TechPowerup - Nvidia Kepler GK104 PCB Drawings and power connector pics

[railven]

Or:
GK104 is BigK. Look at the rumored specs. Would you believe that such a chip would be only midrange?

Nothing is stopping them from doing just that. Bot companies have thrown us curve balls the last two generations in terms of naming/tiering, and I wouldn't be surprised.

 

[blckgrffn]

Of course I believe that. It's called "progress". Did the GTX460 cost as much as the GTX285 initially, even after the price cut? No - it was far cheaper. New midrange beats old highend, nothing new here.

They also had the GTX 470 and GTX 480 on the ground and running when the GTX 460 launched.

If it is comparable to part that AMD is selling for ~$500, it seems ridiculous that nvidia would leave that margin on the table. They aren't your friends or anything....

 

[railven]

Of course I believe that. It's called "progress". Did the GTX460 cost as much as the GTX285 initially, even after the price cut? No - it was far cheaper. New midrange beats old highend, nothing new here.

Bro, have you not realized how things are shifting around in the world? We're paying MORE for stuff nowadays. Why do people act like GPUs are free from the economic crysis P) the rest of the world is seeing?

We've already read about the yield issues, the supposed cost per waffer increase, and magically you still expect things to be linear?

Progress doesn't come cheap.

 

[railven]

They also had the GTX 470 and GTX 480 on the ground and running when the GTX 460 launched.

If it is comparable to part that AMD is selling for ~$500, it seems ridiculous that nvidia would leave that margin on the table. They aren't your friends or anything....

Every time I read this, a little part of me dies inside

 

[railven]

Crap, guess I'll have to start adding Go Premiums to nVidia related too haha.

Go Premiums!

 

[Grooveriding]

GK104 is a 256-bit upper-mid-range Kepler part though. It shouldn't technically beat HD7970 based on its ranking in the NV hierarchy. Basically if you take away the branding (they can call it GTX780 for all we care, etc.), this card is a direct replacement of a GTX560Ti as I have understood it. All NVidia really needs to do is launch it at $399 with performance between HD7950 and HD7970.

I can't see GK104 having any chance outperforming an HD7970 in overclocked vs. overclocked states considering it's supposedly a 256-bit card and given the immense overclocking headroom that HD7970 has.

Your comment about it performing similarly to HD7970 given the die size? I am not sure: 256-bit vs. 384-bit. Also, NV likely has extra transistors allocated for things needed in Quadro, Tesla markets, PhysX. It's hard to imagine an NV chip being more efficient per die size.

If hypothetically this upper-mid-range card can actually beat HD7970 in DX11 games such as Batman, Crysis 2, BF3, Metro 2033, I don't know how NV can price it at $450+, since in the end it's still their mid-range card . . . unless the new mid-range is going to become $400-450 battleground?

I was trying to be diplomatic and not induce flaming. Nvidia claimed they were improving perf/watt with this gen. I am looking forward to seeing how it does, because will see how Nvidia does against AMD at the same die area. The bus width is an interesting thought though.

I know historically nvidia performs worse than AMD at the same die size, see 460 vs 5870. They do know what they are aiming for though and AMD may have clocked the 7970 too conservatively. Nv could clock this card highly and give up much OC headroom but get closer to a stock 7970 in benches.

Also the tri sli connector says to me this is not your 460/560 class card. My honest guess is slower than a 7970 by 10%, $400 and only faster in batman and on psr in BF3. I think the tessellation advantage is over, unless we see games using even more in the future. I do not think it will gave the OC headroom of Tahiti, see earlier comment.

I really want to get some info when reviews come for this card on how delayed their real high end gk110 is and when we will see it. I want to see more performance, not more of the same with a different colour.

 

[f1sherman]

Of course I believe that. It's called "progress". Did the GTX460 cost as much as the GTX285 initially, even after the price cut? No - it was far cheaper. New midrange beats old highend, nothing new here.

It was ~ $240 for custom/overclocked/beefed up circuitry
GTX 460 from EVGA/MSI/GIGA, which blew GTX 285 away.

 

[OILFIELDTRASH]

Bro, have you not realized how things are shifting around in the world? We're paying MORE for stuff nowadays. Why do people act like GPUs are free from the economic crysis P) the rest of the world is seeing?

We've already read about the yield issues, the supposed cost per waffer increase, and magically you still expect things to be linear?

Progress doesn't come cheap.

Wafer costs have increased 15%. Say wafer costs were 100% of what it takes to build a gpu. A 20% price increase would cover the 15% increased cost of wafers and an extra 5% profit just cause they can. AMD has raised the msrp by 50%. Not being confrontational but could you explain to me the 50% increase?

And to take it even a step further wafer costs are not 100% of the costs of gpu's. Being generous they might be 20%. So saying wafer costs are a big factor is not a valid argument. Add $10 to the cost of each card and you more than cover the wafer increase.

 

[RussianSensation]

Or:
GK104 is BigK. Look at the rumored specs. Would you believe that such a chip would be only midrange?

I don't think so. 256-bit for a high-end card with 340-350 mm^2 die size from NV? Not buying it. Unless NV had a complete turnaround of strategy, it will have a 450-500mm^2 GPU to follow, even if it means Q3 2012.

Most HD7970s VRAM runs out at 7000mhz. That means to provide some room for higher yields, at most the shipping VRAM speeds of a modern card are likely to be 6700mhz. Let's assume NV went out all and put 6700mhz VRAM (doubtful). @ 256-bit that's only 214GB of memory bandwidth. That's barely better than GTX580 and is a paltry improvement for a new high-end chip over their previous high-end GTX580.

There is no way GK104 is Big Kepler. It's specs are in-line with NV's new upper-mid-range card: Similar or higher performance than previous high-end with lower power consumption.

I think people are just confused because HD7900 brought such a small performance increase at stock speeds that they can't grasp that GTX580/HD7950 level of performance is what next generation upper-mid-range should be in the first place. GTX460 was a $229 card and it smacked a GTX285 all over the place in modern games. GTX470 @ $350 handily beat the GTX285.

I think NV initially meant for GK104 to be aGTX660Ti, but given what AMD brought to the table, NV can easily position it as a GTX670Ti now.

 

[Elganja]

Yeah the shroud must look funky to accommodate that plug. Don't like the plug, will be a nuisance for situations where you can make the card occupy a single slot, such as water cooling.

What does it portend for the real large die Kepler though ? 4x6pin ? 2x8 & 1x6 ? 1x8 & 2x6 ? It's looks like they are prepping for a card that will need more than 2 PCIE power plugs at some point.

same... this sucks so hard.

gonna pretty much force me into sli instead of tri-sli

 

[railven]

Wafer costs have increased 15%. Say wafer costs were 100% of what it takes to build a gpu. A 20% price increase would cover the 15% increased cost of wafers and an extra 5% profit just cause they can. AMD has raised the msrp by 50%. Not being confrontational but could you explain to me the 50% increase?

And to take it even a step further wafer costs are not 100% of the costs of gpu's. Being generous they might be 20%. So saying wafer costs are a big factor is not a valid argument. Add $10 to the cost of each card and you more than cover the wafer increase.

Are we going down this road again? Look back at PC history, just take a peek, you'll find your answer.

Now answer me this question: Do you expect a company to undersell it's competitor with a superior product in all measurable metrics?

If you answer yes, then nothing anyone can ever say will satisfy you and it would be pointless to even discuss the subject with you.

If you say no, then you got your answer.

 

[tviceman]

I measured 30 pixel lengths for the 12mm side of the memory chip. 47x45 pixels is the size of the die. That gives 340mm2 just as Charlie said.

Is this the actual chip itself that is being measured, or is it the HSF that the chip is contained in that you are measuring? If it's the HSF, then it's going to be slightly smaller than whatever "accurate" measurements are being taken now.

 

[boxleitnerb]

Is this the actual chip itself that is being measured, or is it the HSF that the chip is contained in that you are measuring? If it's the HSF, then it's going to be slightly smaller than whatever "accurate" measurements are being taken now.

Don't know honestly.

 

[OILFIELDTRASH]

Are we going down this road again? Look back at PC history, just take a peek, you'll find your answer.

Now answer me this question: Do you expect a company to undersell it's competitor with a superior product in all measurable metrics?

If you answer yes, then nothing anyone can ever say will satisfy you and it would be pointless to even discuss the subject with you.

If you say no, then you got your answer.

Pc history? lol. Seriously that's your argument? Pc history has shown we get more performance for less money. Look it up sometime.

Like how you abandoned your wafer costs increase theory that quickly.

And yes when it comes to pc hardware I do expect to get more performance for less money. It's been that way since the beginning of PC's. Same with smartphones. I'm not paying more for a 3GS today than I was 2 years ago

 

[tviceman]

The only 9800 that did was an asus custom. Anyway this is a board shot, not a price slide, wait till it is out and see uf you get it for $300... No need for anyone to agree with your opinion/hope for a discussion to continue.

http://www.geforce.com/Hardware/GPUs/geforce-9800-gtx/features
http://www.anandtech.com/show/2492

Looks like all 9800 GTX cards had 3 way SLI.

 

[RussianSensation]

Wouldn't it only be their "midrange" card after they have launched a higher level SKU?

So if NV launched GTX460 before GTX480, GTX460 wouldn't be a mid-range of Fermi generation? The order in which NV decides to launch its mid-range vs. high-end card doesn't change those cards respective standings to each other within that generation. It can simply be a matter of launching into the far more volume-driven market segment first while working out the kinks on the Big K. I obviously don't know for a fact. It's possible NV might call this a GTX680 like I said. But then if in Q3 2012 they launch the real Big K, and rebrand it as GTX780, that's just playing with names. At the end it still would be accurate to call GK104 an upper-midrange of Kepler generation with GTX780 the high-end.

Nvidia claimed they were improving perf/watt with this gen.

Technically speaking, if you deliver GTX580 performance on 28nm, you will automatically improve performance/watt. Improving performance/watt vs. GTX580 does not in any way guarantee better performance/watt than Tahiti XT in games. In other words, releasing a 500-550mm2 Big Kepler could produce 50-60% more performance than GTX580, and still improve performance/watt vs. Fermi despite the large die size. A lot of that can come from a more efficient architecture.

I am not ruling out GK104 @ $399-449. At those prices, it still would be competitive with the HD7950. However, all that means is their real high end card just might be a $650 monster. If GK104 is NV's new high-end for this entire generation, I will be hugely disappointed. But why would NV's high-end chip be only a 256-bit card with 350mm^2 die? I know you don't really believe that for a second

 

[OILFIELDTRASH]

So if NV launched GTX460 before GTX480, GTX460 wouldn't be a mid-range of Fermi generation? The order of which NV decides to launch its mid-range vs. high-end card doesn't change its standing in that generation.

They like the high pricing scheme. I can't for the life of me figure it out.

 

[Grooveriding]

So if NV launched GTX460 before GTX480, GTX460 wouldn't be a mid-range of Fermi generation? The order in which NV decides to launch its mid-range vs. high-end card doesn't change those cards respective standings to each other within that generation. It can simply be a matter of launching into the far more volume-driven market segment first while working out the kinks on the Big K. I obviously don't know for a fact. It's possible NV might call this a GTX680 like I said. But then if in Q3 2012 they launch the real Big K, and rebrand it as GTX780, that's just playing with names. At the end that still doesn't change the fact that GK104 = upper-mid-range and Big K is still not here.

This is what I said before. It is as if they hit the same issues they did before with 40nm fermi and instead of rush something out, they put the 460 out, never released the 480 and went straight to 580.

We have heard a bunch of sources saying this will be a'670' moniker, so that leaves the 680 handle open to the big die later this year.

 

[boxleitnerb]

If it were Nvidia they would be crying foul, greedy and what not. If it's AMD it's okay - they need the money, they had good prices for a long time so now they've "earned" it to be greedy

 

[f1sherman]

Tad smaller then GF104... 320-330mm2

MSRP guesstimate: $249-$349.
High demand will initially pull it closer to $300-$450.

 

[railven]

Pc history? lol. Seriously that's your argument? Pc history has shown we get more performance for less money. Look it up sometime.

Wait:
Halo card - follows trend, more performance for more money
2nd top card - follows trend, more performance for less money.

So, what's changed?

Like how you abandoned your wafer costs increase theory that quickly.

When did I abandon it? Wafer's was just one example of the cost increase in production. With the cost of oil and resources going up - oh wait nevermind, this doesn't really matter to you.

And yes when it comes to pc hardware I do expect to get more performance for less money. It's been that way since the beginning of PC's. Same with smartphones. I'm not paying more for a 3GS today than I was 2 years ago

And that still applies, see my top example. Hey, competition - does a wonderful thing eh? Hopefully we get some, soon, so we can start benefiting from these two giants trying to screw each other.

Sadly, price fixing seems to be happening. /shrug What can I saw - don't like it, don't buy it.

 

[frostedflakes]

Weird design for the power connectors, means it can take up to 2x6-pin+1x8-pin. I wonder if the space for the additional 6-pin is for superclocked versions so companies like EVGA can release heavily overclocked cards without having to worry about power delivery or deviating from the reference PCB design. Wouldn't be surprised if overclocking headroom is nuts for Kepler just like it was for AMD's 28nm products, nVidia's strategy might include a large emphasis on factory overclocked cards, kind of like what they did with the GTX 460.

 

[railven]

So if NV launched GTX460 before GTX480, GTX460 wouldn't be a mid-range of Fermi generation? The order in which NV decides to launch its mid-range vs. high-end card doesn't change those cards respective standings to each other within that generation. It can simply be a matter of launching into the far more volume-driven market segment first while working out the kinks on the Big K. I obviously don't know for a fact. It's possible NV might call this a GTX680 like I said. But then if in Q3 2012 they launch the real Big K, and rebrand it as GTX780, that's just playing with names. At the end it still would be accurate to call GK104 an upper-midrange of Kepler generation with GTX780 the high-end.

Woah, what? If that bold applies then it should have applied to GTX 4/GTX5 series.

This is why I've always argued price should be the limiter for which tier is which. They can launch whatever name they want at whatever price, and in the end, it should only be affected by their price (in my opinion.)

If they launch a GTX 780 - that will be their flagship card for that generation, and if they launch this card as GTX 680 - that would be their flagship card for this generation. Not of this "exceptions" shit. If AMD did something similar I'd argue the same thing.

Unless nVidia is changing the tier names (ie a GTX 690 would be top tier single GPU) we can only assume they are shifting names, and most likely with that possibly prices.

Arguing just because this is the GK104 means nothing, haven't you guys learned that yet?

 

[Skurge]

Stupid PCI-E plug arrangement.

Anyone who thinks GK104 is nv's highend is delusional. Why would they stop making 500mm2 die's when it's served them so well? We can see that the 7970 is not the best that AMD can do and GK104 is not the best nvidia can do. I expect GK104 to compete with 7950. and GK100/110 to compete with The 8970. Cause it's coming on Q3 and AMD doesn't have a refresh ready, I'll very surprised.

 

[boxleitnerb]

Maybe it's just an engineering sample card for testing purposes, thus the odd connectors (power and the other white ones) and the odd layout.