[3dcenter] GK104 specs

[bunnyfubbles]

I find the TMUs to many for 340mm2

I could be wrong

AMD has the same number of TMUs in 352mm^2

 

[lavaheadache]

Like NV set the stage for pricing with the GTX280/260 and then AMD released their cards at prices according to them?
Oh wait, AMD undercut them significantly and cause NV to give rebates to people who purchased the GTX280s on release.

Just because one card is priced at X on launch doesn't mean that sets the market.
The HD5800 series INCREASED in price after launch (although only slightly).

There is no reason to assume that prices will stay static. Even if NV did decide to price according to AMD rather than being aggressive, that doesn't mean AMD wouldn't respond by dropping prices themselves.

The prices are what the prices are NOW. That doesn't mean that in the future everything will revolve around these prices. Never has and probably never will.

It's obvious that they priced the 48xx's the way they did to gain market share and customer confidence. They HAD to after the 2900 debacle. The 8800 reign of supremecy lasted so long and went virtually unanswered for years.

When all things are relatively equal across the board competition will dictate price

 

[maddie]

[/B]

Once you accept what he is and that anything that favors nVidia is positive, he is a fairly straight shooter.

Can this be considered an oxymoron?

 

[lavaheadache]

Can this be considered an oxymoron?

nah, I did say *fairly* straight.

 

[railven]

Can this be considered an oxymoron?

I though it and, my answer is yes and no:

Yes, he will always say pro-nVidia things (thus a straight shooter.)

No, he will always say pro-nVidia things (thus not a straight shooter.)

So, yes, oxymoron, or no depending if you like only hearing pro-nVidia things. I guess...

My head hurts now, haha.

 

[Stingercjc]

Like NV set the stage for pricing with the GTX280/260 and then AMD released their cards at prices according to them?
Oh wait, AMD undercut them significantly and cause NV to give rebates to people who purchased the GTX280s on release.

Just because one card is priced at X on launch doesn't mean that sets the market.
The HD5800 series INCREASED in price after launch (although only slightly).

There is no reason to assume that prices will stay static. Even if NV did decide to price according to AMD rather than being aggressive, that doesn't mean AMD wouldn't respond by dropping prices themselves.

The prices are what the prices are NOW. That doesn't mean that in the future everything will revolve around these prices. Never has and probably never will.

Those were different times in a different situation. AMD saw an opening to gain some market share. They had a product that smoked the competition and knew if it was priced right they could turn a ton of Nvidia users to their side (like Lava said while I was typing this reply - they needed the consumer goodwill). It worked. I replaced my GTX 260 with a 5850 that I just recently upgraded.

This situation is different. Nvidia just has to release a competing product. They can undercut AMD a bit to steal some sales, but they don't need to drop an awesome 300 dollar product that smokes the competition at a price that will make us all rush out to the internets and buy one. I'd LOVE for it to be 300. But in the current landscape of the graphics market, I just don't see it happening.

 

[SirPauly]

How exactly does one garner and gain market share by offering enthusiast price-points?

 

[Grooveriding]

How exactly does one garner and gain market share by offering enthusiast price-points?

By selling the fastest card on the market to enthusiasts.


GOOOOOOOOOOOOOOOOOOOO PREMIUMS.

 

[AnandThenMan]

How exactly does one garner and gain market share by offering enthusiast price-points?

Ask Nvidia.

 

[SirPauly]

By selling the fastest card on the market to enthusiasts.

Interesting logic.

 

[Lonyo]

How exactly does one garner and gain market share by offering enthusiast price-points?

Not as ridiculous as it sounds, actually.


http://www.xbitlabs.com/news/graphi...h_End_Mainstream_Graphics_Cards_Research.html

Notice how the UNIT SALES of $300+ cards is higher than the UNIT SALES of $200-$300 cards.

If you feel a need to complain about the date, feel free to pony up the $10k (yes, TEN THOUSAND DOLLARS) for a subscription to Mercury Research to get more up to date information.
I am just offering a snapshot of how what seems illogical (gaining marketshare by selling high priced cards) may in fact be true.


Other choices of source include Jon Peddie Research. Much more reasonable at only $995 for an issue, or $3500 for a year. They have this to say:
http://www.xbitlabs.com/news/graphi...hics_Cards_Drop_Quarter_over_Quarter_JPR.html

The AIB market is fueled at the high-end by the enthusiast gamer, small in volume (~3 million a year) but high in dollars (average spend for an AIB ~$300). The AIB shipment volume comes from the performance and mainstream segments.

Gamers are the most likely to buy new cards, millions of them in a year, spending an AVERAGE of $300. Once again, that magic $300 number.
Yes, the HD7970 is more expensive than that, but I'm showing that $200 type cards which is where we look for value aren't necessarily always the most important, and having expensive products isn't going to hurt AMD.

 

[SirPauly]

It's ridiculous. Enthusiast prices help with revenue and margins and really doesn't effect market share too much.

 

[Mopetar]

And GK110 will be here by summer as the very, very earliest. It's more probable that cards with it will be on the market on Q3 or Q4 2012.

The earliest reports of GK 104 taping out were in the second half of September. Most are currently estimating that cards will ship in late March or early April. Assuming that GK 110 takes a similar amount of time, we probably won't see any cards until September. Yields could also affect availability and larger cards can be more problematic. Hopefully nVidia has been using their experience with GK 104 to improve yields and get a better design, but I'd guess it will be around the start of October before we see GK 110. I wouldn't mind being wrong as that's a long time to wait, but it seems most likely given historical data.

 

[AtenRa]

Nice, GK110 was taped out. That means it will be ready for Xmas time 2012. :biggrin:

Now lets talk about GK100 launch in April 2012

I find it very hard to believe NV will release a middle end card first, im still betting for High-End (GK100?)first in March/April and Middle-End (GK104?) one/two months later.

 

[sontin]

I find it very hard to believe NV will release a middle end card first, im still betting for High-End (GK100?)first in March/April and Middle-End (GK104?) one/two months later.

I think the same. Why would nVidia launch their performance chip first? Makes no sense. And they need BigK for Quadro and Tesla. Crey want to complete the TITAN system in Q3 2012 and Huang said that Kepler-Quadro will drive their revenue in F-Q2 (May - july).

Look at these GK104 Specs. That's more BigK than a performance chip.

 

[BallaTheFeared]

The memory bandwidth listed in these specs is higher than what a gtx570 has. http://www.nvidia.com/object/product-geforce-gtx-570-us.html

The gtx470's memory bandwidth: http://www.nvidia.com/object/product_geforce_gtx_470_us.html

My bad, I read it as 140MB/s.

That said I'm going to retract my statement anyways, going from 133GB/s to 182GB/s with a 950 core overclock on my 470s in SLI only nets about a 7% performance gain.

I've said that before, but I'm having a hard time figuring out how much the rendering capabilities affect it. I should test again but if I remember correctly in my Crysis 2 tests at stock the same memory overclock at stock core/shader improved fps by just 5%, while the core/shader overclock netted substantial gains and was unaffected by the lack of memory bandwidth for the most part.

 

[blastingcap]

GK104 Pops Up In The Wild

http://semiaccurate.com/2012/02/10/gk104-pops-up-in-the-wild/

"If things go as normal, it takes 4-6 weeks from AIB sampling to cards on the shelves. This would mean late March or early April, just like we have been saying for weeks. The ball is rolling, and the variables are being narrowed quite quickly."

 

[MisterMac]

My god, i want kepler out now and want to see the damn mythical performance numbers.

The whole dropping price of 7970/7950 that 3dcenter suggest seems insane.



How many of you would feel raped in the behind if they drop 150 MSRP cause of kepler?

 

[Crap Daddy]

Here's another source confirming these specs:

http://www.brightsideofnews.com/news...-leak-out.aspx

The interesting part and I've seen this speculated in different forums, is that GK104 will be the GTX680. If this will be confirmed and based on the specs which seem to be right, I am pretty sure this card will be better than the 7970. Can't think that NV will release a GTX*80 that is slower than AMD's flagship.

 

[Arkadrel]

Here's another source confirming these specs:

http://www.brightsideofnews.com/news...-leak-out.aspx

The interesting part and I've seen this speculated in different forums, is that GK104 will be the GTX680. If this will be confirmed and based on the specs which seem to be right, I am pretty sure this card will be better than the 7970. Can't think that NV will release a GTX*80 that is slower than AMD's flagship.

GK104:
256-bit wide GDDR5 memory interface.
1250 MHz actual (5.00 GHz effective) memory, 160 GB/s memory bandwidth.

Nvidia 580:
384-bit wide GDDR5 memory interface.
1002 MHZ actual (4.008 GHz effective) memory, 192.4 GB/s memory bandwidth.


The GK104 will have like ~20% less memory bandwidth than the 580.

If the GK104 was ~20% faster than a 580 (like a 7970) wouldnt it have 20% more bandwidth than a 580 as well? not 20% less?
I think the GK104 will be around 580 levels... so ~20% short or so of the 7970.

I dont think Nvidia calls it the 680..... Im still thinking it ll end up as a 660 or something.
(people that do are assumeing, that nvidia's next card the big one that comes out, late Q3, will be called 780)

 

[BallaTheFeared]

Memory bandwidth doesn't seem to mean a whole lot, I think it's being overstated as far as importance goes.

The really interesting thing was Nvidia is talking about a single monster gpu card to take the place of the dual gpu cards we've been seeing.

2304 CUDA cores/Stream Processors (6GPC)... either 384-bit / 512-bit on the high-end compute side - and GeForce GTX 690, but this time as a single monolithic die, instead of typical mix'n'match of two high-end GPUs.

All speculation of course, but Jesus that would be a monster. Assuming the smaller card competes with the 7970 in anyway.

 

[bunnyfubbles]

GK104:
256-bit wide GDDR5 memory interface.
1250 MHz actual (5.00 GHz effective) memory, 160 GB/s memory bandwidth.

Nvidia 580:
384-bit wide GDDR5 memory interface.
1002 MHZ actual (4.008 GHz effective) memory, 192.4 GB/s memory bandwidth.


The GK104 will have like ~20% less memory bandwidth than the 580.

If the GK104 was ~20% faster than a 580 (like a 7970) wouldnt it have 20% more bandwidth than a 580 as well? not 20% less?
I think the GK104 will be around 580 levels... so ~20% short or so of the 7970.

I dont think Nvidia calls it the 680..... Im still thinking it ll end up as a 660 or something.
(people that do are assumeing, that nvidia's next card the big one that comes out, late Q3, will be called 780)

in memory bandwidth limited scenarios (ie 1440p resolutions and higher) it might very well be limited, but at 1080p it very well might be faster than even the 7970, we'll have to wait and see.

 

[Lonyo]

My god, i want kepler out now and want to see the damn mythical performance numbers.

The whole dropping price of 7970/7950 that 3dcenter suggest seems insane.



How many of you would feel raped in the behind if they drop 150 MSRP cause of kepler?

Since most people who have bought 7900's on this forum (at least) know that they are likely to go down in price when NV finally releases something, it's part of the expectation we had when we bought them.

You can't feel raped if you knew it was eventually going to come, it's just a matter of timing.

When people bought the GTX280 and then AMD counterpunched with much lower prices, that's when you could feel raped in the behind. Luckily NV responded by offering rebates for early adopters so that they didn't feel like that.

This time round, there's no fast reply from NV, and no known pricing or timeline. We know that NV will eventually make something, but that potential $150 hit is the price you pay for having the card now instead of waiting.
It's the same as with buying any one card when there's no competition. The hit is always likely to come in terms of price adjustment, you just don't know how much or when, so there's no feeling of rape when there's a significant (measured in months not days or weeks) delay between launches.

 

[BallaTheFeared]

According to these speculation leaks GK104 is in circulation for testing, it can't be too long before it appears.

I just have a hard time believing Nvidia would release a x80 card with a 256 bit bus, even if they won't have anything bigger coming out soon it would look foolish in the high res benchmarks most likely vs the 7970, but might be competitive with it at lower res'es such as 1200p and 1080p which are by far the most common resolutions currently.

 

[Crap Daddy]

GK104 is, based on these specs, basically 2xGF104 so you can estimate the power of this chip, its has the same ROPs and TMUs like Tahiti XP, only thing lacking is the mem bandwidth.

Some comparisons in bandwidth: GTX570 has 152 GB/s and HD6950 has 160GB/s while being on average 10% faster and the GTX560 Ti has 128 GB/s vs. HD6950 160 GB/s, it's a 25% difference for around the same performance.

I am pretty sure that the guys at NV know what they're doing, maybe over 1900x1200 this lower mem bandwidth will be a problem but up until this resolution I think GK104 will be better than 7970 at stock clocks and the usual gaming benchmarks.

These are only supositions and I hope we'll get the answers sooner than expected.