NVIDIA Kepler speculation thread

Yeah that review is awful. I feel like I still know little about the card's performance. For example, if you just look at the eye candy section or whatever he calls it where he turns the settings (but not resolution) up, it looks like it loses to 7870! Then other parts of the review seem to imply it competes with 7970 (which by default means it will also compete with 670-680, which he ignores completely)

He seems to imply a 249 price too, which I also doubt. But where exact pricing falls whether 299 or what will make a big difference.

Anyway, hopefully it will really push 7870-50 price way down since that's where I'd look to get in.

It'll sell like hotcakes for Nvidia, it's clear to me 90% of GPU buyers just refuse to buy AMD sadly. 670 sold like absolute hotcakes at $400, I doubt Nvidia even needs anything below that price really and they will still dominate the market. With the addition this card I imagine Nvidia will control maybe 90% of sales, as those people who dont want to pay $400 will now have an Nvidia option.
 
249$ would indeed be great for this level of performance but I doubt it as well. I mean two of these babies would obliterate a single 680 while costing less. Not that it hasn't happened before but still...!
 
... note he can use Dirt: Showdown, it will be funny, but not fair for Nvidia
Why not fair? All the new games (such as Nexuiz, Sniper Elite V2, Dirt: Showdown, Ghost Recon Future Soldier) which are used more compute oriented algoritm for some effects then others are not fair? We're going in this direction, so I don't think that any games that will release in the future, and make use of compute shaders more effectively are unfair for NVIDIA. Sure, Kepler won't like these, but this is don't make a comparison unfair.
Dirt: Showdown don't use any unfair code on GeForce. It just use forward+ lighting, which is a very compute oriented algorithm.
NVIDIA made some decision to reduce compute efficiency on Kepler, and not take that route what they started with Fermi. I don't known why NVIDIA did this, but I'm sure that this is not Codemasters fault. They just want to render huge number of point lights effectively with Ego engine. The only way to do this is forward+ lighting.
 
Yeah that review is awful. I feel like I still know little about the card's performance. For example, if you just look at the eye candy section or whatever he calls it where he turns the settings (but not resolution) up, it looks like it loses to 7870! Then other parts of the review seem to imply it competes with 7970 (which by default means it will also compete with 670-680, which he ignores completely)

He seems to imply a 249 price too, which I also doubt. But where exact pricing falls whether 299 or what will make a big difference.

Anyway, hopefully it will really push 7870-50 price way down since that's where I'd look to get in.

It'll sell like hotcakes for Nvidia, it's clear to me 90% of GPU buyers just refuse to buy AMD sadly. 670 sold like absolute hotcakes at $400, I doubt Nvidia even needs anything below that price really and they will still dominate the market. With the addition this card I imagine Nvidia will control maybe 90% of sales, as those people who dont want to pay $400 will now have an Nvidia option.

You are spot on. AMD needs a new strategy on gpu because people are unwilling to pay over $400 on their cards. You can say that the last card to really sell well at that price was the 5870 (which was raised in price to over $400 I believe). The 5870 was something else though.

What AMD should do now is knock the arse out of the midrange market. There is no way Nvidia could compete vs a $150 7850 and $200 7870. AMD has been ramping these cards for ages in comparison, and with smaller dies they could really punish Nvidia at the high end by forcing an instant drop on the 660 Ti. Sure they'd need to drop prices on Tahiti, but if nobody is buying it at current prices where it is a far superior choice, what do they have to lose?

Sadly I get the feeling AMD doesn't even realise the 660 Ti is coming. There is something sadly lacking with them on all fronts just now and if they don't have a proper response for it after such a long delay, they deserve whatever they get.
 
I think it will be better to wait for some serious reviews. With these testing methods (~2,5 years old games, almost no AA, no power numbers, no noise numbers, missing MSRP etc.) even a GeForce FX would appear like a king.
 
960 CCs? If this CC count is true, is it possible that NVIDIA originally planned the GK106 to have 768 CCs but changed it to 960 during development?
 
Come on Nvidia, get it together.

Literally by the time they get out a "true" mid range part, Sea Islands may come out or be very near.

But as I say, given their sales it wont matter much to them. I am sure they are dominating the market with their cheapest card at 400, it will only get worse once they have at least a $300 card out soon. They really dont need any mid range at all.
 
I highly doubt that is the case, there are only a certain minority of graphics card consumers willing to spend US$400+, especially outside the USA.
Or was that comment tongue-in-cheek?

Is Sea Islands going to be a full refresh of the 7000 series, or is it just designed to fill gaps in the lineup like the 6000 series did?
 
Partly offtopic - but sure, Forward+ is apparently compute heavy, but still 670 and 680 are easily beating 580 in it, and 6970 is almost as fast as 570, even though the architecture shouldn't be good in compute (even if it's better than VLIW5)
 
Partly offtopic - but sure, Forward+ is apparently compute heavy, but still 670 and 680 are easily beating 580 in it, and 6970 is almost as fast as 570, even though the architecture shouldn't be good in compute (even if it's better than VLIW5)
I think that's a bit too much of an oversimplification saying the architectures aren't good at compute. The 670/680 and 6970 aren't always bad at compute, some algorithms work reasonably well with such hardware (not necessarily the same ones for these cards). But sure enough GCN and Fermi are better at supporting some classes of compute shaders. I know nothing though about the compute shaders used in Dirt, nor if they are really the reason for the performance we're seeing.
 
Perhaps we ought to question whether there even is a GK106 at all. Has it ever been mentioned by NVIDIA?
 
There is a huge gap between GK107 and GK104. 28nm yields should not be so bad for a 300mm² GK104. So it would be a waste of money to fill this gap with cut-down GK104s, instead of a ~200mm² GK106.
 
There's a huge gap (in a revenue important segment) between CapeVerde and Pitcain, and GK107 is slower than the first, and GK104 faster than the latter...
 
I added up all the newegg user reviews of different cards earlier today (yep, loooser).

I know there are tons of problems with this (so spare me your diatribe on how stupid I am to consider these a indicator of sales, I already know) but I sometimes look at them as a interesting indicator for sales.
Probably the MAIN caveat I'd worry about is I think newegg sometimes just pulls older SKU's. If that SKU had a lot of user reviews, and it's now gone completely from the listings, obviously that could skew the numbers. I did this just cause it was my perception GK104 by itself was probably outselling AMD's whole top to bottom line.

Anyways heres what a I got (number of user reviews, all SKU's):

GTX 690=45
GTX 680=541
GTX 670=786

Total=1372

For AMD:
7970 GE=1
7970=569
7950=218
7870=158
7850=270
7770=240
7750=143

Total=1599

I was kind of surprised AMD actually does have more reviews, even though we are basically comparing 6 main cards (7970, 7950, 7870, 7850, 7770, 7750) to just 2 (680, 670). Of course this is just current gen, I am sure a huge amount of 560 Ti's to name one have been selling throughout.

But yeah, the 660 Ti or whatever at 300 will blow the doors off AMD this gen for good. It will probably make 670's already stellar sales look very bad. If they have enough supply they will sell insane amounts.

At least at newegg, which basically just caters to enthusiasts who buy high end discrete cards.
 
But yeah, the 660 Ti or whatever at 300 will blow the doors off AMD this gen for good. It will probably make 670's already stellar sales look very bad. If they have enough supply they will sell insane amounts.

At least at newegg, which basically just caters to enthusiasts who buy high end discrete cards.

http://www.tweaktown.com/reviews/48..._2gb_reference_video_card_review/index13.html

GTX 660 Ti is crippled by the 192 bit memory controller and 24 ROPs. After spending USD 300 you don't want to play the latest games at anything less than the highest image quality settings. When you turn on MSAA 4x or 8x this card is going to perform like a HD 7870 and not a HD 7950. Moreover the HD 7950 OC cards are very good overclockers with excellent perf scaling and easily run at 1.1 - 1.15 Ghz with voltage tweaking. At those clocks speeds they compete very well with GTX 670 (1250).
In the most demanding games with the highest quality settings including MSAA , GTX 660 Ti has no chance against a HD 7950
 
That tweaktown review is utter shit(Which is no surprise, all of their reviews no matter what they are reviewing are shit) and was the topic for about two pages of mockery if you just read back a page or two.
 
That tweaktown review is utter shit(Which is no surprise, all of their reviews no matter what they are reviewing are shit) and was the topic for about two pages of mockery if you just read back a page or two.

yeah i agree their game test suite is outdated. But in a demanding game like Metro 2033 or Aliens vs Predator with the highest settings and MSAA the GTX 660 Ti was clearly exposed and is no match for the HD 7950.
 
Back
Top