ELSA hints GT206 and GT212

willard's right. It seems like one revision of the chip is shipping now with Tesla, with the last revision they taped out set for GeForce soon. My sources say it's just come back from packaging in Taiwan, so MP clearance should happen soon all going well.

That they needed another spin for GeForce should give a couple of hint as to what they plan to do with it, that they couldn't do with a revision that was OK for Tesla.

 

Hang on a minute.. You're implying that Tesla would be using 55nm GT200b chip. But there has been apparently verified rumours that GT200b would be omitting dual precision. Surely not with the Tesla line

But would this consumer revision omit DP and have possibly higher clocks? But then it wouldn't merely be a "spin". Something is wrong here.

 

@ahu: Rys indicated there's a difference between the GT200b Tesla and GeForce variants. I'd guess they omitted the dual precision for the GeForce variant in favor of higher clocks.

 

GT200b is just a 55nm version of GT200. DP will probably be gone from GT206.

You can't omit something like this via a respin. I think they're going for higher clocks or better yield rates. And higher clocks than current Tesla line means that we'll see GT200b-based "GTX290".

 

But what would be the point of releasing a G206 if there's a G200b? Unless it had less functional units compared to what the rumours said (192 SPs mayhap, but I'm not very convinced about it).
Could it perhaps be possible that nVidia tried to make some minor changes to the G200 design before shrinking it (I mean minor changes like bugfixing, not leaving out DP units or whatever), but instead of fixing it, they've broken something? Then, if the problem was in the part that is not used in Tesla cards, that would explain why graphics need a new revision.
By this time, they could as well have a brand new chip, a G200 derivative at 55nm. Just remember when did R600 and RV670 come out. I'd think that G206 was the logical choice - less SPs, 256 bit, no DP, no need for NVIO, higher clocks... but if Rys says G200b is actually shipping

 

What rumours? =) I've always thought that GT206 will have less than 10 TPCs -- probably 6 or 8. It'll have 256-bit GDDR5 bus and it won't have some GPGPU features of GT200(b). Basically it'll be a mainstream GT2xx chip, a direct competitor to RV770, which means that it'll be slower than GT200 and even slowerer =) than GT200b.
NV needed GT200b for Quadro/Tesla anyway so the real question is why would they want to use GT200b for the GeForce line? Maybe GT206 is still too complex for the GX2-type card and they need something against X2. And GT200b will have substantially higher clocks than GT200 and will be able to outperform RV770X2 while being more cost-effective solution than GT206-based GX2 card? Or maybe GT206 isn't as close to market as we thought and GT200b is a temporary solution to this?

 

Please don't make me search for the link, perhaps it's somewhere in this thread. According to that source, G206 was supposed to have 216 SPs (9 TPCs). But it could still be faster than G200. 55nm manufacturing would allow for a higher core clock (overall) and optimizations made to the design would allow for higher shader-domain clock. nVidia shrinking G200 to 55nm without having the chance of optimizing the design... seems to me the RV770 caught them by *absolute* surprise and they have no new GPUs whatsoever. But the G92 has been on the market for one year now, the G200 was delayed by approximately 6 months, so there has to be something new coming besides 55nm shrinks!

 

There are rumours that NVIDIAs refresh is going to be launched 22 of October this year. This is about two weeks and there are NO benchmarks leaked, NO die shots, NO possible specs, NO clocks and NO prices.

I wonder what will NVIDIA launch 22 of October. Will it be only GT200B or maybe GT206 too? On ELSA slides GT206 is positioned as a little faster than current GTX260 (old or new version?).
Do you think there is any chance that GT206 will be the second G92? Almost as fast as current GTX280 with price about ATIs Hd4870? Thats`s is what NVIDIA needs the most.

 

Could be, but only if its schedule was unaffected by G200 delay.

 

The 9800GTX+ was also a secret until 1-2 days before it's paperlaunch. Oh, and don't forget the 8800GTX...

 

Well, but remember that GF8800GTX performance numbers were on Chinese sites (especially PCinlife) before it was launched.
Take a GT200 or G92 or even G70. There were some info and first 3d Marks benchmarks about at least 2 weeks before they were hit the market. This situation is quite strange for me.

Maybe NVIDIA is not able to release a worthy competitor to HD4870 or HD4850 (with decent price/performance ratio)?

 

Well the 280 is faster than the 4870 non X2 model and I doubt the refresh of the 280 or the 280+ is going to be that much faster. It will be closing the gap against the X2 sure, but thats about it. I still think it a bit unfair and Nvidia is doing themselves no favors by NOT launching cards with multiple GPUs in them. Its a segment product vs segment product comparison not a direct hardware comparison and thats why Nvidia is losing.

 

Inq says now that GT206=GT200b and we can expect the cards (gtx270,gtx290) in november:
http://www.theinquirer.net/gb/inquirer/news/2008/10/08/nvidia-270-290-gx2-roll

That also to some extent explain the lack of leaks etc, if it's just the dieshrink and in a months time.

So time to merge the threads

 

Well, it's the Inq...

...

...

http://www.theinquirer.net/gb/inquirer/news/2008/05/29/nvidia-gt200-sucessor-tapes

 

igg: and the point? Most of the info in that article turned out to be pretty much on spot. NV just had to do another respin, which we heard about around late july, and there's a change in code name. Ofcourse you should see through his rantings and to the actual info, but his sources have been pretty reliable since this spring.

 

I'll believe in a GT200-based GX2 card when I see it. I don't believe for a second that the shrink to 55nm has made such a monstrosity even remotely viable.

 

Agreed.

 

And what's the point of such name change?

 

It would be interesting, if true. Several times now I have confused GT200b with GT206 so I would LMAO if it turned out to be true!

 

GT200b is in line with other 65->55nm shrinks, GT206 isn't. Nobody gives a f about codenames. Why would they rename GT200 55nm shrink from GT200b to GT206? That's pointless. So i'll hold to my info about GT206 having 256-bit bus and thus not being a simple GT200 shrink to 55nm -)