PDA

View Full Version : PS 3.0 support in NV35, 36??


Pages : 1 [2]

simwiz2
09-02-03, 12:40 PM
LOL didn't the Inquirer also say that Prescott would be released at 0.13u?

Then the next day: "never mind..."

Zenikase
09-02-03, 04:03 PM
I feel so lost...what does ILDP mean? :(

Spotch
09-04-03, 12:34 AM
Originally posted by lordtoran
LOL!

Two words: Deja Vu!

'Full support, but some hardware problems'

Don't take that information for full. I just found it on myself and ROFL'ed with my friend for five minutes here.

Typical that this hypocrycy still comes from nVidia when they are already lying aflat.

:D

Yeah hardware problems... like "we forgot to put the hardware into the chip"...

:D

digitalwanderer
09-04-03, 12:38 AM
Originally posted by Spotch
Yeah hardware problems... like "we forgot to put the hardware into the chip"...

:D
Well, you must admit that would be a problem... ;)

Uttar
09-05-03, 03:40 PM
I talked with Fudo about this a while ago ( got him on my MSN list - hey, industry contacts are never a bad thing ;) )

The story is nothing new, far from it. Actually, it might be, but it isn't anything INTERESTING that's new then, lol. It wouldn't even be a silicon change, more of a driver test by lying to DX and say you got things you don't.


Uttar

Hanners
09-06-03, 11:31 AM
Originally posted by Uttar
It wouldn't even be a silicon change, more of a driver test by lying to DX and say you got things you don't.

Sounds right up nVidia's street. ;)

digitalwanderer
09-06-03, 11:54 AM
Originally posted by Hanners
Sounds right up nVidia's street. ;)
Fanboy! :mad:

























j/k! ;)

aapo
09-06-03, 02:38 PM
Originally posted by Zenikase
I feel so lost...what does ILDP mean? :(

It's "Instruction Level Distributed Processing". In the processor front a comparable technique would be Intel's HyperThreading, but this is different of course. The idea is to forget the pipeline architecture and have a bunch of small different calculation units instead. Then one can do many things simultaneously provided that the things done are different (eg. multiply and add different registers at the same time). The Good Thing (tm) would be the possibilty to make eg. more multiplying and addition units than division units, because multiplication and addition are (in average case) used more than division.

So there would be a kind of "floating point number bus" and a lot of "calculation devices" instead of a simple pipeline.

There's a thread in Beyond3d.com (http://www.beyond3d.com/forum/viewtopic.php?t=6303) about the technique.

Shadowx
09-12-03, 04:09 PM
for god's sake ps2 is sh*t why even bother with ps3 in the nv3x family of hardwear!!

ragejg
09-12-03, 04:24 PM
There's rules re: swearing and the swear filter. Read them please.

:-l

Descombobulator
09-12-03, 05:23 PM
Originally posted by digitalwanderer
Well, you must admit that would be a problem... ;)

No problem, fixed in the 6x drivers.

Zenikase
09-12-03, 09:27 PM
Originally posted by aapo
It's "Instruction Level Distributed Processing". In the processor front a comparable technique would be Intel's HyperThreading, but this is different of course. The idea is to forget the pipeline architecture and have a bunch of small different calculation units instead. Then one can do many things simultaneously provided that the things done are different (eg. multiply and add different registers at the same time). The Good Thing (tm) would be the possibilty to make eg. more multiplying and addition units than division units, because multiplication and addition are (in average case) used more than division.

So there would be a kind of "floating point number bus" and a lot of "calculation devices" instead of a simple pipeline.

There's a thread in Beyond3d.com (http://www.beyond3d.com/forum/viewtopic.php?t=6303) about the technique.

So is this like multithreading? 3Dlabs already did this with the advanced command processor and granular architecture in the P10 chip.

aapo
09-14-03, 12:01 PM
Originally posted by Zenikase
So is this like multithreading? 3Dlabs already did this with the advanced command processor and granular architecture in the P10 chip.

I'm not sure about my terminology, but for me multithreading means an operating system concept and it can be done with asingle processor and a scheduler in the OS. Actually, been there done that: I took a course in my university where we had to program an OS scheduler for a simple OS that didn't have a proper scheduler. It was a small OS for teaching purposes, but it still was a hell of a lot of c-code to make it work. Kudos to Linux developers, I learnt to really appreciate that behemoth of the operating system kernels. :D

Anyway, ILDP is more like Intel's hyperthreading: the parts of the (C/G)PU that aren't currently doing anything in the main processing flow can be used to do calculations with a secondary processing flow (the virtual dual processor). But because GPUs already calculate in truly parallel fashion (e.g. 8 pipelines), this is somewhat different.

You could say with ILDP technology you have a changing amount of traditional pipelines depending on the program you are executing. Of course, there would be only one pipeline but it could distribute the operations to different units to be calculated at the same time. So, there could be work done an amount comparing to e.g. 4 - 24 traditional pipelines (numbers are artificial examples) with the single master pipeline. The problem is there has to be some kind of an intelligent bus to move the data into right places, instead of simple traditional 'straight' pipelines.

Actually, it was revealed by Dragunov that NV3X is a design close to that model: it has only one pipeline, that can produce 4 - 8 pixels/texels at a time by distributing them with the calculational units. It's not a true ILDP, though, but it's closer to the ILDP model than the Radeons, which are more traditional. It seems that the NV3X design is indeed a very complicated one.

emacs
09-16-03, 09:13 PM
i thought ps 3.x would only be fully mature with the release of DirectX 10. but, then again, Longhorn is supposed to be out in 2005 which, according to rumours, will require hardware-acceleration via DirectX 9 at the very least. regardless, i don't think we'll see ps 3.x games until q3 2004 anyway ...

StealthHawk
09-16-03, 11:04 PM
Originally posted by emacs
i thought ps 3.x would only be fully mature with the release of DirectX 10. but, then again, Longhorn is supposed to be out in 2005 which, according to rumours, will require hardware-acceleration via DirectX 9 at the very least. regardless, i don't think we'll see ps 3.x games until q3 2004 anyway ...

PS3.0 and VS3.0 are part of DX9.