Semi-Accurate: Fermi A3 silicon is in the oven

[Idontcare]

Fermi A3 silicon is in the oven

IT LOOKS LIKE Nvidia has put the A3 stepping of Fermi in the oven, it happened some time last week. That means that you may see cards as early as February, depending on two or three other factors.

http://www.semiaccurate.com/2009/12/10/fermi-a3-silicon-oven/

A good read, worth hitting the link.

No fire and brimstone, and the logic tree timeline is well laid out.

Good on ya Charlie

 

[nitromullet]

That really is a pretty balanced view on the situation. Everyone should give him a page hit to show that we're really more interested in quality pieces instead of tin foil hat wrapped ramblings.

 

[Arkaign]

Wow, color me shocked, perhaps he's turned a newer, more mature page. What a transition!

 

[Hauk]

A refreshing read. Especially interesting is that last part about clock speeds and their relation to 5870 performance comparison. Stay tuned..!

 

[Idontcare]

The clockspeed numbers are believable too given the power-consumption numbers we've seen from Cypress and the xtor counts involved.

Some magic pixie dust is going to need be involved for a performance-compelling X2 version to be created. I mean sure they can make an X2 and put the clocks down to 400MHz at stock, but will it outperform a 5970 also at stock?

That 300W power-consumption limit is gonna be a bitch with X2 Fermi.

 

[tviceman]

The clockspeed numbers are believable too given the power-consumption numbers we've seen from Cypress and the xtor counts involved.

Some magic pixie dust is going to need be involved for a performance-compelling X2 version to be created. I mean sure they can make an X2 and put the clocks down to 400MHz at stock, but will it outperform a 5970 also at stock?

That 300W power-consumption limit is gonna be a bitch with X2 Fermi.

Which is the same reasons why they couldn't do two gtx 280's or 285's for one card.

Their X2 will likely be a gtx 360 in tandum - but then again they may have to wait until a die shrink until they can do it.

 

[ronnn]

Everyone here loves him when he says what they want to hear. Back in the summer when he said no fermi in 2009 - equally balanced - but different reaction.

That said good news that nvidia is on track. Lets hope tmc follows.

 

[scooterlibby]

Back in the Summer I remember it being quite a bit more vitriolic and everyone knew he had an axe to grind. It does seem like a tone shift, regardless of what the content is.

 

[Nemesis 1]

Good post IDC. I am a bit surprized Charlie is or seems to think NO A4. CHarlie! Thank you for the opening . NO way will A3 stepping work. I believe A4 maybe . But A5 is likely. I have no axe to grind . I just think NV bit off way more than they can chew without choking . Thanks again Charles for the opening . Now lets see who has the better insight.

 

[ronnn]

He had lots to be vitriolic about. Nvidia was leaking all over the place about a launch with win 7 and etc. At one point they were even showing a mock up and pretending it really worked. Do you think anand would have been kind if he had access to the same info (not under nda)? Don't forget this followed bumpgate and all the lies around that.

Now they have cleaned up their act and should be on target for announced release dates.

 

[AnandThenMan]

NO way will A3 stepping work. I believe A4 maybe . But A5 is likely.

What makes you think A3 is not going to be a shipping part? You could very well be right I don't know, it just seems that historically A3 ends up being the final spin. Reasons?

Going all the way to an A5 revision would put Fermi into the territory of being so late it is bordering on meaningless, don't think that is going to happen.

 

[novasatori]

Good post IDC. I am a bit surprized Charlie is or seems to think NO A4. CHarlie! Thank you for the opening . NO way will A3 stepping work. I believe A4 maybe . But A5 is likely. I have no axe to grind . I just think NV bit off way more than they can chew without choking . Thanks again Charles for the opening . Now lets see who has the better insight.

:hmm:

 

[jvroig]

From the article:

If all goes well, risk wafers valid, A3 production went in on Dec 1, and all the stars align, you are looking at Dec 1 +6 weeks +2 weeks, or about Feb 1 for real availability. More realistically, if it waits until the A3 hot lots get back, that would be Jan 1 plus eight weeks, or March 1 for a hard launch. Basically, the best case is Feb 1 for a real launch.

Who'd have thought such best case news would come from Charlie? Charlie ripped into them like a hungry lion on an overfed water buffalo when they were dropping the ball, but he seems to be calling it as it is right now that they seem to be getting back on track. Not sure if he's just doing his job or just trying to boost his cred. Either way, I'm happy to hear good nVidia news coming from an anti-nVidia source.

I'm sure they're aiming to beat Cypress, but there's more than one way to beat a card. Raw performance is great, but performance/power ratio is great as well. Even if they are just on par with Cypress in raw performance, but offer much leaner power consumption at load, that's a pretty big win already.

To ask the tech experts here, when we say "Fermi has 50% more transistors than Cypress" (or whatever percent, I forgot the figure), which two cards are we actually comparing? The 5870 Cypress (not the 5850, I assume) versus which Fermi chip - the chip that would be the GTX375? Since Fermi will actually have an entire line (scale up and down, right?), transistor counts will vary greatly from one specific card to another, but will this "50% more transistors" estimate hold true for almost all product comparisons versus ATi as long as proper chips are matched together? (I mean, budget Fermi vs budget ATi, mainstream Fermi vs mainstream ATi, etc)

 

[jvroig]

Going all the way to an A5 revision would put Fermi into the territory of being so late it is bordering on meaningless, don't think that is going to happen.

A5 would mean about 8-12 weeks more, according to the article, so if Feb is best-case now, best-case then would be April or early May. Not really that much more of a delay than was previously estimated by other sources aside form Charlie.

But even Charlie seems to think A4/A5 is very unlikely. Perhaps Nemesis is just in a gambling mood.

 

[Janooo]

He had lots to be vitriolic about. Nvidia was leaking all over the place about a launch with win 7 and etc. At one point they were even showing a mock up and pretending it really worked. Do you think anand would have been kind if he had access to the same info (not under nda)? Don't forget this followed bumpgate and all the lies around that.

Now they have cleaned up their act and should be on target for announced release dates.

You wish.

 

[Janooo]

I don't see much difference between this article and his previous reports. The same stuff.
It's just that many of you finally recognize that he is very accurate about Fermi.

 

[akugami]

I don't see much difference between this article and his previous reports. The same stuff.
It's just that many of you finally recognize that he is very accurate about Fermi.

I think for many, it's not the content, but the tone of voice. You can occasionally open up the whoop ass on a company when they start messing up and trying to cover up mistakes with lies such as bumpgate or when a company, like Sony, starts churning BS and attempting to shovel it onto us. It's just when there are general problems that can pop up in any production facility we shouldn't really castrate them for it. Fermi would fall into that category as would the (I believe) Radeon X1800 series.

 

[Wreckage]

Wow, color me shocked, perhaps he's turned a newer, more mature page. What a transition!

As long as you ignore that he was saying March/April at the earliest.

It's like saying "pick an number between 1 and 10" and he picks "1,2,3,4,5,6,7,8,9,10'. The when you say "6", he says "see I was right". D:

 

[Wreckage]

NO way will A3 stepping work. I believe A4 maybe . But A5 is likely.

They are not playing Battleship or Bingo or whatever you post seems to imply.

 

[akugami]

As long as you ignore that he was saying March/April at the earliest.

It's like saying "pick an number between 1 and 10" and he picks "1,2,3,4,5,6,7,8,9,10'. The when you say "6", he says "see I was right". D:

IDC who has industry experience says that the time line is plausible if what Charlie says about nVidia only now just "put the A3 stepping of Fermi in the oven." If you've got a link to a more plausible time line then lets hear it.

If what Charlie says is true, March really is when we should be expecting Fermi to hit retail. And again, IDC can't seem to find fault with any of his logic in this particular article and yet again, IDC has industry experience.

Now, Charlie is in the rumor business and I'm sure he's not above taking 2+2 and saying it equals 5 but some of his past rumors have been very accurate and as I've stated previously in a different thread, when you're in the rumor business, hitting above 30% is damned good. If you hit 50% that's amazing. Charlie always seems to have an axe to grind and spins any nVidia news negatively just like how certain posters on this forum will spin any ATI news negatively. But you have to respect him for breaking news on a few really big tech stories that have proven true.

These are still rumors and conjecture and even someone like IDC can't accurately say what will or won't happen since he is not involved with nVidia and he does not have insider information.

They are not playing Battleship or Bingo or whatever you post seems to imply.

You also knew exactly what Nemesis 1 was talking about with his comment about the A3 stepping not being the final one and that an A4 or even A5 stepping would be needed. Another instance where you don't have a valid counter argument/point so instead you attack the person making the comments.

 

[Nemesis 1]

They are not playing Battleship or Bingo or whatever you post seems to imply.

http://www.youtube.com/watch?v=fEkWH8DB7b0

 

[SlowSpyder]

If Fermi launches in Febuary, more than 5 months after the 5870, and doesn't convincingly outpace the 5870, is this a 'win' for Nvidia? I would think that enthusiasts not loyal to Nvidia would have already upgraded... the holiday season is well past by then. AMD will have enjoyed 5+ months of no competition in the high end (It looks like AMD's supply problems are improving, multiple brands of everything but the 5970 is available on Newegg right now).

Now if Fermi is 50% faster, is priced right, and there are games that are out or on the horizon that justify even more video power, then I think Nvidia will be ok. But I get the sense that by the time Fermi arrives it might be too late for Nvidia to salvage this round.

 

[Creig]

Even if Nvidia has put their stamp of approval on the A3 stepping of Fermi, they are still going to have to contend with TSMC and their oh-so-wonderful 40nm yields. Since Fermi has a much larger die than Cypress, this means that Nvidia will have even worse yields than AMD is currently experiencing.

Will TSMC be able to produce Fermi cores in high enough quantity and low enough price per functional die to make the GF100 economically viable?

 

[Keysplayr]

Even if Nvidia has put their stamp of approval on the A3 stepping of Fermi, they are still going to have to contend with TSMC and their oh-so-wonderful 40nm yields. Since Fermi has a much larger die than Cypress, this means that Nvidia will have even worse yields than AMD is currently experiencing.

Will TSMC be able to produce Fermi cores in high enough quantity and low enough price per functional die to make the GF100 economically viable?

TSMC will eventually get their act together. For both of the 40nm customers.

 

[Idontcare]

If Fermi launches in Febuary, more than 5 months after the 5870, and doesn't convincingly outpace the 5870, is this a 'win' for Nvidia? I would think that enthusiasts not loyal to Nvidia would have already upgraded... the holiday season is well past by then. AMD will have enjoyed 5+ months of no competition in the high end (It looks like AMD's supply problems are improving, multiple brands of everything but the 5970 is available on Newegg right now).

Now if Fermi is 50% faster, is priced right, and there are games that are out or on the horizon that justify even more video power, then I think Nvidia will be ok. But I get the sense that by the time Fermi arrives it might be too late for Nvidia to salvage this round.

Now I understand that the viewpoint of "a win" is different be it from a consumer's perspective versus Nvidia's, but the design decisions, trade-offs, and targets were based on Nvidia's perspective so I think it only fair to evaluate the release timeline as either a failure or a win from their perspective as well.

That said, what would determine whether it is a win to come out 5months later than AMD's Cypress comes down to the revenue differential (ASP), cost of production differential (margins) and the investments made in Fermi itself (R&D).

Consider a similar analysis made on AMD vs. Intel for CPU's. Sure AMD tends to be about 1 year behind Intel on node transitions and thus about a year behind on introducing new products (be it architectural refreshes like Phenom II or new microarchitectures like bulldozer)...but it is already accepted that given the disparity in R&D resources AMD is actually doing pretty well to only be that far behind the leading edge.

Now for AMD vs Nvidia we don't have anywhere near the R&D resource disparity, and for all we know it favors Nvidia just going by marketshare and graphics division revenue numbers, but I'm just saying the same sort of "rationalization of investments versus returns" is needed in order to really come out and say their decisions to make those investments in an effort to garner those returns turned out to be a "for the win" outcome.

Given TSMC's node ramp conundrums (which is not something Nvidia planned on, so we can't chalk that coincidence in timing as being brilliance on NV's behalf) how much revenue disparity do we really think Nvidia is incurring from the delayed introduction of Fermi SKU's?

I'll just say from a pragmatic "sustainable business model" viewpoint that if neither of these companies can get their gross margins above 40% and deliver an earnings engine that can consistently generate profits then I don't care what their delivery timelines are like (or the excuses they give for missing them) the entire affair is "for the loss" for everyone.

R&D for tomorrow's GPUs are being resourced by today's revenues generated by today's products. Lackluster revenue on today's products means less resources for developing tomorrow's products.

It is an awesome feedback loop when the company is profitable and growing, a vicious feedback loop when the opposite is transpiring.

Will TSMC be able to produce Fermi cores in high enough quantity and low enough price per functional die to make the GF100 economically viable?

I dare say this question is on more people's minds than the combined headcount of the gaming community the world over.