ATi R600

Crusher disse:
"The X1300 has between a 30% to 44% performance advantage over the X600, which is again a little over the theoretical differences between them, and well above their bandwidth differences, indicating that the architectural differences are helping out fairly sigificantly."

Claro que não é só o Ring Bus a ajudar aqui. Mas se vires os benchs com o uso de AA o beneficio é maior.


Mais uma vez aponto o caso do Oblivion. A X1800XT em outdoors tem uma performance quase igual à X1900XT. A X1900XT tem 48 pixel shaders, o triplo da X1800XT, que nesse caso "extremo" não lhe servem de nada. Nesse caso o controlador de memória mostra que é essencial.

blastarr estás a confundir a cache nos CPUs Netburst que aumentou para colmatar falhas na arquitectura original, com a implementação de raíz de memória EDRAM num GPU que é usada para tirar proveito da largura de banda gigantesca da memória.
O primeiro foi um remendo, o segundo foi pensado desde o início.



Ansatsu disse:
Aliás no Xenos a Microsoft é que impigiu á ATI para por EDRAM.
Fonte?





blastarr disse:
Quando começaram a sair aplicações com shaders, já a Geforce 3 tinha sido ultrapassada pela GF4 e a primeira era considerada lenta demais para ser usável.

Errado.


Os shaders 1.1 na Geforce3 foram um bom investimento a médio-longo prazo e o crossbar tambem.
 
ToTTenTranz,

A diferença nem é tanta se usares Bloom. Se usares HDR já vai uma boa diferença:

oblivion-highend-bloom.png


oblivion-highend.png


Acho que os outdoors acaba por stressar mais os vertex pipelines. Até podia confirmar isso com a minha 6600GT desactivando 1 ou 2 vertex pipelines, não sei é se dá para fazer isso com o Riva Tuner.

EDIT: Não sei porque não saem as imagens. Aqui ficam os links:

http://anandtech.com/video/showdoc.aspx?i=2746&p=4
http://anandtech.com/video/showdoc.aspx?i=2746&p=5
 
Ora bem aqui vai um artigo que vai refrescar muitas ideias de como será o R600 a nivel de Direct_X10, mas tb de DirectX_9

ATI Promotes Forthcoming DirectX 10 Graphics Chip.
ATI’s R600 to Be the Fastest DirectX 9 Chip in History – ATI


ATI Technologies Thursday disclosed some of the details concerning the code-named R600 graphics processing unit (GPU) and said the part will not only support the new features and unified shader architecture, but will also become the company’s highest performing graphics chip when it comes to wide-spread DirectX 9 applications.

R600 – DirectX 9 Champ

The R600 will be [absolutely] the fastest DirectX 9 chip that we had ever built,” said Richard Huddy, the head of ATI Technologies’ software developers relations department, at a press conference in London, UK.

Mr. Huddy was asked whether the R600’s already well-known unified shader architecture will suit well for the wide-spread DirectX 9 games, which were developed with dedicated pixel and vertex processors in mind. The answer implies that users, who will buy the R600-based graphics cards early, will get top-class performance even in existing applications.

Unified Shader Architecture May Boost Performance

Mr. Huddy said that Xbox 360 game console, which sports developed by ATI Xenos graphics core with unified shader architecture and 48 shader processors, loses 20% to 25% performance in pixel-shader limited games, when its graphics chip is configured as non-unified, e.g.,16 processors work strictly on vertex shaders, whereas 32 are assigned for pixel shaders.

Microsoft Corp. pushes unified shader language for pixel, vertex and geometry shaders in its Windows Vista graphics application programming interface (API) – DirectX 10. As a result of that, graphics hardware designers should deliver their chips with unified shader engines at some point in future in order to more efficiently support the new API. Unified shader processors allow graphics chip to configure itself flexibly for every frame, which provides additional horsepower to the most demanding part of the workload.

Traditional GPUs have dedicated pixel shader and vertex shader processors. In cases, where a frame being rendered contains a lot of geometry and just a few pixel shaders to determine colour, the performance is limited by vertex shaders, while pixel processors are standing idle. GPUs with unified shader processors will have special scheduler inside them, which will determine the load from pixel, vertex and geometry shaders and then assign appropriate number of processors for every kind of work. This will allow to utilize all the computing power of the GPU, believes ATI. Nevertheless, rival Nvidia Corp. believes that implementation of a unified shader architecture should be gradual and cautious.

R600 to Bring More Horsepower

Back in 2002 ATI also unveiled the world’s first DirectX 9-supporting GPU, which offered breakthrough performance in DirectX 8-based games, mainly because it sported 8 pixel processors – two times more compared to rival GeForce 4 Titanium – and 256-bit memory bus – also about two times faster than that of the GeForce4 Ti.

The code-named R600 visual processing unit may repeat the success of its grand-grand predecessor, as, according to sources familiar with the plans of ATI Technologies as well as some media reports, ATI R600 will have 64 unified shader processors – an unprecedented number so far, 16 texture units – inline with today’s GPUs, clock-speed beyond 650MHz and support for high-speed GDDR4 memory controller. Rival Nvidia G80 will have lower amount of shader processors and will also have “hybrid”, not unified, micro-architecture. Neither ATI, nor Nvidia commented on specs of future products.

However, there may be a slight problem for ATI: Microsoft is not going to release DirectX 10 API for Windows XP, but only for Windows Vista, which may mean that ATI’s R600 hardware will be able to show itself in all its glory only when used in conjunction with Microsoft’s new operating system.

http://www.xbitlabs.com/news/video/display/20060525104243.html

Teremos uma repetiçao do que aconteceu com a 9700 ja que parece uma placa avançada para a altura em que vai sair e vai ter que esperar pela next-gen API como a 9700 com o DX9?

Para ja fica também outra certeza. Os jogos DX_9 vao ser muito acelerados mesmo com uma arquitectura virada para o futuro. (ao contrario do que algumas pessoas diziam)
Também vamos ter arquitectura unificada Direct_X 10.

No papel tem tudo para ser uma nova 9700. Vamos ver se atrasos ou outros problemas nao estragam o lançamento.
Ate la também nao faz sentido comprar uma high-end.

EDIT: Atençao que quando tou a dizer que poderá ser a proxima 9700 nao tou a hipotecar a hipotese de a Nvidia contra-atacar com outra grande placa. 9700 no sentido de placa next-gen a sair antes do tempo com gigantes potencialidade.
 
Última edição:
ATI's R600 reaches for 1000MHz clock

A whole gigahertz of graphics


R600 WONT be launched that soon, but as we said before it aims to be the fastest and the biggest chip in the computer graphic industry.

ATI is working hard on R600. It won't finish it until the end of the year but it aims to make the fastest chip to date. We heard that the chip might achieve and even surpass the magical 1000MHz clock.

The fastest that Nvidia and ATI can do with 90 nanometres is around 650 to 700MHz and we know that you won't be able to overclock much more with a 90 nanometre chip. This may mean that R600 will be a 65 nanometre component.

We expect that Nvidia might introduce its G80 a few months before ATI, we expect G80 at late summer or early autumn, while we know that R600 wont see the face of the earth until late Q4. Things can always get delayed, as you know. µ

http://www.theinquirer.net/?article=31968
 
1 ghz? Tragam uma VapoRicky para a mesa 4 sff :D

Até tenho medo do consumo, do enorme aspirador que poderá trazer e do calibre do PCB :(
 
34

ya..em termos de consumos deve ser uma coisa descomunal!!!!!

a Gráfica acutalmente já requer pra si à volta de 35% do power consumido num pc..

com a chegada destas bestas de 400 ou mais milhões de transistors, (equivalente aos transistors de um quad-core amd fx-60) mas podem bem ser 500 milhões..já que a x1900xtx tem 385 milhões..

irá chamar a si perto de 45 - 50 % dos watts consumidos por um pc... o que é obra... em crossfire ( se der) deve passar os 50%...

ora bem... tá na altura de começar a fazer um socket na board..pra podermos meter lá o cooler e tal.. isto de placas add-on... embora seja mais facil de fazer o upgrade deste modo..

que venha ela pra eu comprar um X1800GTO baratinha.
 
Em relação aos clocks, tambem me lembro de ver gente dizer que era impossível a R9700 Pro vir a 325Mhz a 150nm, porque a Ti4600 tinha 2/3 dos transístores da primeira e segundo alguns estava puxada quase ao limite nos 300Mhz. Não só a R9700Pro veio a 325Mhz como tambem veio eventualmente a 9800XT a 412Mhz, com quase o dobro dos transístores da Ti4600 e na mesma nos 150nm. E isto sempre com coolers modestos.

Pessoal convençam-se que quem vê realmente quais são os limites das arquitecturas e dos processos de fabrico são os engenheiros da companhias com as amostras ou planos à frente e os montes de equipamentos para os testar, e não o geek maioral do fórum xpto.

Dizer à partida que o R600 vai precisar de um cooler ultra-grande e barulhento etc etc tem tanta lógica como uma criança de 3 anos fazer uma dissertação sobre como governar um país.
 
Última edição pelo moderador:
UMC to Make 65nm X-Architecture Chips.
Cadence, UMC Validate X-Architecture for 65nm


United Microelectronics Corp., a contract semiconductor foundry, and Cadence Design Systems, have announced the readiness of UMC;s 65nm process technology for X architecture-based chip designs. Theoretically, this means that UMC will be able to produce cost-effective high-speed chips or system-on-chip (SoC) devices.

“UMC has been working with Cadence for several years to bring the advantages of the X architecture to mainstream SoC designers. We are delighted to extend our readiness of this technology to the 65nm generation, as leading customers can now leverage the X Architecture with the industry’s most advanced process technology to increase the competitiveness of their products,” said Patrick T. Lin, chief SoC architect at UMC.

The X Architecture is a new approach to chip design whereby diagonal interconnects are employed to reduce chip costs, increase performance and lower power consumption. Targeted at chips with five or more metal layers, the X Architecture rotates the primary direction of the interconnect in the fourth and fifth metal layers by 45 degrees in relation to conventional orthogonal, or “Manhattan”, architecture. Layers one through three remain unchanged, preserving the design community’s investment in existing cell libraries, memory cells, memory compilers, datapath compilers, and IP hard cores. In addition, the X Architecture allows 45 degrees “wrong-way jogs”, which provides an additional four degrees of freedom in each layer of routing.

Cadence and UMC have qualified 65nm X architecture design rules, the Cadence X architecture, and design support manual (DSM) documents, allowing fabless and IDM companies to leverage the cost and performance benefits of the X Architecture for their leading system-on-chip (SoC) designs. The two companies are engaging with mutual customers towards X Architecture production designs at leading process nodes.

The Cadence X Architecture is now available to customers for UMC's 130nm, 90nm and 65nm process technologies.
http://www.xbitlabs.com/news/other/display/20060528142413.html

Depois da TSCM a UMC junta-se com a capacidade de produçao a 65nm.
Visto a placa ainda estar longe do seu lançamento e estar ja as linhas de produçao em andamento parece cada vez mais real uma R600 a 65nm para poder xegar aos 1GHz no nucleo que o TheInq anunciou.

So resta mesmo esperar que o projecto esteja a decorrer normalmente.

E esta arquitectura em X parece estar mesmo a ganhar mercado. Os beneficios que eles anunciam sao claros e muito bons em relaçao á arquitectura classica.
 
DJ_PAPA disse:
UMC to Make 65nm X-Architecture Chips.
Cadence, UMC Validate X-Architecture for 65nm



http://www.xbitlabs.com/news/other/display/20060528142413.html

Depois da TSCM a UMC junta-se com a capacidade de produçao a 65nm.
Visto a placa ainda estar longe do seu lançamento e estar ja as linhas de produçao em andamento parece cada vez mais real uma R600 a 65nm para poder xegar aos 1GHz no nucleo que o TheInq anunciou.

So resta mesmo esperar que o projecto esteja a decorrer normalmente.

E esta arquitectura em X parece estar mesmo a ganhar mercado. Os beneficios que eles anunciam sao claros e muito bons em relaçao á arquitectura classica.
Não há é uma única palavra no artigo que diga que o R600 vai usar ou não esta "X-architecture" :P
 
Tenho fortes dúvidas que a ATI use os 65nm, e logo a começar num topo-de-gama.
Se algo corre mal, é um problema para reparar, e depois acontecem atrasos.

Eu acredito mais que eles vão usar o "half-node" a 80nm.
 
3

sim...mas que vão sair a 65nm vão..isso não tenho a menor dúvida..

que vão saltar os 80nm

até dezembro... e para a quantidade de transistors que se pensa ter o R600.....80nm poderá nme ser opção..senão os 65nm mesmo,..!!!
 
R600 is ATI's first cable-less Crossfire

Plug in two cards and it works


WE HEAR that the upcoming DirectX 10 compatible R600 won't need a master card to work in Crossfire.

ATI will implement this bridge in its upcoming RV570 and RV560 chips and will enable more than one chip on a PCB configuration.
But it won't be until R600 that this cable-less Crossfire comes into action.

R580+ will be the slight redesign of the existing R580, so don’t expect miracles here. You will still need a master card and a cable to plug them together.

ATI can already put two X1800 GTO cards in Crossfire and it doesn’t need a master card, but the bandwidth between those two cards is not that big. Therefore two slave cards will work and will do the job, we tried it here in the lab and it works.

The master card is the last problem that ATI has to solve to be competitive with Nvidia's SLI. ATI's entry-level Crossfire works without a master, and its mainstream card works without a master. The firm just needs a high-end offering and it's sorted. Getting there slowly but surely. µ

http://www.theinquirer.net/?article=32034
 
Já se esperava, aliás, já tinham prometido isso. É a evolução que faltava para poder fazer concorrência adequada ao SLI, com mais versatilidade.

Quanto à arquitectura, faz sentido que quisessem mudar logo para os 65nm, a julgar pela quantidade de transistores que se fala. Como ainda falta algum tempo até sair a R600 talvez ainda consigam usar este processo, mas a elevada complexidade do chip pode não permitir uma passagem muito suave, tenho algumas reservas. O R600 já deve ser um comboio em andamento, pelo que mudanças de rumo a esta altura do campeonato são arriscadas. O facto de irem fazer agora as R580+ a 80nm indicia que isto possa ser um ensaio (com as devidas vantagens já associadas no imediato) para o R600. Talvez até possam adoptar a X-architecture a este processo (dá para 90nm e 130nm) e ganhar aí uns "pontinhos", sem ir logo para os 65nm.
 
Última edição:
A ATI tinha k fazer mais uma ver. do R580 pk n ia ficar até ao final do ano com a mesma coisa.

Adicionar GDDR4 e fazer um pekeno bump dos clock parece facil e deve dar a performance necessaria até vir a grafica DX10.

O facto da R600 ser dongle free faz com k seja a melhor solução High end, já k é muito mais user friendly e diminui os custo pk as master k n são produzidas em tão grande quantidade nunca são muito baratas, axo k a Nvidia tb os devia seguir, só assim e k este tipo de soluções podem passar a ser mass market. :)
 
ATI’s Next-Generation Graphics Chip Taped Out – Web-Site.
ATI’s R600 Has Been Taped Out, Says Web-Site


ATI Technologies, a leading supplier of graphics processors, has reportedly taped out its code-named R600 graphics processing unit (GPU), which means that the company may release its new chip by the end of the year.

A news-story over VR-Zone web-site claims that ATI’s processor internally called R600 has been made at a fab using 65nm process technology. Usually the “tape out” term means that the chip developer had received working sample of a processor from its foundry partner. Usually, it takes chip companies several months before such chip can be finalized and sold to end-users.

While the news about tape-out of ATI’s next-generation high-end graphics chip is generally good one for ATI, it does not mean that the company will release the chip on time. ATI’s code-named R520 GPU was taped out in late 2004, but it took ATI ten months to make the chip work properly and the firm only released its Radeon X1800 product lineup in October, 2005.

According to sources familiar with the plans of ATI Technologies as well as some media reports, ATI R600 will have 64 unified shader processors – an unprecedented number so far, 16 texture units – inline with today’s GPUs, clock-speed beyond 650MHz and support for high-speed GDDR4 memory controller.

Traditional GPUs have dedicated pixel shader and vertex shader processors. In cases, where a frame being rendered contains a lot of geometry and just a few pixel shaders to determine colour, the performance is limited by vertex shaders, while pixel processors are standing idle. GPUs with unified shader processors will have special scheduler inside them, which will determine the load from pixel, vertex and geometry shaders and then assign appropriate number of processors for every kind of work. This will allow to utilize all the computing power of the GPU, believes ATI. Nevertheless, rival Nvidia Corp. believes that implementation of a unified shader architecture should be gradual and cautious.

http://www.xbitlabs.com/news/video/display/20060612121554.html

Venha ele e depressa :D :D
 
Back
Topo