GF 104 is 366mm2

[Keysplayr]

That was when GF104 was supposed to just be an unchanged half of GF100. At launch, about 3 weeks after the article was posted, it became obvious that GF104 was not just a half of GF100, and then ten days after that, he gave a revised value of GF104's die size that was correct to NordicHardware's estimate to within 1mm^2. Yes, he was wrong initially, but he was also right a hell of a long time before anyone else was.


He was saying GF104 couldn't compete profit-wise even when he thought GF104 was going to be smaller than Cypress.


Then why did it nearly take NordicHardware another 2 weeks to do it?

What is the significance of this? What does NordicHardware have to do with Charlie Demerjian? And besides, an answer for this would be that Charlie was starving for dirt on Nvidia and was all that much more hungry to get it out there. Maybe Nordic Hardware didn't care that much about it?
IMHO, Charlie absolutely hates the GTX460. He loves GTX480,475,465. Why? Because it's almost impossible to knock the 460. It's a great GPU all around.
As for the 32mm2 larger die size? Picture a rectangle at 8mm x 4mm. It's a sliver. How much more could it cost than a cypress core? Sure, multiply that by millions, but also consider how many more Nvidia will sell, and how many more they purchase from TSMC.

There is more to consider than just die size. And to be honest, I don't even know why people do. I see it mentioned above that some realize that the current GF104 isn't even a fully enabled core. As yields improve, that will change. Probably already has.

 

[Creig]

What is the significance of this? What does NordicHardware have to do with Charlie Demerjian?

I would hazard a guess and say that Mr. Pedantic was simply showing that Charlie was able to give an accurate measurement of the GF104 die weeks before anybody else did. People tend to dismiss Charlie because of his pro-ATI/anti-Nv stance, yet he does seem to be correct more often than not.

As for the 32mm2 larger die size? Picture a rectangle at 8mm x 4mm. It's a sliver. How much more could it cost than a cypress core?

Hard to say. And yet no matter you look at it, it's more expensive to produce than the 5870. That makes people try to guess how much profit Nvidia/ATi are making per GPU of roughly comparable size.

Sure, multiply that by millions, but also consider how many more Nvidia will sell, and how many more they purchase from TSMC.

???? I'm not sure what you're getting at here. If Nvidia eventually orders/sells millions, that's millions of dies that were more expensive to produce than the flagship ATi core. ATi has already sold millions of Cypress. I don't see that Nvidia will get any sort of price break over ATi at TSMC since 40mm supply is supposedly constrained.

There is more to consider than just die size. And to be honest, I don't even know why people do.

Because people are interested in the technology behind the cards. And die size can affect multiple aspects (power consumption, heat, overclocking ability, pricing, etc).

I see it mentioned above that some realize that the current GF104 isn't even a fully enabled core. As yields improve, that will change. Probably already has.

So you're saying that the GF104 which was released only weeks ago with seven SMs has probably already been bumped up to eight SMs? Do you have any proof of this? Or are you simply guessing?

 

[Scali]

???? I'm not sure what you're getting at here.

He means that larger sales volume means that profit can be spread out over a larger number of units. So a bit less profit per unit would not necessarily result in less profit overall.

So you're saying that the GF104 which was released only weeks ago with seven SMs has probably already been bumped up to eight SMs? Do you have any proof of this? Or are you simply guessing?

It's all about yields.
It's certainly not far-fetched to assume that even today, there are already GPUs that can be binned for 8 SMs.
So it could be that nVidia is stocking up some of these 8 SM-bin GPUs in order to launch a new model later. The volume of these 8 SM-bin GPUs may just not be large enough at this time. nVidia has used this strategy in their last few generations aswell.

 

[SlowSpyder]

High end graphics cards draw a lot of power and require complex boards. On all Cypress, GF100, and GF104 products, the board and components are likely more expensive than the GPU itself. Even if GF104 was more expensive to manufacture than Cypress, it's board will considerably cheaper.

I didn't think that was true...? I thought the GPU was the most expensive part. And why would a Cypress board cost so much more? It's a larger PCB, both use 256 bit (except the 768MB 460), both draw similar power, and both use DDR5 though Cypress uses faster memory. Both seem to have moderate coolers, nothing fancy.

I think it's pretty impressive how AMD is able to squeeze more performance out of less, but then again their objectives are somewhat different. Nvidia is building a part that can do lots of things, AMD is building a video card.

 

[GaiaHunter]

and both use DDR5 though Cypress uses faster memory.

I'm was under the impression that the GF100 and Cypress used similar speed rater GDD5 and the speed differences that can be seen is due to the NVIDIA memory controller having some problems.

Was that changed with GF104?

 

[SirPauly]

Hard to say. And yet no matter you look at it, it's more expensive to produce than the 5870. That makes people try to guess how much profit Nvidia/ATi are making per GPU of roughly comparable size.

Considering the 5870 is the flag ship single chip, one would say it costs more based on its yield when compared to the other sku's utilizing Cypress.

 

[Mr. Pedantic]

What is the significance of this? What does NordicHardware have to do with Charlie Demerjian? And besides, an answer for this would be that Charlie was starving for dirt on Nvidia and was all that much more hungry to get it out there. Maybe Nordic Hardware didn't care that much about it?
IMHO, Charlie absolutely hates the GTX460. He loves GTX480,475,465. Why? Because it's almost impossible to knock the 460. It's a great GPU all around.
As for the 32mm2 larger die size? Picture a rectangle at 8mm x 4mm. It's a sliver. How much more could it cost than a cypress core? Sure, multiply that by millions, but also consider how many more Nvidia will sell, and how many more they purchase from TSMC.

There is more to consider than just die size. And to be honest, I don't even know why people do. I see it mentioned above that some realize that the current GF104 isn't even a fully enabled core. As yields improve, that will change. Probably already has.

You're right, it has nothing to do with it. But that wasn't really the point I was trying to make. The point is that Charlie was right. And he was right before anyone else was right.

 

[OCGuy]

Because people are interested in the technology behind the cards. And die size can affect multiple aspects (power consumption, heat, overclocking ability, pricing, etc).

Well so far the GF104 (460) is priced great, overclocks like hell, and has great heat/power consumption according to anand's review.

So while maybe is can be an indicator, it is just a small piece of the puzzle.

 

[Lonyo]

Well so far the GF104 (460) is priced great, overclocks like hell, and has great heat/power consumption according to anand's review.

So while maybe is can be an indicator, it is just a small piece of the puzzle.

Yes, yes, no. Power consumption isn't great, it's just much improved over the higher end Fermi based cards.It uses more power than the HD5850 despite being slower. Good, but never great.

 

[OCGuy]

Yes, yes, no. Power consumption isn't great, it's just much improved over the higher end Fermi based cards.It uses more power than the HD5850 despite being slower. Good, but never great.

http://www.anandtech.com/show/3809/nvidias-geforce-gtx-460-the-200-king/17


Looks to be better on idle and basically equal on load.

edit: basically equal on both.

 

[Scali]

http://www.anandtech.com/show/3809/nvidias-geforce-gtx-460-the-200-king/17


Looks to be better on idle and basically equal on load.

edit: basically equal on both.

I think the more important issue is this:
The GTX460 is faster than the 5830, and the 5830 has the same (or worse) power consumption as the 5850.
The 5850's slightly better performance for the same power comes at a price.

 

[Skurge]

http://www.anandtech.com/show/3809/nvidias-geforce-gtx-460-the-200-king/17


Looks to be better on idle and basically equal on load.

edit: basically equal on both.

So 2 watts at idle is better, but 6 watts at load is equal? Damn, my math is seriously massed up cause I can't get my head around that.

 

[GaiaHunter]

I think the more important issue is this:
The GTX460 is faster than the 5830, and the 5830 has the same (or worse) power consumption as the 5850.
The 5850's slightly better performance for the same power comes at a price.

The power consumption between these cards isn't really a factor. It isn't like GF100 vs Cypress differences. (BTW the 5830 power consumption is pretty much on par with the GTX460 - slightly higher idle, lower load than 1GB model, higher than 768MB model).

The relevant factors here are performance and price.

The GTX460 is a good card because of good bang/buck.

 

[HurleyBird]

I am aware that he was the first to talk about the 366mm^2, but that happened after the GF104 was released. Anyone could take a card and open the heatsink and measure the Die size at that moment.

GTX460 1GB have 160W

But nobody did... except Charlie's source. Kind of important considering most sites were wrongly assuming GF104 was smaller than cypress.

I thought GTX460 was a bit more than 160W, but the fact remains that it is a low clocked salvage part. I have no difficulty seeing a high clocked, fully enabled GF104 consuming over 200w.

 

[Scali]

The relevant factors here are performance and price.

The GTX460 is a good card because of good bang/buck.

I don't really agree here.
I thought the GTX465 and GTX470 were priced quite reasonably, considering the performance they delivered.
Yet I didn't buy either of them, because of the rather extreme power consumption.

I bought the GTX460 because it has good power consumption. Everything else (performance, price, features) is pretty much equal with the GTX465, so if I didn't think power consumption was relevant, I might as well have bought the GTX465.

 

[HurleyBird]

As for the 32mm2 larger die size? Picture a rectangle at 8mm x 4mm. It's a sliver. How much more could it cost than a cypress core? Sure, multiply that by millions, but also consider how many more Nvidia will sell, and how many more they purchase from TSMC.

Your advocacy is showing pretty strongly here, keys.

You're right that 32mm2 by itself isn't going to make or break Nvidia. It is going to hurt them though. It's not just the slightly bigger die, it's also coupled with the somewhat lower yields. Based on the fact that GF104 has only shown itself in a moderately harvested version, the default assumption has to be that yields aren't that great right now. I love how you just assume GF104 is going to be a higher volume part than Cypress with nothing to back it up. Fact is, cypress is a smaller die, covers a larger range of the market, has significant market penetration according to the steam hardware surveys, and is already very mature with very good yields, and ATI now has more marketshare than Nvidia as well. Nvidia has a lot of work to do if they want to catch up to cypress-level volume.

The significant thing about the GF104 die size is that it means Nvidia is pricing a more expensive part (and yes, it is obviously more expensive right now vs. the smaller more mature more mass produced cypress core. In the future that may or may not change, but right now it is undeniably more expensive) significantly below what AMD is pricing their stuff at. This isn't a part that is going to make huge amounts of money, it's positioned as a marketshare bandage part. The only issue is that soon enough GF104 is going to be the only competitive part Nvidia has left and it isn't going to be a cash cow. Southern Islands is probably going to obsolete GF100, and CPU-integrated graphics in SB and Fusion is going to instantly evaporate Nvidia's low end. That leaves them stuck in the middle with GF104, and at that time they'll need to sell the fully enabled part at $200 because of pressure from SI and derivatives. They still have the pro market, but if the rumors of AMD coming out with much better performing drives for firepro are true, that might start eroding as well.

So, the point is that Nvidia is going to be locked out of the high and low ends fairly quickly, and their mid-range card, while priced and positioned well for consumers, probably isn't going to be enough to stop Nvidia from losing a lot of money.

 

[OCGuy]

Your advocacy is showing pretty strongly here, keys.

You're right that 32mm2 by itself isn't going to make or break Nvidia. It is going to hurt them though. It's not just the slightly bigger die, it's also coupled with the somewhat lower yields. Based on the fact that GF104 has only shown itself in a moderately harvested version, the default assumption has to be that yields aren't that great right now. I love how you just assume GF104 is going to be a higher volume part than Cypress with nothing to back it up. Fact is, cypress is a smaller die, covers a larger range of the market, has significant market penetration according to the steam hardware surveys, and is already very mature with very good yields, and ATI now has more marketshare than Nvidia as well. Nvidia has a lot of work to do if they want to catch up to cypress-level volume.

The significant thing about the GF104 die size is that it means Nvidia is pricing a more expensive part (and yes, it is obviously more expensive right now vs. the smaller more mature more mass produced cypress core. In the future that may or may not change, but right now it is undeniably more expensive) significantly below what AMD is pricing their stuff at. This isn't a part that is going to make huge amounts of money, it's positioned as a marketshare bandage part. The only issue is that soon enough GF104 is going to be the only competitive part Nvidia has left and it isn't going to be a cash cow. Southern Islands is probably going to obsolete GF100, and CPU-integrated graphics in SB and Fusion is going to instantly evaporate Nvidia's low end. That leaves them stuck in the middle with GF104, and at that time they'll need to sell the fully enabled part at $200 because of pressure from SI and derivatives. They still have the pro market, but if the rumors of AMD coming out with much better performing drives for firepro are true, that might start eroding as well.

So, the point is that Nvidia is going to be locked out of the high and low ends fairly quickly, and their mid-range card, while priced and positioned well for consumers, probably isn't going to be enough to stop Nvidia from losing a lot of money.

Your whole post is based on nV doing nothing.

It sounds like the same doom-and-gloom that the Charlie crowd was parroting before GF104 which is looking like it could be a hit.

 

[HurleyBird]

Your whole post is based on nV doing nothing.

I'm assuming that SI will perform somewhat better than GF100, will be much smaller, that GF100 (at least the graphics card) will then become uneconomical to produce. Also assuming that Nvidia isn't going to have anything for awhile to counter SI. Seems pretty safe, but I'd love to be wrong for the sake of more competition and lower prices.

The fact that decent CPU integrated graphics are coming means that it isn't much of an assumption that Nvidia is going to lose the low end, at least up to the point where an add-in card has high enough performance to justify it over the free CPU graphics. The only question is where that cut-off point is going to be. I'm guessing $100 plus or minus $25. In any case, it's not a question of whether or not Nvidia is going to lose volume, it's how much volume they are going to lose.

It sounds like the same doom-and-gloom that the Charlie crowd was parroting before GF104 which is looking like it could be a hit.

Nobody denies that GTX 460 is a great buy at $200 *for the consumer*, the question is how much money Nvidia is making. If they are not making very good money on GF104, and if it is at least somewhat likely that this will be the only competitive part Nvidia has for awhile, a failure here could hurt Nvidia's ability to compete in the future depending on how much money they hemorrhage .

Another example: A GTX 480 at $50 would also be great *for the consumer*, but would be kill Nvidia.

 

[GaiaHunter]

I don't really agree here.
I thought the GTX465 and GTX470 were priced quite reasonably, considering the performance they delivered.
Yet I didn't buy either of them, because of the rather extreme power consumption.

I bought the GTX460 because it has good power consumption. Everything else (performance, price, features) is pretty much equal with the GTX465, so if I didn't think power consumption was relevant, I might as well have bought the GTX465.

Where did I mention GTX470 and GTX 465 in my post? Neither did you mention those cards on the post that I quoted from you.

I'm comparing Cypress vs GF104 (and so were you) - and for all purposes the power consumption is on the same level, give or take a watt.

When you compare a GTX 460/5830/5850 with a GF100 the power consumption (and the by products associated) is indeed a factor, but we weren't talking about GF100 vs GF104/Cypress.

 

[HurleyBird]

I'm comparing Cypress vs GF104 (and so were you) - and for all purposes the power consumption is on the same level, give or take a watt.

Not really. HD 5850 outperforms GTX 460 with lower power consumption. Perf/watt is still very much in AMD's favor... but compared to GF100 Nvidia has taken a significant step forward.

 

[Scali]

Where did I mention GTX470 and GTX 465 in my post? Neither did you mention those cards on the post that I quoted from you.

I mentioned them because especially the GTX465 is pretty much equal to the GTX460 in both price and performance, yet I didn't buy the GTX465, but I did buy the GTX460.
So there's more to the GTX460 than just its bang for the buck factor. The GTX465 had exactly the same bang for the buck, if you only look at price, performance and features.

 

[tincart]

This information is important to me.

I invest in tech stocks and in the past I have owned both AMD and nV shares (presently, I hold neither). Knowing the approximate cost of materials for nV's $200 video card vs AMD's $400 is certainly going to have a bearing on which company is going to make money. I like companies that make money because they make me money.

nV investors would have LOVED to know about the size of GF104 before their revised revenue estimate came out. You would have seen people dumping that stock sooner.

 

[GaiaHunter]

I mentioned them because especially the GTX465 is pretty much equal to the GTX460 in both price and performance, yet I didn't buy the GTX465, but I did buy the GTX460.
So there's more to the GTX460 than just its bang for the buck factor. The GTX465 had exactly the same bang for the buck, if you only look at price, performance and features.

And there you go talking about GTX 470 and 465 (and btw GTX 465 launched 2 months ago at $279 and while you can see some in newegg at $234, many are still at $250+ not the <$230 of the GTX 460).

I already said power consumption is a factor!

But power consumption of a GTX 460 isn't a factor compared to the 5850 or 5830! Is this so hard to grasp?

At the same price I would pick a 5850 over a GTX 460! I would even pick a 5850 over a GTX 460 at $50 more!

Probably most (informed) people that wants to play games would do the same choice - and that choice would be based on price/performance.

Any comparison with the GTX 470/465 is immaterial because then you would have to factor the power consumption too.

 

[tincart]

Your advocacy is showing pretty strongly here, keys.

One must show deference to their patron.

 

[bryanW1995]

I don't care who makes money with what. The competition is finally heating up again. Just need to wait for ATI to get their new stuff out so Nvidia can drop prices.

no, no, no need to mention heat in this thread. you must be thinking about a gf 100 thread...