R520 to have 300 Million Transistors??

90nm has twice the transistors for the same die area than 130nm.

The problem will no doubt be power draw and heat (that most certainly won't be halved for 90nm).

 

I'm hearing 24 pipelines - If I'm to guess (going by what has been said before and relating that to NV43) that would be internal fragment pipelines, but still 16 ROP's.

 

That´s very interesting, so if they go for ~twice the transistor´s of NV40 say around 400million/maybe conservative(?) they still have the same size as a NV40 or even smaller.
The problem as you mention yourself is the power/heat, i guess this is a real challenge for ATI and Nvidia/however at the moment ATI seem´s to be doing extremely well in that area but maybe it will be different at 90nm, who knows?

I was under the impression that it would be a "speed-bump" ala Mhz, and did you mean NV47 not NV48/or both but one is a Ultra and the otherone a GT? Interesting as i never heard this before but i guess i could be reading it at The Inquirer tomorrow anyway..

 

Based on Orton's comments to Dave, it seems that ATI is leaning (note "leaning", not "convinced") towards believing that NV40 die size is an aberration on the high side. I suspect that unless they get convincing evidence of it well before introduction (i.e. in time to do something about it, which may already be passed anyway) that it will take at least another generation of chips to convince ATI that NV has permenantly raised the bar on die size for the flagship chip.

 

Or, of course, they could be lieing their ass off, like they did about R420 pipelines in the run up to introduction. Lot of liar's poker in this industry.

 

I'm guessing NV47 will be the high end part after NV48 (up against R520?), continuing NVIDIA's wacky codenumbering?

 

Doesn't look like an overclocked NV40, not for NV47 at least

I'm guessing that NV47 will be NV's next high-end part and it will contain 24 pixel pipes and 8 vertex pipes. Don't know anything more about it, but 16 ROPs are a possibility, yes, since 24 pipes with todays GDDR3 would still be heavily b/w limited.

NV48 might be the next hi-end part after NV47 designed to compete with R520. I'm thinking about 32 pixel pipes, but that's pure speculation on my part.

 

I´m guess i´m still to "traped" in the NV10-15-20-25-30-35-40 that follows some kind of "rule"...
So we will see more fragment power then, can they do this by adding a third shaderALU so the pipes has 3maths/shader´s and one texture unit compared to the NV40 that have if i understand correct 2shader´s/one texture unit in the pipe?

I forget whit my first reply that you also will have SOI that will help i think.

 

nv40 doesnt have texture units. they have 2 ps units 1 of which is used for textures

 

Hmmm would you elaborate this. :?

 

Have a look at Wavey's preview of the chip:
http://www.beyond3d.com/previews/nvidia/nv40/index.php?p=10

 

NV40 does have separate texture units along the two shader units. It's just that the texture coordinates have to be passed through the first shader unit. So when sampling from a texture, you can only fully utilize the first SU if you do a MUL on the texcoords. Otherwise those MULs are wasted.

 

Its best to look at the previous page, which illustrates that there is a texture sampler but, as XMas said, the first shader unit is responcible for the texture address processing.

 

I think, that this way NV47 would need specific optimizations to work effectively, while simply adding more NV40-like pipelines will save all devrel and driver teams work already done for NV4x. And silicon price may be very comparable.

 

So i take it that Shader 4.0 and Longhorn will have alot to do with the R620/NV60? :?

 

Mmm, this talk of 24 and 32 pipelines designs.. and I was just getting used to the idea of 16 pipeline cards being high-end. I guess we won't be seeing a lasting high-end reign similar to the 8 and 4 pipeline generations with the 16 pipeline generation cards, will we? The high-end pretty much just sneezed a 12-pipeline card out and went straight to 16.

Now we're already talking 24 and 32? How exciting...

P.S. The beginning of this thread is funny. There's always a notion of dislike towards TheInquirer in every post regarding one of their articles, yet just about everyone of their articles are discussed here. If it wasn't for these forums, I'd probably never even go their site. So just admit it, TheInquirer is on many of your start-up homepages and you love them.

 

I'd love them a whole lot more if and when they learn the meaning of "attribution" and the role it plays in journalism. Of course, that would probably eliminate 90% of their articles, which is probably why they don't use attribution in the first place... Nobody who's anybody would be caught dead being quoted making the kinds of statements they routinely cite as coming from their mysterious, unidentified "sources." Heh... But the Inq is not alone, and to their credit they don't try to hide it and it's obvious they are often deliberately trying to provoke a laugh--the Lighter Side of the Industry, etc.--which does have its place.

I once accused Anand Le Shrimp (of AnandTech) of his own mysterious article "source" being in reality a wad of crumpled-up paper he found while digging through a dirty trash can in the women's rest room at a nearby Chapel Hill (N.C.) Shell station, but I could never get him to admit it to it... It was fairly obvious from the article content, though, that my guess as to the identity of his source was near the mark.

 

Dave were you joking? About the 24pipe bit. Jeez they are going fast now. I mean 8->16 is a bigger jump than 4->8 and 16->24 is just as big of a jump again. Of course when are these supposed to come out about, if it is in spring '06 I don't suppose it is a big splash, but if it is next spring heheh I will be broke.

 

Depends how you define bigger. As a percentage 24 is only 50% more than 16.