7800 GTX + Rivatuner = ???

[Commander Suzdal]

I've been wondering something since this morning:

If nVidia is holding back a 32-pipe Ultra to squash the R520 (big IF, I know, could just be a speed bump)...

Wouldn't it make sense that the current core has 32 pipes, but they went with 24 for the GTX because of yield issues (otherwise how could they have a 32-pipe part ready by this fall? No time to tape out a whole new core)...

Then wouldn't those extra pipes be unlockable Rivatuner style?

Can you imagine what a GTX could do, overclocked and unlocked to 32 pipes?

/end techno fantasy camp

 

[FadedSpark]

doubt it, they would be hard wired closed even if they were there me thinks

oddly enough i was thinking about rivatuner and softmodding earlier

 

[TehQuick]

wait and see, who knows what unwinder will come up with

 

[Brent_Justice]

if they did have a 32 pipe G70 without increasing the ROPs as well it'd be unbalanced, but it sure could fill pixels fast

i don't know how that'd scale in real world gaming performance

 

[^eMpTy^]

if they did have a 32 pipe G70 without increasing the ROPs as well it'd be unbalanced, but it sure could fill pixels fast

i don't know how that'd scale in real world gaming performance

The GTX has 16 ROPs right? The 8-pipe 6600GT does just fine with 4 ROPs...so I doubt the 7800 would suffer with the same ratio...

Then again I seriously doubt there are locked pipes...give the transistor count vs the 6800s it just doesn't seem feasible...but I could be totally wrong...

 

[Brent_Justice]

The GTX has 16 ROPs right? The 8-pipe 6600GT does just fine with 4 ROPs...so I doubt the 7800 would suffer with the same ratio...

Then again I seriously doubt there are locked pipes...give the transistor count vs the 6800s it just doesn't seem feasible...but I could be totally wrong...

well like i said i don't know how that would scale

i said it'd be unbalanced, which it is

but your right the 4 ROPs haven't held back the 6600, but, we are talking about many more comparing 16 ROPs to 32 pixels per clock, maybe it would with that many, maybe it wouldn't, i don't know, i'm just saying it'd be unbalanced

i think they should keep the current ratio with the 7800 GTX the same and up the ROPs to 24 if they are going 32 pixels, that's just my opinion

but really, can you imagine a chip any bigger lol, 302 million transistors is a lot already

 

[Commander Suzdal]

Truly! They'd probably have to do a process shrink to 90nm before they could add anything more.

I hadn't thought of the ROPs when I did my daydreaming. How many ROPs are there on the 6800 GT/Ultra by the way?

Yet another random thought--will GPUs ever transition to using the equivalent of a L2 cache like CPUs do? Or do they already do that? Or is it impractical for a high-parallelism architecture like a GPU? If it's feasible, I wonder what kind of optimizing options it would give card makers/game developers? For example, running the driver/compiler in GPU-local on-die cache instead of system memory?

Well, I may have just bared my massive ignorance several times in the above, but I await enlightenment!

 

[Brent_Justice]

Truly! They'd probably have to do a process shrink to 90nm before they could add anything more.

I hadn't thought of the ROPs when I did my daydreaming. How many ROPs are there on the 6800 GT/Ultra by the way?

Yet another random thought--will GPUs ever transition to using the equivalent of a L2 cache like CPUs do? Or do they already do that? Or is it impractical for a high-parallelism architecture like a GPU? If it's feasible, I wonder what kind of optimizing options it would give card makers/game developers? For example, running the driver/compiler in GPU-local on-die cache instead of system memory?

Well, I may have just bared my massive ignorance several times in the above, but I await enlightenment!

the NV40 has 16 ROPs and they already use an L2 Texture Cache that each pixel pipeline has access to in each Quad

 

[thunderblade]

interesting