NVIDIA Kepler speculation thread

I haven't seen anyone reliable state that it is consistently or considerably faster than a 7970. Given more or less equal performance and equal price, I know which card has more bandwidth and 3GB of memory. So does every other enthusiast...

 

I saw quite a bit of gnashing of teeth in the 7970 threads related to its price with the primary complaint being a new generation of GPUs should not cost as much as last cycle's best card and best it by such "small" margins. Ironically, while you would expect pages of outrage rage if NV released a 680 which is roughly ~ to a 7970 and priced in the same ball park and hence a similar situation, much of the gnashing of teeth won't be heard. This shall be very, very interesting

 

http://www.chiphell.com/thread-367824-1-1.html

 

Nice card - just too expensive.

 

Bah, no picture of the display outputs. I'm not going to get even remotely interested until I can see with my own eyes that Nvidia finally has DP outputs standard.

Then again, at the likely price it'll come out at, it's unlikely I'll get excited anyway. At least with 7970, Zeropower made it tempting to break my budget. Here's hoping that Nvidia has something similar.

Regards,
SB

 

I think I was the primary instigator of that. It's not that the price was OTT per se, it's that it was OTT for what the card was overall. A very marginal step up from the GTX580, and now we have NV laughing their arses off and releasing their 3rd fastest chip which shoulda been $299 as their 1st fastest chip at $550. We, the customer, have lost this round. Pathetic on all sides.

 

the 4GB version will be a fine CUDA 3D renderer.

 

I have the feeling we really dont know all of what was originally this card and what it have become thoses last month.

We have heard E. Demers right after the 7970 release ( rage3D interview ) who was hint about the fact they had discover a bit lately the card could have been clocked faster on both core and memory ( spec was allready fixed ).

What we have not see is what have happend for Nvidia during this time.


Something is clear, Nvidia have make some change in his lineup, but im not sure the reason is just " we was too fast for AMD " .

the pcb with 2x SLI connectors is here for tell you this is not a mid range card. ( they will never allow the use of 3-4 midrange cards if they was a model upper )...

I think we have not finish to heard about the Kepler story.. specially if the refresh of it, is not set to be here before a while ( 2 -3- 6 months ?? ).
( I dont want even call it the high end, it could be well different of what we think )

 

Mr. Baumann already pointed out that AMD prioritized timing over performance when launching Tahiti.

So they never intended to start with the "fastest possible card" (which is what you refer to as "the enthusiast card they should have started with") in the first place.

The decision basically came down to either:

• Launch HD7970 four mounth before Nvidia gets anything based on 28nm out of the door. Pro: One quarter of uncontested single-chip performance superiority; good and early profits; card of overclocker's wet dreams; some nice PR; happy shareholders; you can still launch a "Tahiti XTX" card a few months later and make some extra money. Contra: Lack of time to test and readjust for Tahiti's better-than-expected clock potential; stock clocks and performance stay lower than actually possible.
• Push the HD7970 launch back to mid Q1 2012 - and squeeze out another ~20% of performance. Pro: Higher stock clocks and performance; a little more performance pressure on NVidia's still undisclosed 28nm cards; jimbo75 happy. Contra: Loss of considerable timing advantage and pressure on NVidia; delayed profits on chips already payed for and out of the fab; unhappy shareholders; considerable extra investments into re-testing and validation; no more headroom for a "Tahiti XTX" card.

Can you really blame them for going with the first option?

 

but we've been stuck with gddr5 and a 200 watt / 250 watt TDP.
we can't double TDP and bandwith anymore as in going from 8800GT/GTS 250 to GTX 460/560.

both vendors are also busy vending in masses Barts and GF114, everyone who wants a very powerful and affordable gaming rig gets these. they won't pull the plug on them until 28nm is affluent.

hate it? don't buy it
if you have a 5870 or something.. then you don't need a new card.

 

http://forum.beyond3d.com/showpost.php?p=1626393&postcount=2246

 

So it looks like OBR had the upper hand on Charlie this time, regarding the latter affirmation that yesterday was the launch

 

IMHO (big IMHO!)

It could very well be that they have a commitment with the partners (Microsoft, Sony) that they had to release the first HD7000 card before the end of 2011 to support the game development.

 

Nah I don't blame them for that, but losing to Nvidia's performance card (if it happens) with a larger chip is a step too far in the wrong direction. Surely they could have clocked it a bit higher off the bat without a major loss of top bin - I haven't seen any review where it didn't hit over 1 GHz overclock.

For me they got greedy (I'm not complaining about the price) and they are lucky Nvidia decided to go with the 680 - if they'd gone for a $299 660 Ti instead, whatever small gains AMD made in the 3 months up till now would soon have been totally negated.

 

I'm going to speculate the GTX680 will launch at $399, and the GTX670 at $299
I don't believe NV will launch with price/performance parity with AMD, since
1) They are 3-4 months behind AMD, so equal price-performance will seem lame, and not win any kudos...
2) The GK104 looks to be small & powerful, so when NV have a winner, they usually go after market...
3) We know that can sell similarly-sized chips for $249 & still make good money, so, assuming volume are ok, then they have no need to price it so highly.
4) The $550 price rumored price tag, is most likely just a smoke-screen...
5) Even if volumes are low initially, TSMC will fix their problems, and get back to normal fairly quickly, so any shortages are just a temporary blip.

 

Equal price/performance is in reality a win for nVidia due to a stronger brand. Doubt they would care about kudos if the money still rolls in. The clock starts when people are ready to buy so unless the entire market was hankering for a $550 graphics card these past three months GK104 won't suffer any "late" penalty if it's at least on par with Tahiti.

If GK110 is coming in 6-9 months from now it'll probably not be a blockbuster compared to GK104/Tahiti anyway. Maybe +35% or so assuming lots of transistor budget goes to HPC.

 

You think nv will release a 580 stock-killer while there are plenty of 580s in channel? Not likely.
Cheapest 580s are going for 450usd today...figure 350usd post 680 launch. 680 will be 499+usd, 670 399usd... Or not.

 

What about mindshare? Clearly Nvidia enjoyed the title for single GPU since X1900XTX, it seems like they are willing to share it for the first time in x years.

 

Do they even have a choice? There still isn't any confirmation on either clocks or performance compared to Tahiti.

 

A verdetroll edition of Tahiti (aka XTX) should easily keep the performance crown in AMD´s hands.