Nvidia GT200b rumours and speculation thread

[Anarchist4000]

The only logical reason for not having the GT200b would be if they got stuck on a pile of current inventory. With the price cuts I'd think that is unlikely. It's possible the specs are identical to the current parts but they cut the size and improved yields instead of making a whole new generation out of it.

 

[SimBy]

If GT200 shrink is coming, it's not coming until inventory of 65nm is cleared. Nvidia already produces 55nm only. Those are Nvidias CEO words.

I guess they really didn't expect 4800 to be so good so it's a lot harder for them to sell all that 65nm stuff. And of course they can't just start selling 55nm parts cos no one would buy 65nm then.

 

[fellix]

Probably NV thought and tuned its marketing strategy upon the initial 640SP/32TMU design of the RV770, that was widely rumored (and expected). So, I guess it was a final hour surprise, sort of.

 

[INKster]

Probably NV thought and tuned its marketing strategy upon the initial 640SP/32TMU design of the RV770, that was widely rumored (and expected). So, I guess it was a final hour surprise, sort of.

I don't think it had anything to do with that, honestly.
Architectural decisions are made 2 or more years in advance of the actual market launches. There's little that can be done just before launch, outside of pushing clockspeeds as much as the process/design allows.
The 65nm GT200 was at least 6 months late due to various issues (this also explains somewhat the G92-based product invasion up-to, and after GT200, in those 6 months).
Had it launched on time, would you say that it was a failure next to the RV770 ?

I'm thinking that "GT200b" is actually more akin to what G71 was to G70, than what G92b was to the G92.
Not merely an optical shrink, but more of a redundant/test circuitry removal process, with the aim of boosting clockspeed, now that the basic design is further refined/debugged.

 

[DegustatoR]

I'm thinking that "GT200b" is actually more akin to what G71 was to G70, than what G92b was to the G92.

It wouldn't be called GT200b then.

 

[Shtal]

Nvidia might be adding some more horsepower to this chip and the thermals will be much better which might imply that there is a change. GT200 55nm with more Shaders and some additional architectural adjustments might have a chance. http://www.fudzilla.com/index.php?option=com_content&task=view&id=8951&Itemid=1

Here we go again "Might"

 

[Domell]

I think GT200b will only have higher clocks than current GTX280/260 and will be cheaper to produce. The most important thing imo is that NVIDIA doesn`t have to release much faster GPU than GTX280/260 but a little faster GPU (~10-15%) but with lower price. If GT200b will cost about 350$ for GTX280+/GTX290 (??) and about 200$ for GTX260+/GTX270 (??) and will offer the same performance level like GTX280 and GTX260 that would be great and cards will be much more succesful.

But IMO the best choice for NVIDIA is releasing GPU based on GT200 architecture (maybe with some improvements) with specs which were mentioned by Arun a few posts earlier (384-bit MC and 192SP with higher clocks).

 

[Twinkie]

If G200 was 6 months late, i guessing we could see a slightly modified (i.e removal of the unnecessary fat) G200 chip on 55nm could be something along the lines of G70 to G71.

And since when was "GT200b" confirmed?

 

[CJ]

Last I heard about GT200b was that somewhere in July the chip was going back for another spin. That should put it in the September/October time frame for release.... with a possible preview at nVision. But hey, things change a lot... who knows what they've decided about it in the meantime.

 

[trinibwoy]

Here we go again "Might"

Yep, Nvidia is just shadow boxing right now with a bunch of cryptic hints and suggestive comments. But they aren't worth crap.

 

[DegustatoR]

Fuad: GT200 55nm to come in Q4 2008
And another spin?

 

[INKster]

Fuad: GT200 55nm to come in Q4 2008
And another spin?

That story seems like little more than baseless and uneducated speculation.
With a 512bit bus, a straight shrink while adding GDDR5 is completely redundant, since GT200's primary performance limitation doesn't lie in its native memory bandwidth.

It would only make sense if the chip was significantly reworked (dropping ROP's and reducing memory channels to cut on overall chip sizes, adding GDDR5 support, speeding up both core and shader core clocks, etc), which then wouldn't have that much in common with a GT200, and would be totally inconsistent with a die shrink anyway.

 

[ShaidarHaran]

That story seems like little more than baseless and uneducated speculation.
With a 512bit bus, a straight shrink while adding GDDR5 is completely redundant, since GT200's primary performance limitation doesn't lie in its native memory bandwidth.

Not entirely true... It's already been demonstrated that bandwidth is the limiting factor on MSAA performance (a limitation of the ROP design).

 

[DegustatoR]

It would only make sense if the chip was significantly reworked

Maybe it is? Maybe that's why it's coming in Q4 now? Maybe that's why i heard that GT200b won't be released at all? =)

 

[MfA]

Why is the option they fucked up and had to delay the shrink version just like they had to delay the original not considered? They are not immune to fuck ups as the G200 shows.

 

[Jawed]

You've gotta wonder how much inventory of GT200 there is - if NVidia is planning a Q4 chip that isn't GT200-based, is it worthwhile refreshing GT200 to 55nm if they won't be able to sell GT200b until November due to piles of GT200s hanging around?

In the CC they mentioned they had 69 days' inventory - their target is 70 days. But there's no breakdown of inventory by SKU. But with GT200 selling much worse than expected, presumably they've got rather more than 70 days inventory of them. Which takes them well into October.

Jawed

 

[Mintmaster]

You guys should all remember that 55nm may not yet be cheaper than 65nm per transistor, and it most certainly is more expensive per mm2. It depends on whether TSMC is saturated in its 55nm production ability. They'll negotiate contracts with ATI/NV that balance everything out.

In the end, we can only assume that ATI and NVidia are choosing the process which gives them the lowest cost, including one-time design/tapeout costs. Well, sometimes they'll choose a higher cost process if it has higher performance (thus higher ASPs), but that hasn't happened for a while.

Why is the option they fucked up and had to delay the shrink version just like they had to delay the original not considered? They are not immune to fuck ups as the G200 shows.

I wouldn't go so far as to call GT200 a fuckup. It was just a combination of relying on high clock speed in a huge chip and ATI's out-of-nowhere efficiency gains, and the latter was both out of NVidia's control and impossible to predict.

 

[Jawed]

You guys should all remember that 55nm may not yet be cheaper than 65nm per transistor, and it most certainly is more expensive per mm2. It depends on whether TSMC is saturated in its 55nm production ability. They'll negotiate contracts with ATI/NV that balance everything out.

55nm is definitely cheaper for NVidia's GPUs now. During the CC they opined the fact they can't sell their stockpiled 55nm GPUs because they have 65nm GPUs left to sell.

I wouldn't go so far as to call GT200 a fuckup. It was just a combination of relying on high clock speed in a huge chip and ATI's out-of-nowhere efficiency gains, and the latter was both out of NVidia's control and impossible to predict.

You conveniently ignore the fuck-up that is not launching back in 2007Q4.

Jawed

 

[Mintmaster]

You conveniently ignore the fuck-up that is not launching back in 2007Q4.

True, but while an earlier launch would improve GT200 sales, it would have likely cut into their 9800GTX/GX2 sales since then and thus would have barely helped NVidia overall.

NVidia's margins fell the instant RV770 came out. Early GT200 sales were gimped in a large part to huge price cut of G92 based products.

Before RV770, there was no competition at the high end, so the delay really wasn't a fuckup at all.

 

[DegustatoR]

Why is the option they fucked up and had to delay the shrink version just like they had to delay the original not considered? They are not immune to fuck ups as the G200 shows.

GT200 isn't a fuck up and i'm not that sure that they've delayed it because something was fucked, to me it looks more like it was their decision not to counter RV670 and RV670x2 with GT200 but with G92.
They fucked up the initial pricing and underestimated the RV770 performance (which are connected) -- but that aren't a problems of GT200, these are problems of their marketing team.