HOLY COW! R420 12x1 architecture?!?!

[Lonyo]

In a completely unrelated note, coming from another source, the R420 is a 12x1. The NV40 is still a 8 pipelines design AFAIK.

From Notforidiots

They say the reliability is "very high"

Also:

According to Hellbinder on the nV News forums ( paraphrasing ) :
The R420 chip ( *not* board, this is not multichip ) is made out of RV350 ( or RV360, but they're the same thing really ) elements thanks to physical reuse, giving the ( theorical, at least ) power of 4 RV350s ( at Radeon 9600 Pro speeds ).
Such a move makes perfect sense considering the RV350 is currently ATI's only 0.13u chip and that they recently announced they'll use physical reuse for future products.

NVNews news on it
And from NVnews forum about it

 

[rbV5]

Hellbinder is suprisingly accurate in my experience.

 

[Lonyo]

If anyone can do it, ATi can. No offense to nVidia, but there were people (3D Labs I believe) who said that you couldn't get a DX9 part on 0.15micron technology, Matrox and nVidia never did it, ATi managed it, despite people in the field reckoning it wasn't possible.

 

[Schadenfroh]

$

 

[azntiger0586]

Originally posted by: Schadenfroh
$

might i add a couple more $$

 

[rbV5]

I don't think the r420 is targeted as a "value" solution.

 

[Corporate Thug]

**drool**

 

[chsh1ca]

Dang, given the recent trend in pricing, I can only assume this will be about $699USD (meaning about triple the cost of my last upgrade in canadian dollars), so it looks like I wouldn't be able to afford this for about a year or so, barring a lottery win.

 

[Corporate Thug]

wow, $699? i was thing more along the lines of $450

 

[Mingon]

who said that you couldn't get a DX9 part on 0.15micron technology, Matrox and nVidia never did it, ATi managed it,

Really what about nv34

 

[zsouthboy]

Originally posted by: Lonyo
If anyone can do it, ATi can. No offense to nVidia, but there were people (3D Labs I believe) who said that you couldn't get a DX9 part on 0.15micron technology, Matrox and nVidia never did it, ATi managed it, despite people in the field reckoning it wasn't possible.

Uhh.... dude... fx 5600? fx 5200???

 

[stardust]

fx5200 only is running 0.15 (enhanced) lol
12x1 would generate 50% faster fillrates...wow making my 9800 feeling old...

 

[Pete]

$700--are you kidding? Virtually all recent top-end cards have debuted at $400 or $500. I don't see how this would be different.

And 12x1 sounds nice.

 

[zsouthboy]

Originally posted by: stardust
fx5200 only is running 0.15 (enhanced) lol
12x1 would generate 50% faster fillrates...wow making my 9800 feeling old...

5200 is .15?

 

[Ferocious]

Wow.

Sorta reminds of when the V2 came out and blew away the V1 and the competition.

 

[videoclone]

In the rumored spec's of the NV40 and R420 cards they plan on using 16x1 .. pipelines .. or 8x2

 

[zsouthboy]

Keep in mind the nv40 is supposedly a completely different architecture, too.

Also keep in mind, that if you think nv40 and r420 will rock, wait till you find out more about r500 and nv50... they WILL blow you away

 

[Schadenfroh]

Originally posted by: zsouthboy

Originally posted by: stardust
fx5200 only is running 0.15 (enhanced) lol
12x1 would generate 50% faster fillrates...wow making my 9800 feeling old...

5200 is .15?

yup

 

[BFG10K]

12x1 is a very impressive configuration and will improve not only pixel fillrate but also texel fillrate as well, resulting in a win-win situation. If this is true then I am officially excited.

 

[Lonyo]

Just as 3DLabs mentioned to us during our P10 briefing, in order to make the 3D pipeline entirely floating-point you need to be on at least a 0.13-micron process which won't be mature enough (at TSMC at least) until this fall to use for a mass production GPU.

At the same time they work on integer data and not 32-bit floating point values which is required for DX9 compliance. The reason the Parhelia cannot claim these two key features is because of, once again, a lack of die-space

If you think what the 5200 has in its core, you see that there's not really all that much (ie: not all that high transistor count)
And the NV31 (FX5600) is apparently "The NV31 is a 0.13-micron GPU"., plus the FX5600 only has 75m transistors, so it's not very "feature packed"
"47 million transistors" for the FX 5200, and it is rather slow.
The reason 3D Labs thought it could not be done was because of lack of space on the dies, which isn't a problem at the lower end of the market because there is only a limited feature set that's on the cards anyway, so they do actually have a fair amount of space to be able to incorporate the extra transistors needed for a fully FP pipeline, and since they only have something like 2 or 4 pipes, that lowers transistor count for these features as well.

isn't economically viable for 3DLabs with the P10 when you take into account the increase in gates over the present mixed fp/integer setup

Now, with a chip that has 80m transistors, the complexity and difficulty in making it fully FP is quite high. 3D Labs seem to be suggesting higher end parts would have difficulty, and considering there are 8 pipes on the 9700/9500/9800 cards, and they are fully FP with 0.15, this is quite impressive.

the very large and very expensive R300

Suppporting 3D Labs point about it not necessarily being economic, and Matrox's issue of lack of die space.
This is also very different to the FX5200 which is a simple chip, thus has a low transistor count and isn't therefore large. At 110 million transistors for the R300, it's more than twice the number of the 5200's core.

BTW: I have has 2 TNT2's and a GF4 (Ti ofc), those are my only 3 graphics cards, don't call me a fanboy

ATi managed some fairly impressive things before, and I'm guessing may manage good things in the future.

 

[Viper96720]

R300 was out long before NV30 before that is when some were saying couldn't have a .15u DX9 chip. I think it's funny the info comes from a site called nvnews. Nothing against the site or anything.

 

[Regs]

If these new cards are suppose to do what they say they do then I will have no problem shelling out 400-450 dollars. Anything over 500 dollars though is just a tad ridiculous. Some might consider 300 dollars to be a stretch. If these new cards produce 50-100% more out put then their predecessors and turn into the flagship cards for games like D3/D4... then how can you resist paying for such a card? It is going to be a evil marketing campaign. :evil:

And why am I getting the feeling John Carmack knows something that we are not suppose to know yet for a few months....

 

[Schadenfroh]

man, and i thought my maximum of $150 was a stretch

 

[chsh1ca]

Originally posted by: Regs
And why am I getting the feeling John Carmack knows something that we are not suppose to know yet for a few months....

Probably because he just seems like the kind of guy who could have played 'Cancer man' from the X-Files. The guy who is quietly shaping the [gaming] world behind the scenes and knows a lot more than he will ever let on.

Kinda conspiratorial if you ask me. And creepy.

 

[Genx87]

If you think what the 5200 has in its core, you see that there's not really all that much (ie: not all that high transistor count)
And the NV31 (FX5600) is apparently "The NV31 is a 0.13-micron GPU"., plus the FX5600 only has 75m transistors, so it's not very "feature packed"
"47 million transistors" for the FX 5200, and it is rather slow.
The reason 3D Labs thought it could not be done was because of lack of space on the dies, which isn't a problem at the lower end of the market because there is only a limited feature set that's on the cards anyway, so they do actually have a fair amount of space to be able to incorporate the extra transistors needed for a fully FP pipeline, and since they only have something like 2 or 4 pipes, that lowers transistor count for these features as well.

You are defending a defenseless position. Whoever said a DX9 card couldnt be done on .15 is ignorant as hell. How many transistors are in the GF4???? Do you consider that chip to be featureless? Is the 5200 a barn burner now. But is is DX9 compatible? Yes! If you think 47 Million is a low number what do you think about the Athlon XP then?!?!?!?!?!?!?!? Oh that weighs in at 37 million transistors with its 384K of cache.

Now, with a chip that has 80m transistors, the complexity and difficulty in making it fully FP is quite high. 3D Labs seem to be suggesting higher end parts would have difficulty, and considering there are 8 pipes on the 9700/9500/9800 cards, and they are fully FP with 0.15, this is quite impressive.

I dont see where the process issue has anything to with DX9 card technology. I wouldnt really put too much faith into 3dlabs anyways. Do they even have a FP compliant card out yet?????

Kind of like having the have nots say something cant be done. Duh!

Suppporting 3D Labs point about it not necessarily being economic, and Matrox's issue of lack of die space.
This is also very different to the FX5200 which is a simple chip, thus has a low transistor count and isn't therefore large. At 110 million transistors for the R300, it's more than twice the number of the 5200's core.


Whoopie do.................At 110 million transistors it is nearly 4 times larger than the Athlon.