo mogli, ja, malo usmiljenosti do AMDja morajo imet, kot to pocne Intel

m0LN4r je 31. jan 2015 ob 13:33 izjavil:

A nebo še vmes maxwell 2.0 na 20/16nm, ker pač konkurenca še ni tako daleč?

Maxwell 2 je ze. GM204, GM206 in prihajajoci GM200.

Seveda ne. Zakaj bi izdali 2x hitrejšo kartico za enak denar, če lahko vmes naredijo 3 serije, ker jim noben ne konkurira dovolj.
"we have to criple our cards, because we want to be fair against former ATi"

Če bi bila _dejansko_ dirka za preformance (konkurenčnost) in market se to kar se dogaja, nebi dogajalo.
Govorim o trendu zadnjih 3 letih. Torej od 2012 naprej.

Saj ni nic narobe z AMD-jem, malo jih poraba matra, cena je pa kar dobra.
Tako se pač grejo vsake toliko, ko nVidia predstavi naslednjo generacijo kartic.

Če AMD izstopi iz igre bo sel razvoj verjetno še počasneje, dokler se ne pojavi nekdo, ki bo boljši ali enako dober kot Nvidia. To bi pa seveda izredno pohitrilo napredek.

Verjetno tisti, ki ima dovolj denarja, da AMD kupi.

NVIDIA Pascal GPU Still Aimed At 2016 Launch - 10x More Performance Compared To Maxwell With FP16, NVLINK and 3D Memory

http://wccftech.com/nvidia-pascal-gpu-g...

There are currently two Pascal cores that we know about - the PK104 and the PK100. The PK104 will be the GM204 successor while the PK100 will be the flagship Pascal GPU from Nvidia. Both will be manufactured on either the 14nm or 16nm node using FinFETs
The use of HBM2 will let Nvidia put upto 32GB of video buffer on their GPUs and upto 1TB/s of bandwidth.

http://wccftech.com/nvidia-ceo-talks-pa...

10nm bo verjetno drugi gen... prvi ziher ne :)

Kdaj pa pride Gtx 980 Ti?

za 980ti je vprašanje ali bo tako, ker odvisno od 390 serije koliko bo močna in s kakšno ceno, saj so prejšnjič tako naredili, da je bila 680 104 čip in nato 780 pohabljen 110 čip, torej zna biti 990 (v kodah je sicer gm200-310 in ne 400) recimo pa potem šele 990ti kot fully blown 110 oziroma 200

Misliš, da Nvidia nima kakšnega insiderja pri AMDju, da bi jim razkril, kaj zmore 390x? In vice versa.

10nm še nekaj časa ne bo, brez skrbi.

Ajoj kake ti klanfas...
680 in 780 sta popolnoma razlicne grafe, halo.
Zgodilo se bo pa to, kar se je pri 780 in 780ti, s tem da bo odvisno od AMDja, kolko bodo karto odprli. Full odprt gaming cip je samo Titan in 980ti bo nekaj med 780 in Titanom, ceprav je 780ti bla slabsa od Titana samo v computingu in kolicini rama, ce se dobro spomnem, je pa v spilih sisala se Titana v vecini primerov.

veliki GP100 Pascal naj bi Q1 2016 izšel, zadeva bo na 16nm in najprej z velikim GP100 in ne kot pri Keplerju in Maxwellu z GK2014 oziroma GM107 http://www.guru3d.com/news-story/nvidia...

My info says big Pascal has taped out, and is on TSMC 16nm (Unknown if this is FF or FF+, though I suspect it is FF+). Target release date is Q1'16. This is a change from Kepler and Maxwell where the smallest chip (GK107 and GM107 respectively) taped out first. Maybe the experience with 20nm was enough for NV to go back to their usual big die first strategy. Given the huge gains in performance compared to 28nm, and the fact that the 16nm process is both immature and quite expensive, I suspect the die size may be a bit smaller than what we've seen with GK110/GM200.

Glede nato, da maxwell konkurira Fijiju še vedno z GDDR5, bi znalo bit to upanje za resničen napredek v smeri 4K gaminga konc naslednjega leta.

IMHO v primeru, da bo Fiđi samo cenovno konkurenčen pa je možnost za en grdi scenarij od nvidije:
Entuzijast: bo Pascal + 16nm + 3D Mem
Ostalo: Maxwell 2.0 rebrandi na 16nm, mogoče z 3D mem ni pa mus. Če izhajamo iz tega, da se prenese recimo 980ti pa podobno na 16nm je že v bisvtu konkurenca Fijiju.

Vglavnem vse je odvisno od Fijija ali bo superioren ali bo le prehodni izdelek.

Tukaj so tudi že podpisane pogodbe z Oak Ridge, kjer ima NV v bistvu prodanih že cel kup Volta boardov. Torej stvari se nekak še vedno morajo premikat.

NVIDIA Pascal GPU To Feature 17 Billion Transistors and 32 GB HBM2 VRAM - Full CUDA Compute Architecture Arrives in 2016


http://wccftech.com/nvidia-pascal-gpu-1...

Vec rama na grafi, kot ga sam se ene par let v sistemu ne bom imel
Jao, naj AMD ze nekaj strupenega ven rukne, ker ce ne bomo vsi kupovali za prevec denarja ta HW

Govorim o necem strupenem v praksi, ne samo na papirju
Da te asociram na novico v katero si se vpletu

Karkoli bo AMD naredil, bo nvidia se vedno korak spredaj. Tudi ce jim rata, cena ne bo ugodna, kar vidimo sedaj, pa nimajo nic posebnega na voljo.

V novicah poročajo, da se nvidia zelo močno trudi za Pascal:

A composite sketch of the new Pascal based graphics card leaked this morning. The professional expert (who asked to stay anonymous) stated that the card is going to look like this:

Se vedno bolje, kot skrpucálo od fury x.

glede na to skico bo dolga ene 27cm+

hojnikb je 25. jul 2015 ob 12:19 izjavil:

> Jao, naj AMD ze nekaj strupenega ven rukne, ker ce ne bomo vsi kupovali za prevec denarja ta HW

sj so dal ven FURY X :)

Fury X ni nič durgega kot malce izboljša 4 leta star GCN z HBMjem.
Zdaj pa ti malo izboljšaj keplerja na teselaciji pa dodaj HBM.