Nvidia Kepler, Part 2

[KAC]

My problem is not the issue here. 7970 should have launched at 429 and 7950 at 350 back in the day.

Now these cards should be going for 399 and 329 respectively. This is their true value in performance / $ with usual 10-15% improvement generation on generation.

At 549 and 449 these cards were overpriced and nVidia came out and told them their place. Even then, AIBs now feel the need to discount the MSRP further only less than 2 weeks after the official price drop (Diamond is selling for 439 on Amazon.com AR).

Even GTX 680 is overpriced at 499 (should be 449 or 479).

[caveman-jim]

GK104 is in low supply - ~ 10,000 GTX 680 cards worldwide

http://semiaccurate.com/2012/05/01/w...ergk104gtx680/

[shadow001]

The pricing is due to lack of competition since Nvidia only has to watch what AMD is doing and AMD only haas to watch what Nvidia is doing and that's all there is to it....I remember the days when 3DFX, Power VR, matrox, S3 were also competing with Nvidia and ATI and it made it way more interesting overall in terms of price and competition.


We're down to just 2 big players now, and it wouldn't hurt if a third one showed up on scene...

[SIrPauly]

Originally Posted by caveman-jim

GK104 is in low supply - ~ 10,000 GTX 680 cards worldwide

http://semiaccurate.com/2012/05/01/w...ergk104gtx680/

How exactly does Charlie know how many GK104's have been shipped -- how can anyone be privy to that kind of information except for nVidia?

[BababooeyHTJ]

Originally Posted by caveman-jim

GK104 is in low supply - ~ 10,000 GTX 680 cards worldwide

http://semiaccurate.com/2012/05/01/w...ergk104gtx680/

I'm just not buying his explanation as to why. I'm sure that it has more to do with the TMSC shutdown than anything. My favorite line is where he says that "by Nvidia's own admission they have yield issues" and then he links to a investor relations conference call from over two months before GTX680 was even released. You know back before he was even spouting off about GTX680's great physix performance. Charlie has been dead wrong about Kepler so far. I don't see any reason for me to believe him now.

If yields are so bad where are the cards with castrated cores? If rumors are right even GTX670 is almost a full fledged GK104.

I'm not buying it.

[VW_Factor]

Originally Posted by SIrPauly

How exactly does Charlie know how many GK104's have been shipped -- how can anyone be privy to that kind of information except for nVidia?

he isn't.

He has always been a wild speculator and always will be. Nothing more.

[SIrPauly]

MSI GeForce GTX 670 OC Edition Pictured and Detailed\

http://wccftech.com/msi-geforce-gtx-...ured-detailed/

[SIrPauly]

NVIDIA CEO, Two Others, Headline GTC 2012 Keynotes

And, he’ll announce some big GPU news that you’ll not want to miss.

http://blogs.nvidia.com/2012/05/nvid...2012-keynotes/

[Xion X2]

Originally Posted by acroig

You have a point but what Pauly and I are saying is that these cards are overpiced at the moment.

Pricing usually reflects modern day status.. not performance vs historical performance.

The cards that are out now are the max high-end and are priced as such. Just as the 7900GTX, 1900XT, 8800GTX, GTX280, GTX580 etc were before them.

It doesn't make any sense to me that some of you keep stating that 'oh well it's only 15% faster than last gen so should be priced as such.' When has this ever been the case? I remember paying 529$ for my 7900GTX. I remember paying 620$ a clip for my 8800GTXs.

Do I like paying this? Of course not. But this excessive pricing for the high end is nothing new.

Companies will always charge a premium for the high end if they can get away with it.. doesn't matter if it's a 5% or a 50% increase in performance over prior gen. Intel does the same thing with their $1,000 CPUs that they've sold over the years.

[acroig]

Originally Posted by Xion X2

It doesn't make any sense to me that some of you keep stating that 'oh well it's only 15% faster than last gen so should be priced as such.' When has this ever been the case? I remember paying 529$ for my 7900GTX. I remember paying 620$ a clip for my 8800GTXs.

When? AMDs 4870, that's when.

[Meteor|WAR]

Originally Posted by SIrPauly

MSI GeForce GTX 670 OC Edition Pictured and Detailed\

http://wccftech.com/msi-geforce-gtx-...ured-detailed/

I hope mine comes with one of those cool new dispaly ports.

[Xion X2]

Originally Posted by acroig

When? AMDs 4870, that's when.

Good. Name another. That was the exception and not the rule. AMD did that simply to get back in the game and gain market share. Now that they have market share, they're pricing at a premium.

[SIrPauly]

Originally Posted by Xion X2

Pricing usually reflects modern day status.. not performance vs historical performance.

The cards that are out now are the max high-end and are priced as such. Just as the 7900GTX, 1900XT, 8800GTX, GTX280, GTX580 etc were before them.

It doesn't make any sense to me that some of you keep stating that 'oh well it's only 15% faster than last gen so should be priced as such.' When has this ever been the case? I remember paying 529$ for my 7900GTX. I remember paying 620$ a clip for my 8800GTXs.

Do I like paying this? Of course not. But this excessive pricing for the high end is nothing new.

Companies will always charge a premium for the high end if they can get away with it.. doesn't matter if it's a 5% or a 50% increase in performance over prior gen. Intel does the same thing with their $1,000 CPUs that they've sold over the years.

You're confusing price/performance history based on new nodes and arches with price points. The point has never been about the price point but the performance/value for the price point based on a new node and arch. When there is lack luster performance/value at the high end translates into very lack luster performance/value for lesser price-points.

[SIrPauly]

Originally Posted by acroig

When? AMDs 4870, that's when.

Virtually every single new node and arch redefined price/performance -- this may be nVidia's worse generation including the GeForce 5800 when the context is performance when compared to a new node and arch.

Ryan hit the nail on the head at Anandtech.com:

I think ultimately we still haven’t moved very far on the price/performance curve compared to where we’ve gone in past generations, and on that basis this is one of the smaller generational jumps we've seen for a GTX x80 product

That's the constructive nit-pick and so painfully obvious but the key, there is still a lot to like about the GTX 680 -- the price/performance isn't one of them.

[SIrPauly]

Gainward GeForce GTX 680 Phantom Review

http://www.techspot.com/review/525-g...force-gtx-680/

[caveman-jim]

Originally Posted by BababooeyHTJ

I'm just not buying his explanation as to why. I'm sure that it has more to do with the TMSC shutdown than anything. My favorite line is where he says that "by Nvidia's own admission they have yield issues" and then he links to a investor relations conference call from over two months before GTX680 was even released. You know back before he was even spouting off about GTX680's great physix performance. Charlie has been dead wrong about Kepler so far. I don't see any reason for me to believe him now.

If yields are so bad where are the cards with castrated cores? If rumors are right even GTX670 is almost a full fledged GK104.

I'm not buying it.

I don't agree with your assertion that 'Charlie has been dead wrong about Kepler so far'. He said that NVIDIA would win this round, in all metrics, after first saying the would win some/lose some. His claim about PhysX is interesting... makes me think that was part of a leak to find out who was leaking to him, or he added it as seeing who was claiming they had sources but really didn't.

Regarding linking to a call 2mo. before.... he's exactly right. NVIDIA did state they had 28nm yield problems in that call. What product that was referring to is now clear - GK104. NVIDIA were preparing their financial guidance to cover the fact of low shipping volume; or setting a low expectation bar so they can say they beat it through their engineering prowess and hard work. There's probably a third simple, easy explanation as well that I'm not thinking of.

Originally Posted by SIrPauly

How exactly does Charlie know how many GK104's have been shipped -- how can anyone be privy to that kind of information except for nVidia?

A good question. Obviously NVIDIA will know how many wafers they've bought, how many chips they've used in shipping products. TSMC will to some extent as well. AIB partners will know, too - if not the total number of good chips, then how many chips they're allocated, and how much as % of total that is. If Charlie has sources inside NVIDIA, TSMC, and AIB's, then he can put that data together. If depends on if you think he has those sources or not.

However, we do know that he will flat out alter numbers and make things up, so see who will take the bait. This makes it extremely difficult to believe every article and every fact, as you just don't know. So what you have to look for is independent confirmation. Other sites double checking him. I've tried a couple times in the past, and I don't have the contacts to chase down his info. You'd have to look to the guys that have long term, long standing industry relations.

I don't think you can flat out dismiss SemiAccurate. But you do have to know how to read his writing style, which is not easily accessible for some. And you do have to be skeptical. Which I think is kinda his point.

[SIrPauly]

Inno3D Announces the iChill HerculeZ 3000 GTX 680 Graphics Card

http://www.techpowerup.com/165293/In...hics-Card.html

[caveman-jim]

OBR claims Big Kelp won't be a GeForce GPU

http://www.obr-hardware.com/2012/05/...ed-in-two.html

However, OBR.... so,

[Ziklitschli]

Originally Posted by KickAssCop

It is from a perf/$ perspective. Doesn't mean they aren't worth purchasing.
That is like saying never purchase a Ferrari and pay a premium...

Exactly, if I had enough money to spend today I could buy 7970 or 680 as those are great cards, just at a rip-off price. You can buy and enjoy but still be aware of how the price should really look like. Subjectively I consider 800$ high end card still worth it out of the single fact it's PC and best gaming quality you can get, doesn't change the fact that company setting such price would be badly ripping the buyer off.

[SIrPauly]

Originally Posted by caveman-jim

OBR claims Big Kelp won't be a GeForce GPU

http://www.obr-hardware.com/2012/05/...ed-in-two.html

However, OBR.... so,

I think the monolith will be introduced for the professional lines at GTC and when 28nm matures may be offered for GeForce under a GTX 780 moniker in later 2012 or early 2013.

[SIrPauly]

Originally Posted by caveman-jim

I don't agree with your assertion that 'Charlie has been dead wrong about Kepler so far'. He said that NVIDIA would win this round, in all metrics, after first saying the would win some/lose some. His claim about PhysX is interesting... makes me think that was part of a leak to find out who was leaking to him, or he added it as seeing who was claiming they had sources but really didn't.

Regarding linking to a call 2mo. before.... he's exactly right. NVIDIA did state they had 28nm yield problems in that call. What product that was referring to is now clear - GK104. NVIDIA were preparing their financial guidance to cover the fact of low shipping volume; or setting a low expectation bar so they can say they beat it through their engineering prowess and hard work. There's probably a third simple, easy explanation as well that I'm not thinking of.




A good question. Obviously NVIDIA will know how many wafers they've bought, how many chips they've used in shipping products. TSMC will to some extent as well. AIB partners will know, too - if not the total number of good chips, then how many chips they're allocated, and how much as % of total that is. If Charlie has sources inside NVIDIA, TSMC, and AIB's, then he can put that data together. If depends on if you think he has those sources or not.

However, we do know that he will flat out alter numbers and make things up, so see who will take the bait. This makes it extremely difficult to believe every article and every fact, as you just don't know. So what you have to look for is independent confirmation. Other sites double checking him. I've tried a couple times in the past, and I don't have the contacts to chase down his info. You'd have to look to the guys that have long term, long standing industry relations.

I don't think you can flat out dismiss SemiAccurate. But you do have to know how to read his writing style, which is not easily accessible for some. And you do have to be skeptical. Which I think is kinda his point.

Nvidia stated lower than expectations and lowered guidance from around 51 percent to 49 percent.

According to Steam data the GTX 680 is doing as well as the HD 7970 did in a similar window of time, which had no 28nm competition, and doing much better than the 78XX series, which launched around the same time-line as the GTX 680. Also more than the HD 7950.

[acroig]

Originally Posted by Xion X2

Good. Name another. That was the exception and not the rule. AMD did that simply to get back in the game and gain market share. Now that they have market share, they're pricing at a premium.

I named one, not good enough?

You are right, of course. At least we can say it happened once in our lifetime.

[BababooeyHTJ]

Originally Posted by SIrPauly

I think the monolith will be introduced for the professional lines at GTC and when 28nm matures may be offered for GeForce under a GTX 780 moniker in later 2012 or early 2013.

Yeah, there is no way that there won't be a desktop variant.

[SIrPauly]

Forget The GTX 690: The Cloud-Based "GeForce Experience" Could Mean More For PC Gaming

http://www.maximumpc.com/article/new...more_pc_gaming

Offers part of the Keynote in China that touches on the GeForce Experience.

[Roadhog_]

Something.

[SIrPauly]

GTX 670 picture and this site has that picture and some more:

http://we.pcinlife.com/thread-1894715-1-1.html

[SIrPauly]

That product doesn't look like a typical 399 MSRP sku!

[Roadhog_]

Originally Posted by SIrPauly

That product doesn't look like a typical 399 MSRP sku!

Looks like the cheapest pcb i've ever seen.

[SIrPauly]

(PhysX Test) GTX 680 vs GTX 580 vs GTX 480 in FluidMark

http://www.geeks3d.com/20120502/phys...-in-fluidmark/

[SIrPauly]

The GTX 680 did well with the PhysX tests compared to Fermi, which was actually great to see.