Kepler predictions, just for fun.

[bryanW1995]

680=7970+20% $600
670=7970 $450

1. Released in June or July
or
2. Released in May


660=GTX 580 $350

1. Released in April or May
or
2. Released in June or July


Mid range will cost more this round, no $200 460 type card, as nv will price accordingly to amd for profit not undercutting.

I hope you're wrong, but NV would be smart to read the tea leaves. Read has basically said that he's more interested in profits than price wars. I fear that you're right, and we're in for higher prices going forward. I might end up keeping this gtx 480 a bit longer than anticipated...

 

[Madcatatlas]

Predictions predictions and predictions.


Nvidia might do another 480 to 580 launch. A 680 with gimped hardware doing 10% better than a 1000mhz 7970. Fixing that 8 months later with a fully unlocked 780. This time around, AMD will be ready with a 8970 that will equal and slightly beat the geforce.

 

[bryanW1995]

TY Lava 😉

I think the stock GK104 will have the performance of GTX560Ti SLI. GTX660Ti/760Ti

That is of course if the shader rumor turns out true (768). If it's that 576 shader rumor, then I'd say about 10% faster than GTX580. Depending on clocks of course.

This is just going by "Fermi" architecture and not accounting for the new Kepler arch.

It's going to overclock to Mars.

Why do you say this? Judging by how poorly NV did last time there was a die shrink, I lack any confidence whatsoever in their ability to manage thermals this time around. They really didn't rifine the process and come out with something decent until gtx 460. What was that, 10 months after 5870? Are we really going to have to wait that long again? I probably will b/c I'm pissed off at the likely sky-high prices in our immediate future (and I game at 1680x1050), but how many NV loyalists will jump ship if gtx 680 launches in august?

 

[Grooveriding]

Predictions predictions and predictions.


Nvidia might do another 480 to 580 launch. A 680 with gimped hardware doing 10% better than a 1000mhz 7970. Fixing that 8 months later with a fully unlocked 780. This time around, AMD will be ready with a 8970 that will equal and slightly beat the geforce.

This is the $64,000 question with how this node will turn out. We've all seen how Tahiti performs at 1200 core. It gives the kind of boost over a 580 that you'd expect to see nvidia's successor to that card deliver, sometimes even more, with the gains ranging from 50% to 80% over a 580. All that while thermals and power draw still being in the realm of reasonable for a high end gpu.

If the 7970 had released with those clocks people would of been going ga ga over it. So we know the potential is there and the performance is there. The question is how long until and exactly what will AMD do with the refresh. Will it just be a big overclock as stock clocks - likely the easiest, or will they a release a chip with some tweaks, more shaders ? and bigger clocks. 28nm is just in its infancy and we're still hearing TSMC is having issues and some of their customers cannot even get workin chips out of it yet, so there is more room to improve there. AMD's advantage will be clockspeeds imo with their smaller dies on their high end cards keeping thermals sane. Nvidia will have more trouble with clocks I think because they will do the huge chip jammed full of cuda cores.

So will clocks be enough to trump die size ? Who knows ?

Nvidia cannot be happy about their current gaming gpu situation though, with not having any 28nm entries yet or new skus to pr about while their competition does. Not to mention their highest margin part, the 580, being rendered obsolete by their competition. Every day they go without a high end 28nm gpu is another day another consumer buys a 7950 or 7970 instead of an nvidia card. Worst case, a loyal nvidia customer gets tired of seeing the 7970 benches kicking their 580's ass and decides to get a 7970, and finds himself happy in the red camp. Customer lost.

I'm sure they're trying to get something out as fast as possible, but I doubt they want to rush out with whatever they have ready again and get the bad flak for repeating fermi april 2010.

 

[Genx87]

Will match 590 performance in a single GPU config within a 250 watt envelope.

 

[dust]

I'm pretty sure the gtx680 will have equal performance to the new GTX680.()🙂

 

[ctk1981]

Predictions.....people will whine about its price or performance regardless.

 

[IlllI]

it will have a power draw of over 9000

 

[Carmen813]

Im thinking we will see Nvidia take over the price/perfomance crown.

 

[sandorski]

25% performance over 7970, but around current 7970 Prices.

 

[gladiatorua]

Im thinking we will see Nvidia take over the price/perfomance crown.

They might. But how long will they be able to hold it?

 

[Borealis7]

it will be released on Dec 19th, 2012. the first review that comes up on the web will bring about the end of the world on the next day.

 

[tviceman]

it will have a power draw of over 9000

1.21 GIGAWATS

 

[railven]

I just read the other thread - I'm not interested in the GTX x80 unless I can get a GTX 580 for current GTX 570 prices.

Basically, when should I expect the GTX 660/670 (or whichever one would equal GTX 580 performance)? I still need to upgrade a GTX 460 SC - badly! I was hoping the HD 7970 would start a price war, but that didn't work out so waiting for GeForce successor hopefully does.

 

[Hallowed_Dragon]

I think Kepler will be able to cook me dinner and keep me warm 😀

 

[BenSkywalker]

If we ignored the AMD launch altogether, I would reasonably expect nV's x60 parts to perform on par with the 580 for ~$200-$250, the x70 to be in the 50%-60% faster range and the x80 to be in the 75%-100% faster range. I had expected pricing to be in line, give or take a bit, with this generation.

That is simply based on the reality of Moore's law and it's actual premise being seen this generation(a roughly doubling of transistors in the same die space). Unlike CPUs, graphics scale almost perfectly with proportional increases in available resources(yes, edge cases like multi displays using high levels of AA will be bandwidth limited).

Since AMD has launched, I expect the pricing to be scaled up significantly and the relative performance dropped quite a bit(lower target clocks, higher yields). It's a fairly reasonable assumption that they are going to have a rather significant edge over what AMD has shown based simply on the realities of progress in this industry, but market realities and business sense will almost certainly having them lowering all of their targets unless they think they can 'take them out'(not likely at all).

 

[lavaheadache]

If we ignored the AMD launch altogether, I would reasonably expect nV's x60 parts to perform on par with the 580 for ~$200-$250, the x70 to be in the 50%-60% faster range and the x80 to be in the 75%-100% faster range. I had expected pricing to be in line, give or take a bit, with this generation.

That is simply based on the reality of Moore's law and it's actual premise being seen this generation(a roughly doubling of transistors in the same die space). Unlike CPUs, graphics scale almost perfectly with proportional increases in available resources(yes, edge cases like multi displays using high levels of AA will be bandwidth limited).

Since AMD has launched, I expect the pricing to be scaled up significantly and the relative performance dropped quite a bit(lower target clocks, higher yields). It's a fairly reasonable assumption that they are going to have a rather significant edge over what AMD has shown based simply on the realities of progress in this industry, but market realities and business sense will almost certainly having them lowering all of their targets unless they think they can 'take them out'(not likely at all).

By the same metric why wasn't the 480 75-100% faster than the 285 at release?

 

[Grooveriding]

By the same metric why wasn't the 480 75-100% faster than the 285 at release?

Because it wasn't, and it still isn't, apart from a few outliers. There have not been gains like that seen since the 8800GTX or 9700PRO.

Not that gamers have ever stopped wishing for it to happen again, but it doesn't look like it's going to happen for us ever again.

 

[lavaheadache]

Because it wasn't, and it still isn't, apart from a few outliers. There have not been gains like that seen since the 8800GTX or 9700PRO.

Not that gamers have ever stopped wishing for it to happen again, but it doesn't look like it's going to happen for us ever again.

I wasn't really looking for an explaination nor do I think it is going to happen. Just trying to understand why benskywalker thought that Kepler would net see those kinds of gains.

 

[OILFIELDTRASH]

680 will be 40-50% faster than 7970 and will overclock very well. Price will be $599 and they'll sell everyone they make for 6 months out. Expect to see some going for $800+ on eBay and you'll need to get lucky to get one on newegg or amazon. 7970 prices will be dropped to sub $500 levels with some used examples being had for $350ish so they can get teh Kepler.

 

[SirPauly]

By the same metric why wasn't the 480 75-100% faster than the 285 at release?

Because it wasn't a fully enabled core and nVidia had execution problems but based on Computerbase.de's findings:

63-84 percent faster with 1920 x 1080 with x4 and x8 AA.

64-133 percent faster with 2560 x 1600 with x4 AA and x8 AA.

http://www.computerbase.de/artikel/...480/21/#abschnitt_performancerating_qualitaet

 

[Rvenger]

680 will be 40-50% faster than 7970 and will overclock very well. Price will be $599 and they'll sell everyone they make for 6 months out. Expect to see some going for $800+ on eBay and you'll need to get lucky to get one on newegg or amazon. 7970 prices will be dropped to sub $500 levels with some used examples being had for $350ish so they can get teh Kepler.

lol, I hope not. Then again, I could get a good deal on a 7970.

 

[exar333]

My guess - It'll overtake the 7970 by 30%. AMD will respond with a 1.4GHz tweaked 89xx part. Again, this is just a guess based on absolutely nothing. 🙂

This is interesting, because then the question is really how good is Kepler at OCing?

If it launches at 30% better than the 7970 (which can OC almost 30% easily) and only can OC 10%, it's not a great buy unless it is essentially ~ same price. If Kepler, on the other hand, can ALSO OC 30% more, then it is much more interesting when comparing to the 7970. Time will tell...

 

[boxleitnerb]

Midrange:
GK104, HD7950+10%, needs 15% less power

Highend:
GK110, HD7970+30%, needs a tad more power

For both, I'm refering to the top part of each SKU. My 2 cents 🙂

 

[sontin]

Because it wasn't, and it still isn't, apart from a few outliers. There have not been gains like that seen since the 8800GTX or 9700PRO.

Not that gamers have ever stopped wishing for it to happen again, but it doesn't look like it's going to happen for us ever again.

GTX480 is/was much faster than the GTX285. Average 75% in 1080p with 4AA: http://www.hardware.fr/articles/787-23/recapitulatif-performances.html