Nvidia threatens partners at CeBIT

[Kuzi]

http://www.semiaccurate.com/2010/03/02/nvidia-threatens-its-partners-cebit/

Below is a quote from Charlie's article at SemiAccurate:

"The benchmarks are on par with the ATI HD5850 for the GTX470 and about 5 percent above the HD5870 for the GTX480, but both Nvidia cards run much hotter. Clocks are 625/1250MHz, way short of the intended 750/1500, and wattage is through the roof.

Two box samples seen by SemiAccurate had no clocks or specs on them, quite telling for a product that will paper launch later this month. There must be quite a bit of panic in Santa Clara right now. Pricing wasn't set according to four partners talked with by SemiAccurate, and availability for the high-end part, at least according to what Nvidia has promised to deliver, is basically a few hundred units per vendor. Worldwide. Ever."

I find this info hard to believe, but it won't be long before we find out if he's right or not.

 

[Rent]

The sad part is nV won't crush any rumors, some of which seem to have a hint of validity to them. The marketing bullshit is strong with this product.

 

[santz]

Disturbance in the force, feel it I can.

 

[AyashiKaibutsu]

below is a quote from charlie's article at semiaccurate:

I find this info hard to believe, but it won't be long before we find out if he's right or not.

orly?

 

[SunnyD]

Disturbance in the force, feel it I can.

Nah. That's just gas.

 

[Kuzi]

orly?

Yes. My magic ball is telling me that Charlie is either right or wrong, although I think it is the later

 

[Creig]

The only thing my Official Nvidia Fermi 8-ball ever tells me is, "Ask again later".

 

[Leyawiin]

The sad part is nV won't crush any rumors...

Basically, don't feed the trolls. And Charlie is a troll (albeit on a grand scale).

 

[MarcVenice]

I don't think Nvidia threatened anyone. That's the thing with charlie, he overreacts. Charlie is right to some extent though. After talking with a lot of vendors myself, some are preferred Nvidia partners, and some aren't. I've in fact been hearing many of the same things Charlie is writing, it's just the way he puts it, that's wrong.

 

[happy medium]

I don't think Nvidia threatened anyone. That's the thing with charlie, he overreacts. Charlie is right to some extent though. After talking with a lot of vendors myself, some are preferred Nvidia partners, and some aren't. I've in fact been hearing many of the same things Charlie is writing, it's just the way he puts it, that's wrong.

What kinds of things are they saying?
Can you talk about it or are you under NDA too?

 

[SunnyD]

Simply put, if Fermi was going to be a good launch in the wake of ATI's 5-series, they'd want as much info leaking out as possible right now.

 

[SlowSpyder]

Simply put, if Fermi was going to be a good launch in the wake of ATI's 5-series, they'd want as much info leaking out as possible right now.

That's what I think. Their other parts are running dry, and they are launching Fermi at the end of the month. I don't see why they wouldn't release information to stop people from buying AMD parts if Fermi was going to be widely available and give performance worth waiting for.

 

[nitromullet]

orly?

I think he's talking about this specifically:

availability for the high-end part, at least according to what Nvidia has promised to deliver, is basically a few hundred units per vendor. Worldwide. Ever.

You don't think it's a little far fetched to think they might only sell a few hundred GTX 480s worldwide?

 

[MarcVenice]

What kinds of things are they saying?
Can you talk about it or are you under NDA too?

I can't really tell any names, but I'm not with a vendor but indepent press. Thing is Charlie calls it threatening, but it's rather normal that Nvidia doesn't want partners to show anything besides the material they provide. I don't exactly see how that is threatening? It's keeping your products under wraps. I do have a meeting with Nvidia, but that will be under NDA...

Also, when he talks about a few hundred per vendor, he could be right. One source told me that there will be 5000 samples, and after that the product might actually go EOL. Then again, a different source told me they were working on overclocked cards, which would be a waste of time if there's only 5k total.

 

[Lonyo]

I think he's talking about this specifically:



You don't think it's a little far fetched to think they might only sell a few hundred GTX 480s worldwide?

Not if the yields are poor, they don't want to make it, they have a refresh coming "soon" and they want to make as many into Tesla cards as they possibly can (for the monsterously higher margins that offers).

 

[BenSkywalker]

Simply put, if Fermi was going to be a good launch in the wake of ATI's 5-series, they'd want as much info leaking out as possible right now.

Only if they were profoundly bad at business. The GTX2xx parts are still selling, and selling at a premium over what they were selling at a couple months ago. They are closing in on a full flush of their products through the channels right before a new product launch. If the GTX series was OOS all over and their partners had nothing else to sell then it would be wise to start leaking info. Until then, they keep their mouths shut. That is very basic business practice, avoiding write offs on last years products significantly improves operating margins, for nV and their partners.

That is one of those areas where people confuse what they want to see with good business while they are decidedly polar opposites.

Clocks are 625/1250MHz, way short of the intended 750/1500

Ask Charlie to link anything about nV claiming anything resembling a 750MHZ number ever to anyone and he will provide you an ATi provided image. IIRC I thought nV was shooting for 570-600MHZ? That was based on their stated performance metric given the ops per shader core and amount of shader cores. For all I know it may only be clocked at 50MHZ, but Charlie is quoting what ATi thought Fermi might be clocked at before the architecture was even announced as if that was the actual engineering goal.

It will be nice when Charlie is proven wrong in the not too distant future. The comment about manufacturing only a few hundred units- that would push the costs per board into the tens of thousands of dollars, tooling isn't free for PCBs, neither is packaging costs- board vendors wouldn't even consider launching the product at all if the supply was going to be that constrained. I'd say it is within the realm of possibilities still that Fermi won't launch at all in its' original design, but it certainly isn't viable to assume that anyone is going to launch a product that will only ever had a few hundred units into the consumer market in the sub $10K price range.

 

[TheMeanestGuest]

but it certainly isn't viable to assume that anyone is going to launch a product that will only ever had a few hundred units into the consumer market in the sub $10K price range.

Not that it makes it any more viable, but he didn't say a few hundred units, but a few hundred per vendor, which I imagine amounts to several thousand (or, as MarcVenice has been informed ~5000).

Also, while I wasn't particularly interested in video cards at that point in time, I have heard that the x850 XT PE was exceedingly rare, does anyone have any idea how many units were produced?

Edit: Or maybe it was the x800 XT PE, I am now unsure.

 

[evolucion8]

I just hope that Charlie is wrong, if he's right we won't see a nice drop in prices for the HD 5800 series. Because of they performed basically the same or the GTX 280 vs HD 4870 price/performance difference happens again, we won't see a considerable drop in prices, because who will buy a power sucker card that is only barely faster than the ATi's competition?

 

[Kuzi]

Guys Charlie mentioned a few hundred cards "per vendor", so lets say maybe 5k-10k cards worldwide. That's what I find hard to believe, because why even bother release with such a small quantity?

 

[Daedalus685]

Guys Charlie mentioned a few hundred cards "per vendor", so lets say maybe 5k-10k cards worldwide. That's what I find hard to believe, because why even bother release with such a small quantity?

Well, it is not very nice for sure.. but if (big if) Charlie is right, they will try to sell whatever they can. The chips are there and made, it is worth their time to release what they can even if they plan to EOL the line for Fermi 2 almost immediately. Seems pretty certain they would manage to sell every one of the limited run rather easily, which is money they would otherwise go without, and if the things are good performers it at least helps with the mind share on the low end.

 

[dzoner]

Only if they were profoundly bad at business. The GTX2xx parts are still selling, and selling at a premium over what they were selling at a couple months ago. They are closing in on a full flush of their products through the channels right before a new product launch. If the GTX series was OOS all over and their partners had nothing else to sell then it would be wise to start leaking info. Until then, they keep their mouths shut. That is very basic business practice, avoiding write offs on last years products significantly improves operating margins, for nV and their partners.

That is one of those areas where people confuse what they want to see with good business while they are decidedly polar opposites.

The GTX 470/480 competition isn't the GTX 2xx cards, it's the HD 58xx cards.

The GTX 2xx cards were EOLed months ago, the channel is already mostly flushed out (unless Nvidia un-EOLed them) and the revenue trickle they gain from GTX 2xx sales due to their information clampdown on Fermi is a pittance compared to the river of money they are losing to AMD 58xx defectors due to that same policy.

 

[BenSkywalker]

Also, while I wasn't particularly interested in video cards at that point in time, I have heard that the x850 XT PE was exceedingly rare, does anyone have any idea how many units were produced?

Not sure on unit quantity but that was an ATi part(as in BBA)- if nV were going to take the hit on costs and sell their board partners already finished boards then it could be seen as a viable use of resources as nV would write it down as a marketting expense. Board vendors aren't going to do it themselves, and if this were the case we would be certain to see every board from every vendor identical, which the talk of OC varriants would rule out.

Not that it makes it any more viable, but he didn't say a few hundred units, but a few hundred per vendor, which I imagine amounts to several thousand (or, as MarcVenice has been informed ~5000).

If I were nV, and I had a rather limited supply of chips, I'd be focusing on my dedicated partners and shorting those that make competitors parts- I can't see that nV won't be of precisely that mindset for numerous reasons(if a dedicated partner isn't able to produce a high end card at all it puts undue burden on them while those that have a diverse lineup can get chips from team red to protect their margins).

It does seem a bit odd that they would have a number for end of life on the product already, no matter what the situation, for a product that has just started to ramp. Looking at it from a straight business perspective, if I'm only getting 5K chips out of the high end run every one of them is becoming a Tesla/Quadro- it's just smart business.

The GTX 2xx cards were EOLed months ago, the channel is already mostly flushed out (unless Nvidia un-EOLed them) and the revenue trickle they gain from GTX 2xx sales due to their information clampdown on Fermi is a pittance compared to the river of money they are losing to AMD 58xx defectors due to that same policy.

Over the Christmas holiday quarter nV outsold ATi 2:1. Enthusiasts that hold their breath from benchmarks are a miniscule portion of the overall market.

 

[cbn]

the revenue trickle they gain from GTX 2xx sales due to their information clampdown on Fermi is a pittance compared to the river of money they are losing to AMD 58xx defectors due to that same policy.

I think that term "defectors" you are using is interesting.

I don't think ATI realized how many new customers they would get due to Nvidia's 40nm issues.

 

[dzoner]

Guys Charlie mentioned a few hundred cards "per vendor", so lets say maybe 5k-10k cards worldwide. That's what I find hard to believe, because why even bother release with such a small quantity?

Because that's all they've got and since they have to do a base layer re-spin to get chips that are remotely cost competitive, ordering more A3 wafers from TSMC with <10% yields is a no go.

So they sell what they've salvaged and spin, spin, spin until they get something that can compete on price sometime this fall.

 

[BenSkywalker]

Because that's all they've got and since they have to do a base layer re-spin to get chips that are remotely cost competitive, ordering more A3 wafers from TSMC with <10&#37; yields is a no go.

nV isn't going to lose the ability to make a proper business decision because of poor yields. Sell Tesla/Quadro parts and make a nice chunk of money, or sell GeForces and lose a ton? That is what really doesn't make any sense.