TechPowerup - Nvidia Kepler GK104 PCB Drawings and power connector pics

[railven]

Sorry had stuff to do, hopefully I can get this out before I gots to go!

Exactly. Had NV launched GTX460 back then and called it GTX480 and called GTX480 == GTX580, GTX460 ("GTX480 as you want to call it hypothetically) is still mid-range. The only thing that matters is HOW much GTX460 costs and HOW fast it is relative to the fastest card in that chain. You are fixating way too much on the name of the SKU or its launch date.

Using your logic, since HD7770 launched at $159, it's AMD's new mid-range card at the moment since the next level up is a $460 HD7950. If HD7970 was launched 3 months later after HD7770 launched, you'd say that's AMD's new high-end HD7000 series card (until HD7970 is released)?

GTX460 is a mid-sized GPU, with performance roughly equal or better than previous NV high-end GTX280/285. That's in-line with NV's strategy of mid-range GPU. GTX460 is by all definitions a mid-range product and its pricing reflected that as well.

To your blue, no I wouldn't call HD 7770 mid, I'd call it mainstream, and if HD 78xx comes in for the $200-300 price point I'd call that Mid.

Anyways, I tried not to include AMD in my responses since I'm only focusing on nVidia and their stack.

Again, you are assuming there will be a next card, and I understand why, but my question to you is now - if the card launches at this moment with no announcement of a successor, would you still call it mid-range? It would be at the top of the nV line up regardless what is planned in the future.

Not speaking in hindsight. Until you realize that NV has delivered a new upper-mid-range card with performance similar to previous high-end card, this discussion is going nowhere. I don't care if they call it GTX640, 670Ti or GTX880. Price and performance relative to other cards within that generation from NV will determine its own standing. If NV released a new upper-mid-range chip that beats GTX580 (and that automatically means HD7950), that's just what's expected in the first place based on NV's performance improvements going back GeForce 3. NV developed Kepler for 2-3 years. It's not their fault HD7950 is barely faster than Fermi. They probably expected HD7950 to be at least 30% faster than GTX580. Whose fault is that?

I get that, and I'm the one trying to ask you - would you accept it if nVidia approached it differently. IE, what if this GK104 was their top tier until later (or just top tier for gaming.) I'm only asking you if you'd still call it mid-range, if - not even sure why you can't just accept my rhetorical scenario - that is the only card in the pipeline (ie no BigK.)

Because it seems you'd always just argue there is something bigger on the horizon, which again I can understand why you'd argue it, but you aren't even taking yourself out of your position to address mine.

Why would it? It provides similar performance to HD7950. HD7950s cost $460. You can find GTX580 for $430. 5% performance delta. NV is playing the same game AMD is. If consumers are willing to pay, we will price it as such. I think GTX580 and HD7950 are both overpriced because by now we should have had 30-50% more performance over GTX580 @ $500. How is HD7950 doing on that front? Oh right.

I dunno, I don't follow market trends. I'm just a consumer. I buy what I like at a certain price point and move on. I'm not the ones in forums arguing die sizes and marketshare.

Either a significantly improved previous design (i.e., 4870 --> 5870) or a brand new architecture, in either case accompanied by a performance increase of at least 40-50% above the previous best high-end card. This has been true going back to Radeon 8500 at least. The only exceptions are rebrands such as 9800 series or the flop that was HD2900XT.

So again, as in historically, if you haven't noticed - they haven't been doing that lately. I personally think both are taking different new approaches and the ideologies of days gone are going to show cracks.

Thus my even asking - would you still call it mid-range if it was the only top single GPU card announced by nVidia.

If GK104 is NV's fastest card that generation (regardless what AMD brings), that's NV's high-end chip.

If GK104 is 50% faster than HD7970 but NV has GK110 that's 50% faster than GK104, GK104 is a mid-range NV chip.

It's not how it does relative to AMD, but how it does relative to its own generation. If GK104 beats HD7950, is it a high-end 28nm Kepler chip? It can still be a mid-range chip in NV's stack if GTX690 blows it by another 30-50%. All it means is AMD didn't bring enough to the table, so they have no business calling HD7950 a high-end chip within the context of a 28nm generation. And therefore, it has no business being priced at $450 since it's only giving us upper-mid-range level of performance based on what's expected from a brand new architecture + node shrink.

Don't blame me for moving the goal posts that define a new generation. I am not moving anything. It has been that way for NV for 10 years. Go blame AMD for releasing a next generation line-up that barely beats last gen 40nm cards without 30-40% overclocks.

At the end, all that matters is GK104's performance and price.

Sometimes I wonder if you just talk to yourself. I specifically asked in nVIdia's stack. My question has no bearing on what AMD does with their line up - simply nVidias. And this isn't even a question about who's cards are better or who's team you root for (clearly you are so bias you can't even entertain my scenario.)

The question is simply, yet you keep not answering it based on how I'm asking it and keep introducing fluff to this post that is now pages long (haha), so let me try again:

Conditions:
Ignore AMD - they are irrelevant to thsi question
Ignore what nVidia has done in the past - yes I get this is hard to do, the past predicts the future

If, here is the hyopthetical, GK104 is the only card based and annoucned by NVidia for the current time, they launch it with the GTX 680 name, relative to nVidia's line up, would you still call it a mid-range card?

(ie the line up would be dual GPU 690/GTX680 (GK104)/GTX 670/and so forth)

Would you still call it mid-range?

 

[Concillian]

Consumers don't care about die sizes, unless die sizes negatively affect the end product such as having negative consequences that may include high power consumption, or delayed product due to lower yields as a result of a large die size, etc. Once the product is on the shelf, they look at the price, if their PSU can handle it, how much performance they need, what features they want, etc. No one goes up and asks: "Excuse me sir, does this HD7970 have a smaller die size than a GK104 or not?"

Regardless, your die comparison should be thrown out the window since NV has been dominating discrete GPU market share since GeForce 8. Just more evidence that most consumers don't care about die sizes in the end product.

Die size doesn't matter, features and performance at a given price level matter. This is a moving target, since even if performance is the same on a given card, the price is not usually all that steady.

By the same token, market share is a poor metric of performance as well. the general consumer knows nothing about performance of a given GPU, as witnessed by the number of people I knew with Geforce FX 5200 and MX200 / 400 cards, as well as people with 1GB cards with a 64bit memory bus in an era of 256 MB cards. People buy what they think is good, and nVidia branding does a better job of hooking the uninformed consumers.

People here are not uninformed consumers. Some are irrationally loyal to a brand, some are paid to be loyal to a brand (paid with deep discounts or free hardware) but most are not uninformed.

Which brand is better also depends on the price range you're discussing. In the 6xxx / 5xx generation, there was some debate at the middle price range, but at the lower price range, there was little debate which brand was a better value. The other brand held the absolute performance crown... If you were willing to shell out the steep markup.

Most generations have had these kinds of exchanging blows depending on price range. I think this generation will be no different. The 7750 / 7770 are already dropping in price, and I expect AMD will price them to be unbeatable at the low end of the spectrum. nVidia has a brand rep to maintain, and will continue to flex that muscle against the little guy by pricing the low end higher than AMD. The upper end... eventually... will likely be won by nVidia with a significantly larger die than the 7970. And the midrange is again going to be the true battleground. How it's shaping up is pretty clear.

 

[GaiaHunter]

This is one of the most unintelligent posts I have ever read on AT.

Sure it is.

Because die size isn't related with the amount of things you can cram on a chip, and that amount of stuff isn't related with the performance you get at all.

That is why chips don't keep getting bigger until a point you need to shrink the transistors so you can pack more of them and have a lower power budget.

No, the increase in performance hasn't been related to increase in transistor counts which is related to die size/process at all.

 

[GaiaHunter]

Even then, node shrinks allow more transistors to fit into a given die size and different 28nm node manufacturers can fit different amounts of transistors. Different architectural designs can also dictate how many transistors can fit even in the same node.

This is what I was talking about.

Assuming both companies are similarly skilled, one should expect similar performance at same die sizes.

Before the larger NVIDIA die size for same performance could be explained by higher GPGPU performance and the lower density due to higher shader clocks.

Assuming NVIDIA didn't abandon the GPGPU strategy on their non-high end, similar performance at similar transistor budget is the most sensible expectation.

My point was that one can't just compare GPUs to decide each architecture is better based on top card vs top card, if a card is 30-50% larger die size/transistor.

For example the 7970 has 30% higher transistor count than a GTX580, so its performance increase is in line.

From the rest of stuff, that price and price/performance is important, I agree.

The rest isn't addressing my point so I'll ignore it since it isn't relevant to it.

 

[sontin]

Hm, besides the smalltalk, here is little K:
http://we.pcinlife.com/forum.php?mod=viewthread&tid=1844549&page=1#pid33452412

384 Cores, twice the compute performance (FLOPs/s) over GF108. That should going down or up to GF114:
768 Core, 2x compute performance of GK106 over GF116
1536 Core, 2x compute performance of GK104 over GF114

And BigK?!
I think:
2304 Cores, 2,5x+ compute performance over GF110.

Looks great so far.

 

[RussianSensation]

IE, what if this GK104 was their top tier until later (or just top tier for gaming.) I'm only asking you if you'd still call it mid-range, if - not even sure why you can't just accept my rhetorical scenario - that is the only card in the pipeline (ie no BigK.)

I have a hard time accepting it for a few reasons:

1) 256-bit memory width for their highest end single-GPU
2) Based on the huge performance leaps NV has made in the past with each generation and the fact that NV desperately wants the single GPU-performance crown each generation -> unless you believe a 256-bit 350mm^2 die GPU from NV will be 20% faster than HD7970....

Let's just wait and see how it performs.

Because it seems you'd always just argue there is something bigger on the horizon, which again I can understand why you'd argue it, but you aren't even taking yourself out of your position to address mine.

The onus is on you to show why it's reasonable that NV's highest end Kepler card would only be a 350mm^2 256-bit offering, not on me to prove why that seems hard to believe considering in the last 5 years NV hasn't made any high-end GPU with such a small die. It seems more reasonable than not that GK104 is just an upper-mid-range card from NV that was never meant to combat HD7970. But since HD7970 only beats GTX580 by 20%, NV might have decided to launch GK104 now. Let's wait for benches.

I dunno, I don't follow market trends. I'm just a consumer. I buy what I like at a certain price point and move on.

If you don't follow market trends, why are you following the launch of Kepler? If you are ready to buy now, you'd get HD7900 series.

Did you buy an HD7950/7970 yet? If yes, why do you care about how Kepler performs? If not, why didn't you buy those cards yet as they've been out for 1-2 months? It is because you want to see how Kepler performs for your gf's rig? If you don't follow market trends, then how are your arguments about what's expected price/performance in terms of history relevant? If you admit you don't follow market trends, then how can you say that Nvidia's next generation upper-mid-range performing ~ previous high-end is not reasonable when it has been the case for 10 years?

And this isn't even a question about who's cards are better or who's team you root for (clearly you are so bias you can't even entertain my scenario.)

Your scenario is GK104 is the highest end single-GPU Kepler card. All right, in that case, I expect the next highest end single-GPU NV card to beat GTX580 by 40-50%. If not, I'd consider it as NV changing its historical strategy.

If, here is the hyopthetical, GK104 is the only card based and annoucned by NVidia for the current time, they launch it with the GTX 680 name, relative to nVidia's line up, would you still call it a mid-range card?

If NV has 5 cards, GTX680, 681, 682, 683 and 684 and GK104 is GTX682, it's mid-range. If it's priced in the middle of Nvidia's card, it's mid-range, if it's performance is roughly the middle of NV's cards, it's midrange for single GPUs. If NV plans to launch GK104 first and it's faster than HD7970, but it has full intentions to follow up GK104 with a GK110, the GK104 is mid-range. If NV's fastest card for 28nm Kepler is GK104 with no other card planned, than GK104 is the new single-GPU high-end for NV. <-- see I highlighted that point for you.

In the end, I honestly don't care if GK104 is upper-mid-range or high-end. If it costs $700 and performs worse than GTX550Ti, it's low end for me. Like DeeJayeS says, it's just semantics. The main point is NV's new upper-mid-range card (whatever it's called) should end up being as fast as a GTX580 and their high-end card should end up faster than GTX580 by 40-50%.

If they beat that, the new generation has overachieved vs. their past performance. If they don't meet those metrics, the new generation has underachieved vs. their past performance. If they meet that, then their new generation is in-line with how they've operated their business in the past.

 

[DeeJayeS]

At the end, all that matters is GK104's performance and price.

You forgot timing. And as they say, timing is everything.

Seems like a lot of argument over semantics regardless.

 

[RussianSensation]

You forgot timing. And as they say, timing is everything.

Seems like a lot of argument over semantics regardless.

Agreed.

 

[railven]

I have a hard time accepting it for a few reasons:

1) 256-bit memory width for their highest end single-GPU
2) Based on the huge performance leaps NV has made in the past with each generation and the fact that NV desperately wants the single GPU-performance crown each generation -> unless you believe a 256-bit 350mm^2 die GPU from NV will be 20% faster than HD7970....

Let's just wait and see how it performs.

Yet, again, it seems like you just talk to yourself or don't even read what is being asked of you.

I only asked you a specific question with set conditions (mine) just because I was curious if when the stars align (my conditions) if you'd still call it "mid-range." I wasn't even trying to argue anything about performance or specs.

The onus is on you to show why it's reasonable that NV's highest end Kepler card would only be a 350mm^2 256-bit offering, not on me to prove why that seems hard to believe considering in the last 5 years NV hasn't made any high-end GPU with such a small die. It seems more reasonable than not that GK104 is just an upper-mid-range card from NV that was never meant to combat HD7970. But since HD7970 only beats GTX580 by 20%, NV might have decided to launch GK104 now. Let's wait for benches.

Again, what onus? I'm not making any claims. Jesus, now I get why the little back-and-forths with you go so long. You aren't even reading what I'm writing. At this point I'll just cut my losses before this nonsense keeps on going.

If you don't follow market trends, why are you following the launch of Kepler? If you are ready to buy now, you'd get HD7900 series.

Did you buy an HD7950/7970 yet? If yes, why do you care about how Kepler performs? If not, why didn't you buy those cards yet as they've been out for 1-2 months? It is because you want to see how Kepler performs for your gf's rig? If you don't follow market trends, then how are your arguments about what's expected price/performance in terms of history relevant? If you admit you don't follow market trends, then how can you say that Nvidia's next generation upper-mid-range performing ~ previous high-end is not reasonable when it has been the case for 10 years?

You asked me how the:

How is HD7950 doing on that front?

I answered you. I don't have spreed sheets or trackers on who is moving what inventory, at what rates, and etc. Do you really lack reading comprehension THAT much?

I follow card's launches as a consumer and when it is my turn to buy - I buy. After that, I don't really care who sold what to who, die sizes, etc etc. Is that hard to believe? That, gasp, some people here only really care about the technical information and not the back-and-forth power struggle by the two sides? Sheesh.

Your scenario is GK104 is the highest end single-GPU Kepler card. All right, in that case, I expect the next highest end single-GPU NV card to beat GTX580 by 40-50%. If not, I'd consider it as NV changing its historical strategy.

Now we're getting some where. That's all I was asking. Would you still call that card "mid-range?"

If NV has 5 cards, GTX680, 681, 682, 683 and 684 and GK104 is GTX682, it's mid-range. If it's priced in the middle of Nvidia's card, it's mid-range, if it's performance is roughly the middle of NV's cards, it's midrange for single GPUs. If NV plans to launch GK104 first and it's faster than HD7970, but it has full intentions to follow up GK104 with a GK110, the GK104 is mid-range. If NV's fastest card for 28nm Kepler is GK104 with no other card planned, than GK104 is the new single-GPU high-end for NV. <-- see I highlighted that point for you.

Thanks, considering I set these conditions in my first post - I don't get why we had to go down this rabbit hole. Does it change anything? Of course not. I was just curious. Kind of how like you asked me how I broke the cards up - and I answered. Isn't that so much easier?

In the end, I honestly don't care if GK104 is upper-mid-range or high-end. If it costs $700 and performs worse than GTX550Ti, it's low end for me. Like DeeJayeS says, it's just semantics. The main point is NV's new upper-mid-range card (whatever it's called) should end up being as fast as a GTX580 and their high-end card should end up faster than GTX580 by 40-50%.

If they beat that, the new generation has overachieved vs. their past performance. If they don't meet those metrics, the new generation has underachieved vs. their past performance. If they meet that, then their new generation is in-line with how they've operated their business in the past.

I never cared about price. I have my budget and I buy within it. If the cards are dripping with perf : price then I, as a consumer, win big. If they aren't, well damn that sucks hope next time I'm in the market to buy they stack up better.

Frankly I just play games And hopefully the money I spend lets me do that a little better. You guys can argue about die sizes and market share - I'll just go back to lurking once I buy

 

[RussianSensation]

My point was that one can't just compare GPUs to decide each architecture is better based on top card vs top card, if a card is 30-50% larger die size/transistor.

Agreed. Which is why it's impossible to say if Fermi architecture is better or worse than GCN is. They aren't made on the same nm node. No debate here. The only way to judge how much more efficient GCN is is to put it up against 28nm Kepler and then running both at the same clocks and with the same memory bandwidth. That still doesn't prevent AMD or NV from one-upping one another by launching a larger die sized GPU than HD7970 and taking the overall performance crown. I am sure many here would welcome a 30% faster HD7970 even if it's die size grew 30% in the process.

For example the 7970 has 30% higher transistor count than a GTX580, so its performance increase is in line.

HD7970 (4.3B) has 65% more transistors than HD6970 (2.6B) and only 45% more performance.
HD7970 (4.3B) has 43% more transistors than GTX580 (3.0B) and only 20-25% more performance.

HD7970's performance per transistor is worse than both the 580 and HD6970. However, someone who plays games isn't going to care that HD7970 has worse performance/transistor, are they?

Similarly, performance/mm^2 of die size is pretty useless for gamers. Who says performance/mm^2 of die is a more important metric than performance/transistor? Instead of engaging in argument on which GPU is better from an "engineering" (i.e., transistor/ die size point of view), it's better to measure GPUs based on what they are meant to do - PLAY GAMES. If we were electrical engineering students studying GPU design at Waterloo or MIT, then yes, everything you have said would matter to us.

That's why most consumers focus on performance per $ and frames per second, not performance/mm^2 or performance / transistor (something engineers might care about).

Thanks, considering I set these conditions in my first post - I don't get why we had to go down this rabbit hole. Does it change anything? Of course not. I was just curious. Kind of how like you asked me how I broke the cards up - and I answered. Isn't that so much easier?

Just to make the point clear, if GK104 is NV's highest end but it only matches HD7970 in performance and costs $550-600, I would still consider both not living up to expectations. By extension, that would mean the entire 28nm generation will only give us what normally would be considered upper-midrange levels of performance but at high-end GPU pricing. I don't believe this will happen. I think NV's highest end single GPU will be much faster than a stock HD7970.

 

[tviceman]

Possible GK107 sighting: http://we.pcinlife.com/forum.php?mod=viewthread&tid=1844549&page=1#pid33452412

Supposedly, based on reference of chips present in pic, this GPU is around 130-140mm^2, or about the same size or slightly bigger than cape verde. Benchmark shows 50% faster than gt550m in 3dmark.

 

[notty22]

NVmind

 

[kidsafe]

Why on earth would they design the PEG connectors that way. As already mentioned, it possibly inhibits single slot usage. In addition it's just a poor physical design...imagine the amount of force on the solder joints when numbskulls start forcible pushing down on and tugging up on their PEG.

 

[Concillian]

it says GT650m

Doesn't that imply mobile chip? Aren't all nVidias 6xx mobile chips Fermi?
Screenshot also shows hotclocks. The screenshot looks like a slow clocked 560 Ti and doesn't match with info that is either official (mobile will be Fermi) or rumored (no hotclocks).

 

[tviceman]

it says GT650m

Doesn't that imply mobile chip? Aren't all nVidias 6xx mobile chips Fermi?
Screenshot also shows hotclocks. The screenshot looks like a slow clocked 560 Ti and doesn't match with info that is either official (mobile will be Fermi) or rumored (no hotclocks).

It shows hotclocks because GPUz doesn't have Kepler in it's database yet, so it's reading it like a Fermi card. For further explanation read here: http://forum.beyond3d.com/showpost.php?p=1624375&postcount=2014 starting with this post and after this post.

But yeah, it's in a notebook. So hence, the "m" at the end of the GPU's operating name. And no, this cannot be a fermi based GPU with those specs. The chip is way too small.

 

[GaiaHunter]

HD7970 (4.3B) has 65% more transistors than HD6970 (2.6B) and only 45% more performance.
HD7970 (4.3B) has 43% more transistors than GTX580 (3.0B) and only 20-25% more performance.

HD7970's performance per transistor is worse than both the 580 and HD6970. However, someone who plays games isn't going to care that HD7970 has worse performance/transistor, are they?

Bigger memory controller.

GPGPU performance, not that I care about it yet.

Higher tess performance, not that I've seen anything to draw my attention yet.


Additionally people don't care that 7970/7950 is 28 nm vs 40 nm GTX580 or that one is 15 months old, but apparently that doesn't prevent it from being used as an argument.

Whatever price NVIDIA can sell a 28nm 340-360mm^2, if the cards perform similarly, AMD can also do so.

And I seriously doubt the GTX580 is that much more expensive to make than a 7970.

 

[RussianSensation]

Additionally people don't care that 7970/7950 is 28 nm vs 40 nm GTX580 or that one is 15 months old, but apparently that doesn't prevent it from being used as an argument.

You said you can't compare how good an architecture is without both GPUs being on the same nodes. I agree. That statement also makes it difficult to conclude that 20% increase for a new 28nm generation over GTX580 is good, not until NV shows its products on 28nm. The fact that it took AMD a brand new architecture + 28nm node shrink just to barely beat a GTX580 with HD7950 isn't a good start.

Basically that's like saying last generation HD7950 could have easily been a competitor to a GTX580. AMD hasn't really set that bar that high. Beating it should be a given. The only question is price.

Whatever price NVIDIA can sell a 28nm 340-360mm^2, if the cards perform similarly, AMD can also do so.

What happens if GK104 is NV's midrange card? What's AMD going to do to beat a 500mm^2 Kepler? Release a 20-30% faster clocked HD7970? That's a win-win for everyone.

What if the card in the picture has a memory bandwidth of 512-bit? That would more strongly support the notion of leaked benches showing 40-50% faster performance than GTX580. That 6+6 pin connector can just as easily be a 6+8 pin connector in the final card.

 

[toyota]

You said you can't compare how good an architecture is without both GPUs being on the same nodes. I agree. That statement also makes it difficult to conclude that 20% increase for a new 28nm generation over GTX580 is good, not until NV shows its products on 28nm. The fact that it took AMD a brand new architecture + 28nm node shrink just to barely beat a GTX580 with HD7950 isn't a good start.

Basically that's like saying last generation HD7950 could have easily been a competitor to a GTX580. AMD hasn't really set that bar that high. Beating it should be a given. The only question is price.



What happens if GK104 is NV's midrange card? What's AMD going to do to beat a 500mm^2 Kepler? Release a 20-30% faster clocked HD7970? That's a win-win for everyone.

sadly a stock 7950 has zero advantage over a gtx580 when using 4x AA. lol, now that is real progress for a next gen card at nearly the same price as the older piss poor value gtx580 from last gen.


jpeg image hosting

 

[Vesku]

See as a consumer die size is mainly important from how it effects the factors of price, performance, and power draw. But this statement you make is not a consumer statement, it took AMD a brand new architecture + 28nm node shrink to beat NVIDIA's *BIG DIE* product using their *sweet spot die* strategy.

All signs point to the GK104 being of similar 28nm size to the 7950/7970 AMD offering. Will make for an interesting head to head comparison when NVIDIA finally gets some out the door.

The fact that it took AMD a brand new architecture + 28nm node shrink just to barely beat a GTX580 with HD7950 isn't a good start.

 

[RussianSensation]

See as a consumer die size is mainly important from how it effects the factors of price, performance, and power draw. But this statement you make is not a consumer statement, it took AMD a brand new architecture + 28nm node shrink to beat NVIDIA's *BIG DIE* product using their *sweet spot die* strategy.

In one case a poster was implying that die sizes affect purchasing decisions by suggesting that a card with a larger die but similar performance is a turd, etc.

In my case, I am not making any claims about purchasing decisions because I don't know anything about price or AMD's possible response should they deem it necessary to lower prices. I am simply estimating performance of NV's next gen 28nm parts based on the fact that just to catch up to (beat by 20%) NV's 2-year-old architecture, it took AMD a brand new architecture + a node shrink. That makes it reasonable to estimate that the NV's brand new architecture + its own node shrink should easily surpass that mark. If not, well then NV really dropped the ball given how low the mark has been set thus far.

All signs point to the GK104 being of similar 28nm size to the 7950/7970 AMD offering. Will make for an interesting head to head comparison when NVIDIA finally gets some out the door.

Agreed. That would make for an excellent clock-for-clock comparison. We don't know if GK104 is NV's mid-range or high-end though. If GK104 is only NV's upper-midrange card, that would automatically relegate HD7900 series to being upper-midrange in their performance as well, possibly paving way for a faster HD7970 to combat the faster GK110. That would be very good for consumers.

sadly a stock 7950 has zero advantage over a gtx580 when using 4x AA. lol, now that is real progress for a next gen card at nearly the same price as the older piss poor value gtx580 from last gen.

But firms aren't a charity....You have to study the technology sector more. You wait 15-18 months to upgrade your single-core smartphone to a single-core smartphone. iPad 2 turns into iPad 3 without a new HD screen and more processing power, just better power consumption. You just get new "features" and lower power consumption. You should be happy with that. Where have you been the last 10 years?

Frankly, it might not be fair to put the blame on AMD. It's not their fault that GPU consumers are OK with spending $450 for 15 months old performance level. Sounds like AMD is onto something. Maybe Read and JHH should have a meeting and time it so that each other's GPUs beat the other by 5% every 6 months and prices stay at $450-550 level forever.



I hear there is new trend in high-end GPUs: launching a next generation high-end card 30-50% underclocked from the factory. Rumor has it, GK104 will launch with 600mhz clocks at $500. Its performance will only be 5% faster than GTX580, but NV expects people to overclock it 20-30% to beat GTX580 by 25-35% to show their overclocking skills and overclocking knowledge. It makes new GPUs that much better since to extract what was previously added performance for "free" given Moore's Law, NV will also expect people to play the overclocking lottery to achieve it.

From now on, all next generation GPUs will require overclocking to achieve significantly better performance than the previous high-end GPU. The main selling features will be power consumption :sneaky:

 

[jacktesterson]

Die size explains performance.

If a card is bigger die size and slower/similar speed, it is a turd.

IF a card is bigger die size and faster, it is explained by its bigger die size.

If a card is faster and smaller its great or the competition is a turd.

If NVIDIA K104 is similar die size and as fast as 7970/7950 it is only natural.

wow... terrible info here

 

[GaiaHunter]

In one case a poster was implying that die sizes affect purchasing decisions by suggesting that a card with a larger die but similar performance is a turd, etc.

Quote for that.

wow... terrible info here

Give me examples of cards that are smaller and faster and aren't a good product.

If it is bigger and slower you can always price it down, I guess.

 

[BallaTheFeared]

GK104 is mid ranged, I dunno why there is so much fuss over it.

 

[Vesku]

For AMD's "sweet spot" die strategy to catch up to NVIDIA's "as big as it can be made" strategy. It's silly to ignore die size when talking about the non-consumer purchasing side of things. Again, from an architecture perspective we will have some of the best means to compare (gaming, GPGPU, and such) when NVIDIA launches a 28nm part of near the same die size as AMD's part.

I am simply estimating performance of NV's next gen 28nm parts based on the fact that just to catch up to (beat by 20%) NV's 2-year-old architecture, it took AMD a brand new architecture + a node shrink. That makes it reasonable to estimate that the NV's brand new architecture + its own node shrink should easily surpass that mark. If not, well then NV really dropped the ball given how low the mark has been set thus far.

 

[RussianSensation]

Quote for that.

If a card is bigger die size and slower/similar speed, it is a turd.

So basically, according to you, NV hasn't made any good videocards in at least the last 5 years.

Using your logic, GTX460 was a turd since it had a larger die size than HD6850 I believe and it was slightly slower at stock speeds, but it became one of the most legendary price/performance cards in the last 5 years. HD6850 will forever be remembered as the card that existed during the time when GTX460 redefined price/performance for mid-range gamers at the time, not the other way around. No one is going to give HD3870 credit for getting smacked by 8800GT even if HD3870 did have a smaller die. Too bad. AMD should have made HD3870 with a larger die and beat 8800GT. Die size excuses don't count.

If in 2013, HD8790 is a 600mm^2 die and beats the crap out of Kepler, no one is going to be saying, let's buy the slower Kepler card with a smaller die....

Give me examples of cards that are smaller and faster and aren't a good product.

Give me examples of any generation outside of GeForce 5 where NV didn't have a good price/performance card and/or a top-end GPU that almost always was nearly as fast as ATI/AMD's high-end or faster. Your argument about smaller die sizes is so arbitrary. Show anyone how it matters? It didn't gain AMD market share, it didn't make AMD more profitable. Therefore, it doesn't matter for GPU buyers. It might matter to you if you are in charge of manufacturing the chip. I am too lazy to look up die sizes of GPUs before 2900XT between the 2 brands. But I know I don't remember any of those cards die sizes relative to each other - which even proves the point how meaningless die sizes are for consumers. I remember that 6800GT smacked X850Pro all over the place, especially once overclocked and that it could actually run HDR and SM3.0. That's what I remember. I also remember that X1900XT cards did really well in shader intensive games and 9700Pro/9800Xt smacked GeForce 5 in DX9 games.

You can argue until you are blue in the face that HD5870 had better performance / die than GTX480 did and you'd be right. But play modern games such as Batman: AC, Crysis 2 with Tessellation or BF3 with 4x MSAA and you can throw that HD5870 into the garage. In the pantheon of history it will always be remembered as the card that beat Fermi by 6 months (+), provided unbeatable price/performance at the time (+) but was completely useless for Tessellation in modern games that later to followed (-) and ran out of VRAM left right and center in games like BF3 (-) while GTX480 kept on trucking. No one is going to be sitting there and arguing that HD5870 had a smaller die size than a GTX480.

In 5 years from now, Fermi architecture will be remembered as the generation that jump-started the era of Tessellation and renewed focus on GPGPU compute. Cypress architecture will be remembered as one that was obsoleted because VLIW-4 wasn't good enough to serve the purpose of modern GPUs that required GPGPU functionality. In history books, Fermi will have played a far more important and dramatic role in graphics than VLIW-4 ever will. What die sizes?

If it is bigger and slower you can always price it down, I guess.

If it's smaller and slower you can always price it down, I guess.

Same logic since HD4000 series.