The first "proper" Kepler news Fri 17th Feb?

[Ghøst]

Nothing interesting & new coming from the Kepler rumour factory then?

 

[Phage]

No - nothing.
Looks like an AMD solution for me when the tax cheque comes in.

 

[ALXAndy]

What roadmap that Nvidia released?

If you mean this old chestnut, then what do you have to suggest that this roadmap was released by Nvidia? Or that it's any more genuine than the other 1000 rumours that cite "unspecified but reliable sources"?

I was referring to the one where a guy from Nvidia was standing below a huge screen.

http://hothardware.com/News/NVIDIA-Exposes-GPU-Roadmap-Kepler-Arrives-2011-Maxwell-in-2013/

However, it seems with all of the rumour and speculation I've gotten my dates mixed up. Kepler should have been released last year according to Nvidia. They were supposed to go to production late in 2011.

I know that I read (pre 7970) that Kepler was being released in two phases. One in Q2-3 2012 as low end solutions (IE laptop chips and mobile devices) and that the desktop GPUs would show up in Q4.

However, given Nvidia said it would be out last year I do suspect they are having problems with it.

 

[Meaker]

That's already out...

It's called GT6xxM (mobile) series, and is basically die-shrunk Fermi from 40nm to 28nm. It was done to reduce power consumption for Fermi architecture mobile GPUs and as nVidia's test bench for 28nm process.

The fact that die-shrunk Fermi has taken the GT6xx moniker has prompted the speculation that Kepler GPUs will be called GTX7xx.

GT6xxM are pure rebrands and no shrink has been made. Identical clocks, identical TDP.

 

[andybird123]

GT6xxM are pure rebrands and no shrink has been made. Identical clocks, identical TDP.

some of them are 40nm, the ones going in to the new ivybridge ultrabooks that intel are holding back are supposed to be 28nm kepler derived (like the 670m going in to the Asus ROG laptops), as far as I can find on google

 

[andybird123]

However, it seems with all of the rumour and speculation I've gotten my dates mixed up. Kepler should have been released last year according to Nvidia. They were supposed to go to production late in 2011.

also don't get confused between chips being manufactured at TSMC, vs. CARDS being made at AIB partners - Nvidia need to manufacture the chips first and build up enough volume of "good" ones before they can start giving them to the board makers to actually make in to cards

saying that yields are low indicates that Nvidia HAVE started manufacturing them at least, and may have been for some time



AMD didn't wait for a show before releasing their cards, they did it entirely over the internet and then turned up to a show a couple of weeks later, I would imagine Nvidia will be doing the same, I'm expecting reviews to start appearing as the "release", not some big announcement at a show

 

[CAT-THE-FIFTH]

Supposedly,the GK104 will be called the GTX670TI and will outperform the GTX580 and HD7950:

http://www.techpowerup.com/160933/Top-NVIDIA-GK104-Part-Gets-GeForce-GTX-670-Ti-Branding.html

 

[mfj1985]

Too bad everything about Kepler is "supposedly"

 

[Harlequin]

that tpu link sources vr zone

 

[sxndy]

AMD and NVidia both have their marketing gimmicks. 99% of users will never require triple monitor support (Eyefinity) but that does not stop AMD and AMD fanboys pronouncing it's usefulness.
Neither are necessary.

You may be correct, I have 3 but most of my m8s have 4 monitors hooked up. Olds school Eyefinity is useless to them.

 

[ALXAndy]

OK more wind for the rumour mill.

http://wccftech.com/nvidia-kepler-g...i-arrives-march-2012-performs-gtx-580-hd7950/

Apparently the 670ti will be the first card to launch, is faster than a 7950 and 580 (yet no mention of the 7970) and will cost $299.

But, none of that adds up with reality. You've got Nvidia making statements about the shortage of wafers, so selling them cheap isn't a good idea.

Then when doing more poking around I found some information about that Charlie D guy saying that it wasn't as fast as a 7950.

Either way there is no way on earth that initial launch Kepler cards (no matter how poor) are going to be selling for $299. If it's faster than a 580 it will cost more than a 580.

And it still doesn't solve Nvidia's problem with being able to take the top end crown, so isn't very exciting.

Given that we have seen that AMD can charge whatever they like for that pleasure (of having the world's fastest single GPU card) I can't see Kepler changing that price unless it can compete. And the 7950 can be clocked to perform as well as a 7970 so I can't see the price on that dropping to unexpected lows either.

 

[DamoTourettes]

OK more wind for the rumour mill.

http://wccftech.com/nvidia-kepler-g...i-arrives-march-2012-performs-gtx-580-hd7950/

Apparently the 670ti will be the first card to launch, is faster than a 7950 and 580 (yet no mention of the 7970) and will cost $299.

But, none of that adds up with reality. You've got Nvidia making statements about the shortage of wafers, so selling them cheap isn't a good idea.

Then when doing more poking around I found some information about that Charlie D guy saying that it wasn't as fast as a 7950.

Either way there is no way on earth that initial launch Kepler cards (no matter how poor) are going to be selling for $299. If it's faster than a 580 it will cost more than a 580.

And it still doesn't solve Nvidia's problem with being able to take the top end crown, so isn't very exciting.

Given that we have seen that AMD can charge whatever they like for that pleasure (of having the world's fastest single GPU card) I can't see Kepler changing that price unless it can compete. And the 7950 can be clocked to perform as well as a 7970 so I can't see the price on that dropping to unexpected lows either.

Unfortunately Andy, you're making complete sense.

 

[Stanners]

OK more wind for the rumour mill.

http://wccftech.com/nvidia-kepler-g...i-arrives-march-2012-performs-gtx-580-hd7950/

Apparently the 670ti will be the first card to launch, is faster than a 7950 and 580 (yet no mention of the 7970) and will cost $299.

But, none of that adds up with reality. You've got Nvidia making statements about the shortage of wafers, so selling them cheap isn't a good idea.

Then when doing more poking around I found some information about that Charlie D guy saying that it wasn't as fast as a 7950.

Either way there is no way on earth that initial launch Kepler cards (no matter how poor) are going to be selling for $299. If it's faster than a 580 it will cost more than a 580.

And it still doesn't solve Nvidia's problem with being able to take the top end crown, so isn't very exciting.

Given that we have seen that AMD can charge whatever they like for that pleasure (of having the world's fastest single GPU card) I can't see Kepler changing that price unless it can compete. And the 7950 can be clocked to perform as well as a 7970 so I can't see the price on that dropping to unexpected lows either.

You maybe correct, But you are making some assumptions that may not. How do we know that the wafer shortage isnt infact in relation to the higher performing parts?

 

[Rroff]

GT6xxM are pure rebrands and no shrink has been made. Identical clocks, identical TDP.

Those going into laptops now are 40nm but as of the end of Jan 2012 all new cores produced for them are supposedly going to be 28nm.

 

[DamoTourettes]

Look, it appears to me that the "wafer shortage" story was a load of old ******** to disguise the fact that engineering issues have stopped Kepler getting to market faster. Charlie Dribbling Demerijian might be full of it, but it's pretty hard to refute 5 different companies all saying they haven't had any problems with the 28nm manufacturing process. Nvidia's dropped the ball, again, like Fermi - the only question is, are people willing to wait for the performance parts? Much as I think the 7970 is nowhere near worth the premium they're charging for the relatively minor performance increase it offers over my current GTX 580, it's still something I'm considering until high end Kepler comes out, which is seemingly Q4 this year at the earliest.

 

[Ghøst]

Right underneath that article that reports on a March release date there's another article from last week claiming there will be another 8 cards launched in April including a 690, no mention of a 680 though.

wccftech.com/nvidia-expected-launch-28nm-kepler-gpus-april-2012/

 

[DamoTourettes]

Right underneath that article that reports on a March release date there's another article from last weeking claiming there will be another 8 cards launched in April including a 690, no mention of a 680 though.

wccftech.com/nvidia-expected-launch-28nm-kepler-gpus-april-2012/

Pinch. Of. Salt.

 

[Ghøst]

For sure a pinch of salt, I ain't believing **** till its beyond reasonable doubt after the fiasco so far.

 

[Phage]

- the only question is, are people willing to wait for the performance parts?

My tax check is here now.

- Much as I think the 7970 is nowhere near worth the premium they're charging for the relatively minor performance increase it offers over my current GTX 580, it's still something I'm considering until high end Kepler comes out, which is seemingly Q4 this year at the earliest.

I'm only upgrading because I missed a whole generation. If I were on a 580, I probably wouldn't bother. It's rarely worthwhile to upgrade every generation.

 

[555BUK]

$299 does seem a little on the cheap side for 7950 performance, but that is mainly due to AMD's revised pricing.

With Fermi, NVidia started by launching the big guns. The GF100 based GTX480 and GTX470 offered stonking performance, but they also ran hot and were quite expensive. It took NVidia a further 6 months before they released their mainstream GF104 based GTX460. In comparison, the GTX460 offered close to GTX470 performance for much less money and with a significantly smaller die. The GTX460 also comfortably outperformed the best card of the previou generation (GTX285), for significantly less money It's UK launch price was ~£160 at most retailers.

This time around, NVidia are doing things the other way around and releasing GF104's replacement (GK104) first. This makes sense, especially when you consider how poor the yeilds were for Fermi's massive GF100's die's during initial launch of the GTX470/480's. Smaller die's always equal higher yields, and yeilds are always worst at the start of a production process. Start small, grow big as the process improves, simples.

Regarding pricing, it is probable that GK104 will have a similar die size and memory bus to GF104. This being so would mean production costs should be similar, give or take inflation, reduction in the cost of VRAM pricing, increases in precious metals etc. Given that GF104 cost £160 at launch and outperformed the previous generations top card, it is not behond the realms of possibilty that GK104 (GTX660/670?) will arrive at a reasonable price and outperform the GTX580.

$299 / £240 is certainly possible. The major obstacle will be AMD's current rip-off pricing for 7900 cards, and whether NVidia want to continue AMD's price reallignment trend.

Perhaps the GTX660/670 will be this generations 8800GT or 5850/5870? With a 256bit bus and 2GB VRAM they should certainly be much cheaper to produce than the 3GB 7900's, and may offer very similar performance. Perhaps this generation, the boot will be on the other foot? Atleast until full-fat GTX680 Kepler arrives at a horrendous price and destroys every other card in it's path.