Gráfica NVIDIA Turing (GTX 1600 series, RTX 2000 (Super) series, Quadro RTX)

@lagadu para o caso de não saberes, mudar de processo de fabrico implica redesenhar quase tudo. Nestes novos processos não é apenas um optical shrink como acontecia há uns anos atrás ;) A maioria dos macro blocos foram mantidos (daí o IPC semelhante) mas a implementação dos mesmos não tem nada a ver. Para além disso tiveram efectivamente mudanças de arquitectura como se pode ver a diferença de desempenho com async ;)
 
Se agora a nVidia lançar um "refinamento" com 5% de melhor performance para espremer a vaca, então sim, tens uma situação análoga. Que foi a hipótese que o muddymind colocou.
Já tivemos um cenário parecido, em que vias praticamente uma x60 a ter o perfomance da x70 anterior , x70 a ter o perfomance de um x80, daí terem usado as TI para espremer ao maximo.

Ps: Não me importava de esperar pelas 1170 e terem o perfomance das 1080 , desde que passem a custar como as antigas x70 ( 400 euros preço de lançamento)
 
Quase dobrou a performance porque vem quase com o dobro dos clocks base e extra cores. Em termos do hardware fazer coisas novas ou ter arquitectura diferente quase que não teve nada.

Aliás, o roadmap de há uns anos atrás passava de Maxwell para Volta directamente.

Na pratica, que é o que interessa, sairam bem mais rapidas. Em laptop então a diferença é enorme.
Efectivamente esticaram um bocado a corda, mas acredito que se deve às memorias hbm não estarem ainda disponiveis com volumes decentes. O volta só deve aparecer lá para 2018.
 
Ps: Não me importava de esperar pelas 1170 e terem o perfomance das 1080 , desde que passem a custar como as antigas x70 ( 400 euros preço de lançamento)

Até mais do que isso . a 1070 tem uma performance identica à 980ti , algo que ja tinha escrito aqui antes das 1xxx terem saído ..o maior problema é não haver concorrência para baixar os preços
 
Até mais do que isso . a 1070 tem uma performance identica à 980ti , algo que ja tinha escrito aqui antes das 1xxx terem saído ..o maior problema é não haver concorrência para baixar os preços

Era mais no caso se o cenário ficasse como ficou a uns anos :P Tambem concordo se houvesse concorrência por parte da amd as 1070 estava a custar menos 125 euros do que o preço atual.
 
NVIDIA reportedly set to unveil Volta-based GeForce cards in Q3 2017 - Tweakdown

NVIDIA is dominating the high-end and enthusiast markets with so many graphics cards that AMD will have trouble fighting them off with Radeon RX Vega, but the GTX 1070, GTX 1080, GTX 1080 Ti, and TITAN X(P). Oh, and don't forget about the new GTX 1060 9Gbps, GTX 1080 11Gbps, and the new TITAN Xp.

We've reported last October that NVIDIA were preparing their next-gen GeForce GTX 2080 flagship graphics card, with a full GTX 2080 line up expected in 2018. It seems NVIDIA has "changed plans" and will launch a new series of graphics card later this year. Now we have rumors flying at us from MyDrivers, with the Chinese website teasing:
  • Pascal lineup is complete now that NVIDIA launched Titan Xp and is about to launch GeForce GTX 1030
  • Based on their traditional upgrade cycle we should expect Volta by the end of the year or early 2018, but NVIDIA changed plans
  • Market situation in Q1-2017 was bad, graphics and other hardware sales in decline, price war, meager profits, sales not increasing the way they want to (vicious circle)
  • According to the latest news, NVIDIA plans to release GeForce 20 series in advance, it is now expected to be released in the third quarter of this year
  • GeForce 20 series product prices, market positioning will be replanned
  • By launching sooner than expected they will be able to increase prices of single cards and increase profits
Although it's strange, as the rumor states NVIDIA is preparing to launch the GTX 1030 - and until now, we haven't heard a peep about the GTX 1030. The next part tackles what we're all here for; Volta. According to this rumor, NVIDIA has "changed plans" and will launch a Volta-based graphics card this year, instead of the previously reported early 2018 window.

I think we'll see:
  • GTX 2070 with 11GB of GDDR5X @ 12Gbps
  • GTX 2080 with 11GB of GDDR5X @ 14Gbps
  • TITAN Xv with 16GB of HBM2 @ 16Gbps
In early/mid 2018, I think NVIDIA will launch:
  • GTX 2080 Ti with 16GB of HBM2 @ 16Gbps
  • GTX 2070 Ti with 11GB of GDDR5X @ new speeds (16Gbps??)

NEWS SOURCES:News.mydrivers.com, Hardforum.com, Tweaktown.com, Tweaktown.com

http://www.tweaktown.com/news/57177/nvidia-launch-next-gen-volta-based-geforce-q3-2017/index.html
 
não devia ser 1180/1170 em vez de 2070/2080 ou isso será dual gpu ou simplesmente marketing? Estou a curtir a Titan Xv se se for realmente chamar assim :bounce:
 
A diferença de GDDR5X para HBM2 é de apenas 14Gbps para 16Gbps? (teoricamente)

A diferença vai muito além disso. São arquitecturas totalmente distintas e não são comparáveis nesses termos, não sei onde é que ele foi buscar aqueles 16Gbps.

A GDDR5(X) corre a frequências elevadas em canais de 32bits e buses totais múltiplos deste valor. Um chip da GDDR5X mais rápida actualmente disponível, de 11Gbps, dá 44GB/s num canal de 32bits. No bus de 256bits da 1080 isto dá um total de 352GB/s, por exemplo.

A HBM corre a frequência baixa mas em buses muito largos. Cada stack de HBM2 pode correr a 2GT/s e usa um canal de 1024bits, para um total de 256GB/s. A R9 fury usa 4 stacks de HBM a 1GT/s, num bus total de 4096bits para uma largura de banda total de 512GB/s. Uma implementação semelhante com HBM2, que funciona ao dobro da velocidade, pode chegar a um massivo 1TB/s.
 
A diferença vai muito além disso. São arquitecturas totalmente distintas e não são comparáveis nesses termos, não sei onde é que ele foi buscar aqueles 16Gbps.

A GDDR5(X) corre a frequências elevadas em canais de 32bits e buses totais múltiplos deste valor. Um chip da GDDR5X mais rápida actualmente disponível, de 11Gbps, dá 44GB/s num canal de 32bits. No bus de 256bits da 1080 isto dá um total de 352GB/s, por exemplo.

A HBM corre a frequência baixa mas em buses muito largos. Cada stack de HBM2 pode correr a 2GT/s e usa um canal de 1024bits, para um total de 256GB/s. A R9 fury usa 4 stacks de HBM a 1GT/s, num bus total de 4096bits para uma largura de banda total de 512GB/s. Uma implementação semelhante com HBM2, que funciona ao dobro da velocidade, pode chegar a um massivo 1TB/s.
Alem de ser energeticamente mais eficiente. A GDDR5x por causa das frequências altissimas consomem muita energia.
O problema do HBM é a integração da memoria em stack e as ligações ao processador gráfico que vai encarecer o bastante o produto .
A Amd em 2015 teve dificuldades na 1ª geração de HBM por causa disso e ao facto de na altura não existir densidades de memoria flash que permitissem capacidades superiores a 4GB total, 1GB /stack x 4 stacks. em 2016 a Samsung aumentou a densidade por stack até 4 GB.
A R9 fury X não vingou porque a concorrência usava 8 GB GDDR5X contra os 4GB HBM alem de mais cara de produzir, mais cara para o consumidor final fazia com que a R9 fury não fosse um sucesso de vendas.
Mesmo assim só com 4Gb consegue ainda hoje bater-se com placas gráficas com 6 e 8 GB GDDR5x.
Nada mau para um chip gráfico de 2015
 
Techpowerup:

NVIDIA Announces Its Volta-based Tesla V100

Today at its GTC keynote, NVIDIA CEO Jensen Huang took the wraps on some of the features on their upcoming V100 accelerator, the Volta-based accelerator for the professional market that will likely pave the way to the company's next-generation 2000 series GeForce graphics cards. If NVIDIA goes on with its product carvings and naming scheme for the next-generation Volta architecture, we can expect to see this processor on the company's next-generation GTX 2080 Ti. Running the nitty-gritty details (like the new Tensor processing approach) on this piece would be impossible, but there are some things we know already from this presentation.

This chip is a beast of a processor: it packs 21 billion transistors (up from 15,3 billion found on the P100); it's built on TSMC's 12 nm FF process (evolving from Pascal's 16 nm FF); and measures a staggering 815 mm² (from the P100's 610 mm².) This is such a considerable leap in die-area that we can only speculate on how yields will be for this monstrous chip, especially considering the novelty of the 12 nm process that it's going to leverage. But now, the most interesting details from a gaming perspective are the 5,120 CUDA cores powering the V100 out of a total possible 5,376 in the whole chip design, which NVIDIA will likely leave for their Titan Xv. These are divided in 84 Volta Streaming Multiprocessor Units with each carrying 64 CUDA cores (84 x 64 = 5,376, from which NVIDIA is cutting 4 Volta Streaming Multiprocessor Units for yields, most likely, which accounts for the announced 5,120.) Even in this cut-down configuration, we're looking at a staggering 42% higher pure CUDA core-count than the P100's. The new V100 will offer up to 15 FP 32 TFLOPS, and will still leverage a 16 GB HBM2 implementation delivering up to 900 GB/s bandwidth (up from the P100's 721 GB/s). No details on clock speed or TDP as of yet, but we already have enough details to enable a lengthy discussion... Wouldn't you agree?

0b9fd209ced2.jpg

Guru3D:

Nvidia Shows and Demos Testla Volta V100 has 5120 Shader processors


Over at GDC Nvidia has announced the Testla V100 processor, this is a Volta based GPU based on Tensor architecture. Tesla Volta V100 will be made on TSMC’s 12nm finfet process, pushing the limits of photo lithography as this GPU is huge.

The Tesla Volta V100 graphics processor has 5,120 CUDA / Shader cores and is based upon a massive 21 Billion transistors. It offers what Nvidia calls 120 Tensor TeraFLOPS of performance. Gaming wise it would perform in the 15 TFLOP region, delivered by a type of processors called Tensor Cores. The R&D behind this did cost Nvidia about $3 billion, CEO JHH says. The first server and deep learning segmented products based on Tesla Volta V100 will become available in Q3 2017.


index.php



The new Tensor Core is based on a 4×4 matrix array and fully optimized for deep learning. Nvidia stated, they felt Pascal is fast, but isn't fast enough. I already stated that the GPu is huge, it's 815mm2 huge and would fit roughly the palm of your hand.

  • Massive 815mm2 die size
  • 12nm FinFet (TSMC)
  • 21B Transistors
  • 15 FP32 TFLOPS / 7.5 FP64 TFLOPS
  • 120 Tensor TFLOPS
  • 16GB HBM2 which manages @ 900 GB/s
  • 5120 Shader processor cores


index.php



Tesla Volta V100 is capable of pushing 15 FP32 TFLOPS and much like Pascal GP100 is once again tied towards HBM2 graphics memory (stacked on-die cache). The unit will get 16GB of it divided over four stacks (4GB per stack). The memory is fabbed by Samsung. That HUGE die at 815 mm2 is fabbed by TSMC on a 12nm FFN fabrication process. In Q3 you will see the first enterprise based products based on Volta that start at 69.000 dollar. For us gamers, when GeForce GTX 1180 or 2080 will be released. That remains to be topic of a long discussion :-) Below a comparative specification list of the primary Tesla GPUs running up-to Volta, which runs in the 1.45 GHz marker for Boost frequency btw. And that is fairly comparable towards the current Pascal GP100.


index.php


nyJkCQp.png

PcPer:

NVIDIA Announces Tesla V100 with Volta GPU at GTC 2017

During the opening keynote to NVIDIA’s GPU Technology Conference, CEO Jen-Hsun Huang formally unveiled the latest GPU architecture and the first product based on it. The Tesla V100 accelerator is based on the Volta GPU architecture and features some amazingly impressive specifications. Let’s take a look.

f2933f1f-de1e-46f6-838f-ec5f48fdc6d3
f2933f1f-de1e-46f6-838f-ec5f48fdc6d3
ocWlfbR.png



While we are low on details today, it appears that the fundamental compute units of Volta are similar to that of Pascal. The GV100 has 80 SMs with 40 TPCs and 5120 total CUDA cores, a 42% increase over the GP100 GPU used on the Tesla P100 and 42% more than the GP102 GPU used on the GeForce GTX 1080 Ti. The structure of the GPU remains the same GP100 with the CUDA cores organized as 64 single precision (FP32) per SM and 32 double precision (FP64) per SM.







Interestingly, NVIDIA has already told us the clock speed of this new product as well, coming in at 1455 MHz Boost, more than 100 MHz lower than the GeForce GTX 1080 Ti and 25 MHz lower than the Tesla P100.





Volta adds in support for a brand new compute unit though, known as Tensor Cores. With 640 of these on the GPU die, NVIDIA directly targets the neural network and deep learning fields. If this is your first time hearing about Tensor, you should read up on its influence on the hardware markets, bringing forth an open-source software library for machine learning. Google has invested in a Tensor-specific processor already, and now NIVDIA throws its hat in the ring.

Adding Tensor Cores to Volta allows the GPU to do mass processing for deep learning, on the order of a 12x improvement over Pascal’s capabilities using CUDA cores only.






For users interested in standard usage models, including gaming, the GV100 GPU offers 1.5x improvement in FP32 computing, up to 15 TFLOPS of theoretical performance and 7.5 TFLOPS of FP64. Other relevant specifications include 320 texture units, a 4096-bit HBM2 memory interface and 16GB of memory on-module. NVIDIA claims a memory bandwidth of 900 GB/s which works out to 878 MHz per stack.

Maybe more impressive is the transistor count: 21.1 BILLION! NVIDIA claims that this is the largest chip you can make physically with today’s technology. Considering it is being built on Samsung’s 12nm FinFET technology and has an 815 mm2 die size, I see no reason to doubt them.






Shipping is scheduled for Q3 for Tesla V100 – at least that is when NVIDIA is promising the DXG-1 system using the chip is promised to developers.

I know many of you are interested in the gaming implications and timelines – sorry, I don’t have an answer for you yet. I will say that the bump from 10.6 TFLOPS to 15 TFLOPS is an impressive boost! But if the server variant of Volta isn’t due until Q3 of this year, I find it hard to think NVIDIA would bring the consumer version out faster than that. And whether or not NVIDIA offers gamers the chip with non-HBM2 memory is still a question mark for me and could directly impact performance and timing.

More soon!!
 
portanto, a Volta é um shrink do Pascal com mais transístores e, consequentemente, mais cuda cores. O tamanho é monstruoso, o preço... nem quero imaginar
 
Bom, vamos começar os trabalhos...

Começando pelo flagship o GV100. Nem quero imaginar os yields de um animal deste calibre. 815 mm2 é um record de tamanho :D

O interessante é que tem um aumento de 40% nos transístores e shaders, um clock boost semelhante ao GP100 e o TDP mantém-se nos mesmos 300W! Portanto se "transportar" isso para as Geforce's, xx80 com 3500 shaders e xx60 com 1800. Depois claro deve vir a xx80 / Titan Xv com um shader count semelhante ao GV100, mas sem FP64 e Tensor core...

Portanto quando a AMD Vega sair, já a nVidia vai destruir elas ao pequeno almoço, nem duvido nada que a 2060 fique ao nivel da Vega de entrada e basta a 2070 para estoirar a flagship...
 
Se vier com HBM2 isso é mais do que garantido. Mas se os primeiros skus vierem com gddr5x ainda são capazes de lançar/anunciar qualquer coisa até ao fim deste ano.
 
A não ser que:

SK Hynix is displaying its next generation GDDR6 memory modules at Nvidia's GTC event in San Jose, California. According to the memory maker, GDDR6 operates with an I/O data rate of 16Gbps and a theoretical bandwidth of 768GB/s when paired with a 384-bit I/O bus. This new memory standard allows for twice the bandwidth per pin as GDDR5 and delivers 10% lower power consumption.
http://www.tomshardware.com/news/sk-hynix-gddr6-wafer-gtc,34377.html
 
Back
Topo