NVIDIA GT200 Rumours & Speculation Thread

[Jaaanosik]

Is it? I didn't bother to check, but G80 was made on 90 nm, not 65 nm... So the physical diesize might not be all that different.

Besides that, G80 was 2 years ago, these days production costs are different.
Heck, even the 7800 was a very expensive card at its introduction, or the 6800, or the 5800 etc.

Aside from that, GTX260 will be a 'broken' GTX280, so that changes economics. In the technical sense it's just 'leftovers'. The diesize itself isn't all that important... What's more important is how much they can improve yields by disabling various broken modules on the chip, and how much performance is left.
Again, look at the 8800GTS, which is basically a 'broken' 8800GTX/Ultra. By simply disabling 32 of the 128 processing units, and reducing the clockspeed a bit, they had a card that was only about 20% slower, but which they could sell at only about 60% of the price of an 8800GTX, making it an incredible bang-for-the-buck card, and probably ATi's biggest nightmare at the time (even their 2900XT had problems competing).

The die size is very important. A die that was supposed to be sold on $600 card is now being sold on $400. But they both cost the same money to manufacture.

 

[DegustatoR]

To compete with G92b?

GTX260 should be quite faster then G92b.
And from what was leaked 4850 is more or less on par with old G92.

(Maybe my wording was wrong. I meant to say that GTX260 will be heavily stripped down version of GTX280 because NV knows the yields of GT200. If these yields were good they woulnd't produce a GTX260 as we know it, they'd probably lower the frequencies but they wouldn't turn off 2 out of 10 SMPs and 1 ROP.)

 

[INKster]

To compete with G92b?

Jawed

Given that all signs point to G92b as a strictly midrange product for the 9800 GT (meaning, no 8800 GTS 512MB/9800 GTX-like variants), i'd say there's still quite a gap to be filled between 199 and 399~449 dollars.
Especially the 9800 GTX, because i believe it's a very short term solution, just like the 9800 GX2.

GTX260 should be quite faster then G92b.
And from what was leaked 4850 is more or less on par with old G92.

(Maybe my wording was wrong. I meant to say that GTX260 will be heavily stripped down version of GTX280 because NV knows the yields of GT200. If these yields were good they woulnd't produce a GTX260 as we know it, they'd probably lower the frequencies but they wouldn't turn off 2 out of 10 SMPs and 1 ROP.)

Are those yields so bad, or are we talking "The Inq-rumor" craziness ?
There are signs that eVGA, for instance, is readying no less than 10 different GTX 2xx SKU's (5 for GTX 280, another 5 for GTX 260), with multiple factory OC'ed cards.
Remember that OC'ed versions of G80 only came out much later ?
So, this rumor of the so called "bad yields" just doesn't jive.

 

[DegustatoR]

A die that was supposed to be sold on $600 card is now being sold on $400.

Is it? Were?
You have NO idea what price range was 'supposed' for the GTX200.
You CAN'T make any assumptions out of a die size alone.
Let's leave this BS to Fuad and Inq, they're getting increasingly good at it.

 

[Thorburn]

Is it? Were?
You have NO idea what price range was 'supposed' for the GTX200.
You CAN'T make any assumptions out of a die size alone.
Let's leave this BS to Fuad and Inq, they're getting increasingly good at it.

You can make a reasonable assumption, G92 and GT200 are fabbed on the same process node, so a wafer of them should cost the same more or less.
The chip is large, so it'll be expensive to make, and if they have to sell them at a lower price then the margins fall (obviously).

 

[v_rr]

There are signs that eVGA, for instance, is readying no less than 10 different GTX 2xx SKU's (5 for GTX 280, another 5 for GTX 260), with multiple factory OC'ed cards.
Remember that OC'ed versions of G80 only came out much later ?
So, this rumor of the so called "bad yields" just doesn't jive.

That´s ridiculous. 10 diferent version with only G280/G260 chips?

The guy who said that have a very solid imagination

The core of GT200 cost as much as 100-130$ with good yields. If the yields where bad the GTX 280 pricing in 1000$ wasn´t enough.

100-130$ for the core + 512bit PCB + 1Gb GDDR3 + Super-expensive cooler leaves no space to decrease pricing.
Only if they make a new core with less transistores like 9600GT is the only way to reduce prices.

 

[INKster]

You can make a reasonable assumption, G92 and GT200 are fabbed on the same process node, so a wafer of them should cost the same more or less.
The chip is large, so it'll be expensive to make, and if they have to sell them at a lower price then the margins fall (obviously).

There are two sources for the current 65nm G92, TSMC and UMC (so, it had to be designed to accommodate both foundries' technologies, libraries, etc).
But all top-end large die chips from Nvidia are fabbed only by TSMC, which still has the upper hand in process tech against UMC.

Take that as you will.

That´s ridiculous. 10 diferent version with only 2 chips?

The guy who said that have a very solid imagination

Maybe. Maybe not...

 

[Jawed]

Given that all signs point to G92b as a strictly midrange product for the 9800 GT (meaning, no 8800 GTS 512MB/9800 GTX-like variants), i'd say there's still quite a gap to be filled between 199 and 399~449 dollars.
Especially the 9800 GTX, because i believe it's a very short term solution, just like the 9800 GX2.

Doesn't make sense to produce G92b just to make 112-SP GPUs, unless it's actually a smaller chip - and therefore isn't G92xxx.

So I expect G92b will result in a 9900GTX, and we could see it coupled with the very fastest GDDR3, 1250MHz+, presumably. A bandwidth bump should make 9900GTX that little bit more competitive with HD4850.

Are those yields so bad, or are we talking "The Inq-rumor" craziness ?
There are signs that eVGA, for instance, is readying no less than 10 different GTX 2xx SKU's (5 for GTX 280, another 5 for GTX 260), with multiple factory OC'ed cards.
Remember that OC'ed versions of G80 only came out much later ?
So, this rumor of the so called "bad yields" just doesn't jive.

Until the quantities of these GPUs are known we don't know if they're nothing more than benchmark-editions. NVidia AIBs are used to competing with each other with the OC, XXX etc. variants so the fact that there'll be GTX280 OC products doesn't indicate volume.

Jawed

 

[Jawed]

GTX260 should be quite faster then G92b.
And from what was leaked 4850 is more or less on par with old G92.

It would appear that HD4850 will be as fast as any G92/G92b that NVidia can produce.

(Maybe my wording was wrong. I meant to say that GTX260 will be heavily stripped down version of GTX280 because NV knows the yields of GT200. If these yields were good they woulnd't produce a GTX260 as we know it, they'd probably lower the frequencies but they wouldn't turn off 2 out of 10 SMPs and 1 ROP.)

OK, thought you meant that there'd be a version of GT200 with even less clusters/partitions than GTX260.

Jawed

 

[INKster]

Doesn't make sense to produce G92b just to make 112-SP GPUs, unless it's actually a smaller chip - and therefore isn't G92xxx.

So I expect G92b will result in a 9900GTX, and we could see it coupled with the very fastest GDDR3, 1250MHz+, presumably. A bandwidth bump should make 9900GTX that little bit more competitive with HD4850.

Why not ?
Have you seen many G98-based boards with 16 enabled sp's variants lately ?

Until the quantities of these GPUs are known we don't know if they're nothing more than benchmark-editions. NVidia AIBs are used to competing with each other with the OC, XXX etc. variants so the fact that there'll be GTX280 OC products doesn't indicate volume.

Jawed

G80 OC'ed cards were banned by Nvidia for a few months back in late 2006 due to yield and card return rate concerns, so they would definitively step in if G200 suffered from the same issue.
They're an interested party.

 

[Slappi]

280s do not have any yield issues.

Figment of a biased disgruntled fiction writer.

 

[satein]

280s do not have any yield issues.

Figment of a biased disgruntled fiction writer.

Are you sure Are you working for TSMC

No yield issues hopefully does not mean by 0% defect If yes, why bother to produce 260 since NV can produce a lot of 280 and earn profit

 

[Sxotty]

Wow. Charlie must really dislike nVIDIA. And thats an understatement!

His articles are hilarious now. Coherency is no longer even required if he can bash MS or Nvidia.

 

[INKster]

His articles are hilarious now. Coherency is no longer even required if he can bash MS or Nvidia.

Notice the not-so-subtle jab at Nvidia right in the end, like if that theoretical number somehow had anything to do with the cards' actual performance, or even with the subject of the article in the first place...

 

[Thorburn]

His articles are hilarious now. Coherency is no longer even required if he can bash MS or Nvidia.

Calling Vista 'Me II' and babbling on about DRM which doesn't actually affect you in day to day usage never gets old, he keeps his style hip and fresh yo!

 

[Slappi]

Are you sure Are you working for TSMC

No yield issues hopefully does not mean by 0% defect If yes, why bother to produce 260 since NV can produce a lot of 280 and earn profit

When I say no yield issues, that comes from a perspective of what Nvidia anticipated the yield to be.

The bogus rumor going around for a few weeks that was pulled from thin air was that Nvidia was having yield problems with their 280s.

The bogus claim implied that Nvidia was not meeting their predetermined yield goals for the 280s. That claim is WRONG.

 

[Jaaanosik]

Is it? Were?
You have NO idea what price range was 'supposed' for the GTX200.
You CAN'T make any assumptions out of a die size alone.
Let's leave this BS to Fuad and Inq, they're getting increasingly good at it.

OK, my bad. I should've been more specific. I don't know what's going to be the real price.
I wanted to point out that bigger die is being sold on a cheaper card but still costs the same money as the one on more expensive card.

 

[Scali]

The die size is very important. A die that was supposed to be sold on $600 card is now being sold on $400. But they both cost the same money to manufacture.

But the thing is that there's little if any relation between what it costs to build the card and what the card is sold for.
The actual manufacturing costs of processors isn't very high at all.
What makes chip manufacturing expensive is mostly the development of the design and the investment in the production facilities for these chips. Once all that is in place, building the chip doesn't cost anywhere near $400 to produce, let alone $600.
Most of the price of GPUs, CPUs and that sort of hardware is for return on investment, not for the manufacturing costs of the actual physical product.
I'd be surprised if a GTX260/GTX280 chip would cost more than $50 to manufacture.

You just can't compare it with most other products where you mainly pay for the materials and the labour. A chip is very small, with very little material, but most of them cost more than their weight in gold. And it doesn't take much labour/time to build a chip.

 

[Jawed]

Why not ?
Have you seen many G98-based boards with 16 enabled sp's variants lately ?

Haven't seen any G98 boards actually. But more to the point, because NVidia has nothing but coarse-grained redundancy they are, perhaps, forced to be brutal about it that price/die-size level.

Has it launched in the West yet?

G80 OC'ed cards were banned by Nvidia for a few months back in late 2006 due to yield and card return rate concerns, so they would definitively step in if G200 suffered from the same issue.
They're an interested party.

If NVidia has better testing then their confidence in their test result of each die is prolly higher. That wouldn't alter the actual yield rate though. It just means they'd be more willing to offer a warranty on the chips that meet OC criteria.

I think you're reading too-much into the availability of OC cards. Since GT200 is ~8 months late the chip has clearly been problematic - now it's just a question of volume. Whatever happens, it'll be spun with tales of unprecedented demand, I'm sure.

Jawed

 

[Jaaanosik]

Doesn't make sense to produce G92b just to make 112-SP GPUs, unless it's actually a smaller chip - and therefore isn't G92xxx.

So I expect G92b will result in a 9900GTX, and we could see it coupled with the very fastest GDDR3, 1250MHz+, presumably. A bandwidth bump should make 9900GTX that little bit more competitive with HD4850.


Until the quantities of these GPUs are known we don't know if they're nothing more than benchmark-editions. NVidia AIBs are used to competing with each other with the OC, XXX etc. variants so the fact that there'll be GTX280 OC products doesn't indicate volume.

Jawed

Performance yes, money no. This new GTX would have more expensive parts.