nv skipping 8900's for 65nm parts?

[withmyshadow]

I read some articles on it, which makes a lot of sense.
8800 ultra was not on the original 2007 map, IIRC
and 8900s @ 80nm dont' seem to be fast enough
if you google "tsmc 65nm nvidia", you get articles on their mobile chips. will we see 65nm parts before fall?

[retsam]

8900 is just a name....if they canceled parts then thats a whole nother story..i have yet to hear of any cancelations...

[withmyshadow]

8900 is just a name...
65nm parts should be much faster than 90nm 8800 parts so it should be a new line, i.e. 9x00. imho

[retsam]

65nm parts should be much faster than 90nm 8800 parts so it should be a new line, i.e. 9x00. imho
it could just be a die shrink, usually when nvidia goes from a x000 to another x000 product they is usally a major revision ion hardware design.

[Madpistol]

So why the heck did they rename the 6xxx series to 7xxx series??? There was one new feature on the 7 series that wasn't in the 6. (If you run Mad Mod Mike on a 6600/6800, some of the renderings don't show up properly.)

[rhink]

So why the heck did they rename the 6xxx series to 7xxx series??? There was one new feature on the 7 series that wasn't in the 6. (If you run Mad Mod Mike on a 6600/6800, some of the renderings don't show up properly.)

7x00's I think were a little more than a die shrink, it was more a refined 6800, with a few tweaks (like transparency AA). Biggest difference between the two was still just speed though.

Wouldn't be surprised if we start out with a 9000 series this fall that's just a refined 8800, we'll probably be on that basic architecture for a couple more "generations". If you remember the 5000,6000,7000 series were all really the same "family" (though at least the 6000 series added a slew of new features over the 5000's, they still built directly off of each other).

[DMx]

Don't the 79XX series have a smaller die than the 78XX? If so, then the 89XX could be a smaller die.

[rhink]

7800->7900 was a die shrink from 110 nm to 90 nm. G70 (7800) was 334 mm^2, G71 was 196 mm^2.

[withmyshadow]

I can believe this. I just want to know the time frame.
http://www.vr-zone.com/?i=4682
NVIDIA is using back the GTS naming for their mainstream cards and is preparing to launch their GeForce 8600 series codenamed G84 in April. There will be 2 models; GeForce 8600 GTS and GeForce 8600GT and they are priced at $249-$299 and $199-$249 respectively. Interestingly, there is no G81 or GeForce 8900 series we heard. There is something more powerful brewing over at NVIDIA.

[911medic]

I read some articles on it, which makes a lot of sense.
8800 ultra was not on the original 2007 map, IIRC
and 8900s @ 80nm dont' seem to be fast enough
if you google "tsmc 65nm nvidia", you get articles on their mobile chips. will we see 65nm parts before fall?I think the pattern has been lately to do the die shrink on the midrange/low end/mobile cards first, then the following generation they offer the top-of-the line cards with the smaller die size.

Didn't somebody get burned trying to do their initial die shrink on their "creme de la creme" cards? I vaguely remember this caused poor yields moving to the smaller process for their most complicated products right off the bat. I think it resulted in vaporware or very low quantities of the cards at the time. Since then it's been "die shrink the low end first," IIRC.

[rhink]

If I remember right the Geforce 3 was much delayed due to it being designed for a certain process size, and very difficult to fabricate on anything larger. But the TSMC process wasn't ready, so the card was delayed and nvidia lost quite a bit of money. Eventually it did come around and eventually it was a good card, but it could have come out much sooner and been very dominant. That's about the time they signed a deal with IBM to use some of IBM's fabs to help out for limited quantities of high speed products (though I don't think IBM fabs any of their stuff anymore). Ever since then they've always introduced their new high end products on a proven process.

Other companies have been introducing their new processes in low to mid range products as well- for one, those products are high volume, so saving the die area on those products saves a lot more money. For another, it's easier to learn the process and work the kinks out with a lower complexity product then move to the high complexity ones once you're reasonably confident it works well. AMD did this moving to 65nm.

[911medic]

For another, it's easier to learn the process and work the kinks out with a lower complexity product then move to the high complexity ones once you're reasonably confident it works well. AMD did this moving to 65nm.That's what I was thinking.

I was thinking the problem with the high-end parts moving to a smaller process was more recent than the GF3, though... But my memory's not what it used to be. :retard:

[magitek]

Whatever it is, I wish they'd hurry it up. My mobo hates my current ATI card, and I'd like to get something top-of-the-line to replace it.

[withmyshadow]

8800 ultra just came out with specs very similar to the proposed 8900GTX. the only difference between them is 80nm vs 90nm. I guess that means either the yield of 8900GTx isn't so good at 80nm so they had to go back to 90nm for the ultra, or, there's no big pressure from ATI so they put out 8800 ultra for the time being and go directly to 65nm.
but it's interesting when I go back to see the table of all G8x chips scheduled for 2007 ( you can find that table somewhere in this board.) 8900GTX was supposed to be 700/2200 @80nm while 8800GTX was 575/1800@90nm. It looks like the jump from 90 to 80 isn't so much worthwhile. imho,...

[retsam]

8800 ultra just came out with specs very similar to the proposed 8900GTX. the only difference between them is 80nm vs 90nm. I guess that means either the yield of 8900GTx isn't so good at 80nm so they had to go back to 90nm for the ultra, or, there's no big pressure from ATI so they put out 8800 ultra for the time being and go directly to 65nm.
but it's interesting when I go back to see the table of all G8x chips scheduled for 2007 ( you can find that table somewhere in this board.) 8900GTX was supposed to be 700/2200 @80nm while 8800GTX was 575/1800@90nm. It looks like the jump from 90 to 80 isn't so much worthwhile. imho,...
you do know that the decument that you saw was a fake, right?

[withmyshadow]

you do know that the decument that you saw was a fake, right?
I didn't.
but I came across it in the rumor mill and I am talking about it as a rumor.
nv doesn't make any official comments on future products so anything regarding 8900 or 9x00 is not solid.

but since 8800 ultra just came out so I don't think they are gonna put out either 8900 or 9x00 before oct. I guess 8800gts should stay on the "sweet spot" until thanksgiving, at least.
just my speculation. but I may be getting one of those next month.

[Domell]

I wonder what NVIDIA push against r650/rv670 which will be released in Q3....
I hope they will do the same thing like in 2004/2005 when skipping GF6800s refresh and a few months later launch much faster GF7800.... Then they could push G9x series against ATIs r600 refresh - r650....

[Madpistol]

I wonder what NVIDIA push against r650/rv670 which will be released in Q3....
I hope they will do the same thing like in 2004/2005 when skipping GF6800s refresh and a few months later launch much faster GF7800.... Then they could push G9x series against ATIs r600 refresh - r650....

I honestly hope that next time we get 2 very powerful cards from ATI and Nvidia around the same time. If we do, we get a massive price war, and everyone wins. $699 for a video card is madness. 4 years ago, the premium for a video card was $400... What the hell happened to that?

Greed... that's what happened.

[MUYA]

Greed... that's what happened.
Actually it is probably to do with BOM and availability of the part listed in it. High speed memory...yeilds on silicon wafers, PCB complecity, HSF costs...all factor in while trying to maintain a reasonable price point...got to drawt he line somewhere, it's not like NV or ATi ot Intel can sell stuff at a discount like consoles...where console makers can get licensing or royalties from game dev houses!

[nrdstrm]

Actually it is probably to do with BOM and availability of the part listed in it. High speed memory...yeilds on silicon wafers, PCB complecity, HSF costs...all factor in while trying to maintain a reasonable price point...got to drawt he line somewhere, it's not like NV or ATi ot Intel can sell stuff at a discount like consoles...where console makers can get licensing or royalties from game dev houses!

QFT....But you may want to explain your accronyms to some here, such as Bill of Materials...I don't think everyone knows all of those :P

[coldpower27]

I dunno, 65nm at this point is still much to immature It would be something though if Nvidia went straight to 65nm from 90nm without touching the half node for high end SKU's.

[911medic]

I dunno, 65nm at this point is still much to immature It would be something though if Nvidia went straight to 65nm from 90nm without touching the half node for high end SKU's.If the 80nm process is really to blame for the new AMD part's less-than-expected performance, they might do well to skip it.

[Daneel Olivaw]

QFT....But you may want to explain your accronyms to some here, such as Bill of Materials...I don't think everyone knows all of those :P
Does anyone else smell the telltale smell of ERP software around?

[withmyshadow]

If the 80nm process is really to blame for the new AMD part's less-than-expected performance, they might do well to skip it.
I agree.
there's not much pressure from ati's 80nm chips now that it's officially released. so nv does have some time working on 65nm, with the 8800 ultra staying on top until around thanksgiving.

a couple of R600 reviews mentioned ati's chance of filling the performance gap between 8600GTS and 8800GTS320 with a new board around $250. does anyone think 8800GTS320 is going to drop a little in the coming weeks? ;)

[SH0DAN]

7800->7900 was a die shrink from 110 nm to 90 nm. G70 (7800) was 334 mm^2, G71 was 196 mm^2.



302(G70) million trannys versus 278(G71) million transistors.