G70 poderá ter pipelines escondidos...
- Autor do tópico blastarr
- Data Início
Horus-Anhur
I folded Myself
Até pode ser, mas então a saída de uma 7800 com 32 pipes dependerá mais da TSMC conseguir melhores yelds que propriamente da vontade da nVidia.
A questão é se a 7800gtx tem esses pipes desactivados, se é possivél activa-los e qual a percentagem de 7800gtx com os pipes em boas condições.
Mas também pode ser q a nVidia tenha mandado para o mercado apenas os chips com 24 pipes bons e guardado os que tem os 32 para acumular-los para ter mais chips para um futuro lançamento.
A questão é se a 7800gtx tem esses pipes desactivados, se é possivél activa-los e qual a percentagem de 7800gtx com os pipes em boas condições.
Mas também pode ser q a nVidia tenha mandado para o mercado apenas os chips com 24 pipes bons e guardado os que tem os 32 para acumular-los para ter mais chips para um futuro lançamento.
Ansatsu
Power Member
Clues indicate that G70/RSX might be even more powerful than originally thought
Piecing together several disparate clues, all indications seem to be that NVidia's G70 chip is packing a surprise. R520 insurance built in...
By Carl Bender - June 26, 2005
We've been following the RSX chip and it's brother on the PC side - NVidia's G70 - ever since the name of the chip was revealed at this years E3. With the official launch of the GeForce 7800 GTX earlier last week and the confirmation that G70 was indeed the base architecture for Sony's RSX chip, it seemed everything could be taken at face value; RSX would be a higher clocked G70 on a smaller process with some logic adjustments to facilitate it's interactions with Cell. Well, that may still be the case, but our understanding of what that base G70 architecture actually is... is beginning to change.
Talk is building that G70 may in fact be a 32-pipeline part, with 8 of those pipelines presently deactivated.
The first clue that all may not be as it seems with G70 is the observation that early die-size estimates place it at 334mm^2 - a gigantic chip - and a gigantic chip that should seemingly be able to hold more than the ~300 million transistors claimed by NVidia. Could it be that the G70 actually has more transistors than stated by NVidia, that those 300 million transistors only indicate active transistors on the die?
The second clue that there's more to G70 than meets the eye is recently leaked information on the Internet that NVidia's Forceware 75.90 drivers show five different parts to be based on the G70. When examining their more recently released Forceware 77.72 drivers however, we find that the GeForce 7800 GTX actually occupies the second G70 slot indicated by the 75.90 drivers, rather than the first. Why is this significant? Because the initial card listed in the architecture 'line-up' is almost always the top-performance flagship card, with subsequent cards being variations on the first. For example, the GeForce 5900 Ultra was 0330.1 in the drivers; the GeForce 5900 vanilla was 0331.1. The GeForce 6600GT was 0140.1; the basic 6600 was 0141.1. The GeForce 6800 Ultra was 0040.1, and the GeForce 6800 was 0041.1. The GeForce 7800 GTX carries the code 0091.1 - so where's 0090.1?
Additional common-sense clues that the GeForce 7800 GTX will not represent the top-of-the-line offering from NVidia this generation are apparent as well. It's great that it is a single slot, relatively low power card, but in the high-performance market the card targets, there would be no reason to hold back on a higher clocked, dual-slot graphics card. Especially considering how well the card has overclocked so far. A future card meant to combat ATI's upcoming R520, which is rumored to be a 32-pipeline part, would offer the perfect opportunity to launch the 'full' G70 chip, with additional pipelines, higher clock speed, and a dual-slot cooling system.
In the meantime, what launching with a crippled G70 offers NVidia is the opportunity to set a higher price point allowed to the #1 graphics card, and significantly reduce yield-issues by needing only to focus on getting chips out the door with 24 of the 32 pipelines functional. This scenario would also explain why this is one of the few chip/card combinations to be available immediately at launch in recent memory. With the R520 currently experiencing manufacturing troubles and expected to launch late this summer, this gives NVidia the chance to exploit the $600 price-point with GTX until R520 launches, and then counter with a 'full' G70 card to maintain that price point, and hopefully the performance lead.
Since RSX will be based on the G70, and Kutaragi has recently indicated that like Cell and for fabbing reasons RSX is likely to have a portion of 'redundant' parts deactivated, it stands to reason that RSX might in fact be a 32 pixel pipeline part, scaled down to 28 pipelines. This would reconcile statements made by both Sony and NVidia that the RSX will be more powerful than the GTX, but that NVidia will have a higher-performance PC part shipping when PS3 launches. For now, we'll call that part the 7800 GTX Ultra.
Piecing together several disparate clues, all indications seem to be that NVidia's G70 chip is packing a surprise. R520 insurance built in...
By Carl Bender - June 26, 2005
We've been following the RSX chip and it's brother on the PC side - NVidia's G70 - ever since the name of the chip was revealed at this years E3. With the official launch of the GeForce 7800 GTX earlier last week and the confirmation that G70 was indeed the base architecture for Sony's RSX chip, it seemed everything could be taken at face value; RSX would be a higher clocked G70 on a smaller process with some logic adjustments to facilitate it's interactions with Cell. Well, that may still be the case, but our understanding of what that base G70 architecture actually is... is beginning to change.
Talk is building that G70 may in fact be a 32-pipeline part, with 8 of those pipelines presently deactivated.
The first clue that all may not be as it seems with G70 is the observation that early die-size estimates place it at 334mm^2 - a gigantic chip - and a gigantic chip that should seemingly be able to hold more than the ~300 million transistors claimed by NVidia. Could it be that the G70 actually has more transistors than stated by NVidia, that those 300 million transistors only indicate active transistors on the die?
The second clue that there's more to G70 than meets the eye is recently leaked information on the Internet that NVidia's Forceware 75.90 drivers show five different parts to be based on the G70. When examining their more recently released Forceware 77.72 drivers however, we find that the GeForce 7800 GTX actually occupies the second G70 slot indicated by the 75.90 drivers, rather than the first. Why is this significant? Because the initial card listed in the architecture 'line-up' is almost always the top-performance flagship card, with subsequent cards being variations on the first. For example, the GeForce 5900 Ultra was 0330.1 in the drivers; the GeForce 5900 vanilla was 0331.1. The GeForce 6600GT was 0140.1; the basic 6600 was 0141.1. The GeForce 6800 Ultra was 0040.1, and the GeForce 6800 was 0041.1. The GeForce 7800 GTX carries the code 0091.1 - so where's 0090.1?
Additional common-sense clues that the GeForce 7800 GTX will not represent the top-of-the-line offering from NVidia this generation are apparent as well. It's great that it is a single slot, relatively low power card, but in the high-performance market the card targets, there would be no reason to hold back on a higher clocked, dual-slot graphics card. Especially considering how well the card has overclocked so far. A future card meant to combat ATI's upcoming R520, which is rumored to be a 32-pipeline part, would offer the perfect opportunity to launch the 'full' G70 chip, with additional pipelines, higher clock speed, and a dual-slot cooling system.
In the meantime, what launching with a crippled G70 offers NVidia is the opportunity to set a higher price point allowed to the #1 graphics card, and significantly reduce yield-issues by needing only to focus on getting chips out the door with 24 of the 32 pipelines functional. This scenario would also explain why this is one of the few chip/card combinations to be available immediately at launch in recent memory. With the R520 currently experiencing manufacturing troubles and expected to launch late this summer, this gives NVidia the chance to exploit the $600 price-point with GTX until R520 launches, and then counter with a 'full' G70 card to maintain that price point, and hopefully the performance lead.
Since RSX will be based on the G70, and Kutaragi has recently indicated that like Cell and for fabbing reasons RSX is likely to have a portion of 'redundant' parts deactivated, it stands to reason that RSX might in fact be a 32 pixel pipeline part, scaled down to 28 pipelines. This would reconcile statements made by both Sony and NVidia that the RSX will be more powerful than the GTX, but that NVidia will have a higher-performance PC part shipping when PS3 launches. For now, we'll call that part the 7800 GTX Ultra.
blastarr
Power Member
Tal como eu suspeitava.
Por outro lado isto levanta uma questão porventura mais interessante:
Se esta G70 fabricada em 110nm já tem os 32 pipelines para fazer uma Ultra, que chip mistério é o G80, fabricado em 90nm e agendado para o último trimestre de 2005 / primeiro trimestre de 2006 ?
Certamente não é uma versão da G70 para as gamas inferiores, pois estas devem estar no mercado muito antes, possivelmente aínda antes do fim do Verão...
Nova geração Geforce 8 ? Responderia à questão que a Nvidia colocou quando afirmou que, por altura do lançamento da PS3 com o chip RSX, já existiria um chip para pc mais potente.
É que não acredito que uma versão da G7800 com 32 pipelines fosse mais rápida do que a RSX, costumizada, com acesso directo à memória sem pci-express nem DirectX HAL pelo meio, etc, etc. Não faria sentido.
Por outro lado isto levanta uma questão porventura mais interessante:
Se esta G70 fabricada em 110nm já tem os 32 pipelines para fazer uma Ultra, que chip mistério é o G80, fabricado em 90nm e agendado para o último trimestre de 2005 / primeiro trimestre de 2006 ?
Certamente não é uma versão da G70 para as gamas inferiores, pois estas devem estar no mercado muito antes, possivelmente aínda antes do fim do Verão...
Nova geração Geforce 8 ? Responderia à questão que a Nvidia colocou quando afirmou que, por altura do lançamento da PS3 com o chip RSX, já existiria um chip para pc mais potente.
É que não acredito que uma versão da G7800 com 32 pipelines fosse mais rápida do que a RSX, costumizada, com acesso directo à memória sem pci-express nem DirectX HAL pelo meio, etc, etc. Não faria sentido.
Última edição:
blastarr
Power Member
Crusher disse:
A tendência actual é para desfasar o nº de pixel pipelines do nºs de ROP's (a 6600 GT tem 8;3).
É por isto que a 6600 GT, mesmo com menos largura de banda da memória em relação a uma 9800 pro (128 vs 256 bit), consegue batê-la nos jogos que usem e abusem de pixel shaders, ao invés de usarem o método tradicional de simples texturização e filtragem.
Na esmagadora maioria dos casos actuais, ter 24 pixel pipes é mais importante do que ter uma relação 1:1 com o nº de ROP's tradicionais.
Quem sabe, talvez a próxima versão do Rivatuner traga novidades ao nível do unlocking de quads extra...
Sim, Madril, referi a HAL, pois é a própria MS a reconhecer que o modelo do DX actual pode consumir até 30% de tempo de processamento, criando grandes latências.
O ganho de velocidade dos WGF, especialmente a 2.0, vai ser por causa da reformulação e optimização dessa layer.
Horus-Anhur
I folded Myself
A G80 pode ser uma revisão do G70. Assim teriamos uma Geforce 7900.
Esta politica de desactivar partes de um chip está a tornar-se moda. A Sony está a fazer algo parecido com o CELL. Desactivaram um dos seus 8 SPEs para melhorarem os yelds.
Esta politica de desactivar partes de um chip está a tornar-se moda. A Sony está a fazer algo parecido com o CELL. Desactivaram um dos seus 8 SPEs para melhorarem os yelds.
