Semi-Accurate: Fermi A3 silicon is in the oven

[T2k]

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.

Even without direct fab/asic design experience it was published zillion times that it takes approx. 16 weeks from respin to have actual cards ready for shipping (best case scenario.)

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

Sure. Obviously they will, the question is when? The later they do the better for NV and the worse for ATI - up until Fermi is ready for launch: then everything suddenly goes haywire... a strong Fermi held back by TSMC problems is good for ATI but a so-so one is not necessarily, especially if they got their refresh part ready to counter it...

 

[akugami]

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.

I also think that whether Fermi is a "win" or not I think we also have to evaluate nVidia's GPGPU efforts. Early efforts like Adobe adopting CUDA for Adobe CS5 is definitely a win though from what Adobe says on their froums they may be switching to OpenCL or Directcompute for CS6 so that all Adobe customers enjoy the benefits of GPU acceleration.

 

[T2k]

I also think that whether Fermi is a "win" or not I think we also have to evaluate nVidia's GPGPU efforts. Early efforts like Adobe adopting CUDA for Adobe CS5 is definitely a win though from what Adobe says on their froums they may be switching to OpenCL or Directcompute for CS6 so that all Adobe customers enjoy the benefits of GPU acceleration.

Adobe's CUDA support doesn't mean jack*** outside of Huang's warped mind - in CS4 every single feature is available in OGL and since Adobe's interests dictate to try to provide accelerated features to the widest audience (Windows, OS X) with the least R&D cost, they will use the only platform-agnostic and vendor-agnostic API and that's OpenCL.

NV might sponsor a CUDA sidetrack - Adobe won't do it for free, that's for sure -but it's only good to please current CUDA-shops (like we are here ), nothing else...

 

[Wreckage]

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.

The point is Charlie is now saying February and he's mentioned April before. It's clearly just wild speculation on his part. So he's got 3 months covered.

 

[SlowSpyder]

The point is Charlie is now saying February and he's mentioned April before. It's clearly just wild speculation on his part. So he's got 3 months covered.

Febuary is best case? Actual decent retail availability is March-April? No? (according to Charlie that is)

 

[Forumpanda]

The point is Charlie is now saying February and he's mentioned April before. It's clearly just wild speculation on his part. So he's got 3 months covered.

In other words he is saying not january, which some people still hoped for, and he is saying no chance of volume before march at the earliest, and thats not even considering TMSC's yields, which may make it uneconomical for nV to produce volume if they have to compete on price/performance with a much smaller cypress.

Not to mention that as IDC pointed out, at the top end we are now hitting the actual power cap of the PCI-e specs, which means for nV to reclaim the highest performing single card crown they will need to have the best performance/power ratio, not impossible, but so far that has not been their strategy.

 

[akugami]

Adobe's CUDA support doesn't mean jack*** outside of Huang's warped mind - in CS4 every single feature is available in OGL and since Adobe's interests dictate to try to provide accelerated features to the widest audience (Windows, OS X) with the least R&D cost, they will use the only platform-agnostic and vendor-agnostic API and that's OpenCL.

NV might sponsor a CUDA sidetrack - Adobe won't do it for free, that's for sure -but it's only good to please current CUDA-shops (like we are here ), nothing else...

Yeah, but I was talking about Adobe CS5 not CS4. Adobe is indeed adopting CUDA for acceleration of some features in the whole CS5 suite though it is not available in CS4 outside of some plugins. CS5 will contain CUDA acceleration within the Adobe apps themselves.

The point is Charlie is now saying February and he's mentioned April before. It's clearly just wild speculation on his part. So he's got 3 months covered.

I'm not surprised that you don't want to comment on my pointing out that you were attacking Nemesis the person rather than his opinion and comments.

Instead you want to shift it to what Charlie says. That's fine. But what you say about Charlie is invalid. Why? Because Charlie and his ilk are in the rumor business. Their very business is speculation and passing on speculation. They call it a rumor because it is unsubstantiated. The very nature of their business dictates that a lot of what they say will be proven false. And as the rumor mongers hear more or get new information, they have to revise what they say. Standard fare for his type of business. To criticize someone for doing his job is ridiculous.

Charlie has had enough hits on the Fermi issue that we have to at least weigh what he says. He said Fermi would be massively delayed and it was. If I recall correctly, he also said that it would be out in December and at the time it might have even been true and his sources as well as people inside nVidia might have thought this was so. Then more problems arose and the possible release dates got pushed back further and further causing Charlie to revise and update his information.

Charlie is also not saying February is the new release date. He's saying February assuming an absolute best case scenario as he outlined it and with some large risks involved. If you read and are able to understand what he, Charlie, is saying then the only conclusion you can come to is that March is the safest bet for Fermi's release assuming everything goes well. February is only if nVidia takes some risks and April or even later if another revision to Fermi is needed. And again, one needs to take what Charlie says with a grain of salt because he is in the rumor business. Wild speculation and passing on wild speculation is his job.

 

[Wreckage]

I'm not surprised that you don't want to comment on my pointing out that you were attacking Nemesis the person rather than his opinion and comments.

No I simply ignored the comment as what you are implying is false.

Instead you want to shift it to what Charlie says. That's fine. But what you say about Charlie is invalid. Why?

I already made it crystal clear. He's picking such a broad range of dates that it's clear he has no clue. If 4 months ago he would have said that Fermi will be released on March 14th and that came true, I would be impressed. However picking a 3 month span (when NVIDIA already confirmed Q1). Is just spitballing.

 

[Lonyo]

The point is Charlie is now saying February and he's mentioned April before. It's clearly just wild speculation on his part. So he's got 3 months covered.

Yes, he's giving all the possible outcomes.
He's not saying anything particularly new, but based on what's recently happened (A3) he is saying what that could lead to and what it would mean for possible availability of cards.
The reason he's got 3 months covered is because that's all you can say, that there's a likely 3 month window of availability, depending on how smoothly things go.
That's pretty much the POINT of the story. Calling him out on his point is a pretty weird thing to do.

 

[akugami]

No I simply ignored the comment as what you are implying is false.

The fact is you saw fit to ridicule Nemesis and not refute what he actually says. That is a fact. My implication was not false. What I said was correct, instead of proving what he said false you ridiculed him. While his comments are not always clear, he was very clear in what he was trying to say in this thread if you have anything above a 9th grade reading level.

I already made it crystal clear. He's picking such a broad range of dates that it's clear he has no clue. If 4 months ago he would have said that Fermi will be released on March 14th and that came true, I would be impressed. However picking a 3 month span (when NVIDIA already confirmed Q1). Is just spitballing.

You're side stepping the issue. No one knows the future and it's likely that not even nVidia can give you a firm date at this time. You completely ignored the fact that Charlie is in the rumor business and instead are trashing him because he can't give you an exact date? I've already given an example of why by the very nature of his business a lot of what he says will be proven false or ambiguous. With Fermi's release date at least another two months in the future that at this time even nVidia can't give you a firm date on Fermi's release date. If nVidia can't give you a firm date how do you expect Charlie to do so?

The only thing crystal clear is that you're showing bias towards Charlie because he has something to say about nVidia however accurate it may be (and it's still just rumors and speculation) because it is bad news for nVidia and because he can't pin every last detail down. But if it was a rumor that was positive about nVidia, such as Tegra being used for the next Nintendo DS, let's just ignore the fact that it was a rumor and post it as being the next best thing to true. Let's not question it at all.

 

[Idontcare]

I already made it crystal clear. He's picking such a broad range of dates that it's clear he has no clue. If 4 months ago he would have said that Fermi will be released on March 14th and that came true, I would be impressed. However picking a 3 month span (when NVIDIA already confirmed Q1). Is just spitballing.

Not even Nvidia knows if you want to get technical about it. They hope for the best and plan for the worst just like the rest of us.

If someone says March 14th and it just so happens to be the launch day come March 14th that would be pure coincidence (this far out from that date, if today was March 11th then it would be reasonable insider info), not foresight or insider knowledge.

What Charlie has done is the only thing that can be done, a bracketed justified assessment of the feasibility timeline for Fermi's release.

Even TSMC doesn't know when the A3 stepping hotlots will exit the fab, they have bracketed projections based on realistic expectations but nothing can be any more precise than that.

Anyone can spout an opinion..."I predict no Fermi till July 21st!"...it is mostly worthless though unless there is something realistic and reasonable as the basis of the opinion...that is what makes it a justified opinion. Charlie justified his opinion quite nicely, and reasonably, and I have no qualms acknowledging that.

 

[Wreckage]

The fact is you saw fit to ridicule Nemesis and not refute what he actually says. That is a fact. My implication was not false. What I said was correct, instead of proving what he said false you ridiculed him. While his comments are not always clear, he was very clear in what he was trying to say in this thread if you have anything above a 9th grade reading level.

I was replying to his statement, I said nothing about him. Stop trying to make something out of nothing. However, your "reading level" comment, is clearly a personal attack.

You're side stepping the issue. No one knows the future and it's likely that not even nVidia can give you a firm date at this time.

Sigh. He claims to know the future. You keep side stepping the fact that he has already picked at least 3 different months in which Fermi will be released. How you don't see that as just pulling dates out of...somewhere.... Is beyond me. Just as I don't see why you feel compelled to argue with me and try and make things up about me.

 

[Voo]

Sigh. He claims to know the future.

Not more than any scientist who predicts the outcome of a experiment on the basis of a hypothesis.

He may be wrong, but based on the facts and his assumptions his conclusion sounds rather logical. Nobody knows at the moment at which particular date fermi will be released, neither Charlie nor Nvidia.


Changing your opinion based on new facts isn't something bad - your attitude reminds me on a particual quote by Colbert on Bush back in 2006 "He believes the same thing Wednesday as he did Monday. No matter what happened Tuesday".

 

[T2k]

Yeah, but I was talking about Adobe CS5 not CS4.

Me too. CS4 was metioned because they promised CUDA in CS4 as well but never happened. Which part you can't get? :awe:

Adobe is indeed adopting CUDA for acceleration of some features in the whole CS5 suite though it is not available in CS4 outside of some plugins.

There's nothing "indeed" there. You are simply writing your speculations about an unreleased product, based on some very early promises - which, as I've proved already, historically had little to do with final products.

CS5 will contain CUDA acceleration within the Adobe apps themselves.

Or it won't.
And just WTH "within the Adobe apps themselves" supposed to mean? You're not making sense, I'm afraid.

Again, so far this is the same story they said about CS4 - and it never happened there, CUDA never got any exclusivity.

Mark my words: it won't happen in CS5 either, Adobe is not stupid to alienate half of its userbase.

 

[yh125d]

He's not saying "Oh, it will def be out Feb 1 to April 1" and hoping it fits, he's basically saying he can't give an accurate date because of variables that have not been sorted out yet (A3 being good or not, waiting to see what A3 looks like or not, risk wafers needing or not needing scrapped), so he gave out dates based on the diff possible scenarios. It's a bit better than just stabbing in the dark

 

[tviceman]

If A3 is the good-to-go final hardware, then maybe we'll get some benchmark leaks in January and hopefully this extra time has allowed Nvidia to polish the crap out of their drivers in anticipation of Fermi's release.

 

[SlowSpyder]

I already made it crystal clear. He's picking such a broad range of dates that it's clear he has no clue. If 4 months ago he would have said that Fermi will be released on March 14th and that came true, I would be impressed. However picking a 3 month span (when NVIDIA already confirmed Q1). Is just spitballing.

But he has been claiming that we'd be looking at ~March availabilty for some time now. He is not being any less precise than Nvidia... Q1 2010. It seems Charlie knew Fermi would be released in Q1 2010 before Nvidia did, or at least before they wanted to admit it.

Love him or hate him, he's been pretty spot on so far with Fermi.

What I wonder about is clock speed. How much clock speed improvement can be expected in a stepping? If A2 wasn't able to clock high enough to beat Cypress, than what can be reasonalbly expected in A3? Another 100MHz? Another 300MHz? I have absolutely nothing to base this on, but unless A2 -> A3 can improve clockspeeds by a very decent amount, I have a feeling that Fermi will end up not too exciting (from a gaming perspective).

 

[nyker96]

interesting article, even if it's just rumors, at least it's educated guess on the Fermi release date. He seems to say if things go well then March will be likely if not then further back. Personally, I don't think it mattered too much if it's pushed back another month or two, maybe to Nvidia's bottom line it does. Now the question is how NV and ATI is going to split the available TSMC 40nm capacity considering ATI who is using 40nm now is having trouble to fill their existing orders on the DX11 parts.

Another thing is that considering how much more time it takes NV to design this monster chip, I am surprised that they are not going the route ATi is taking with simpler but scalable design instead of one monolithic chip design. Also the size of it will definitely hurt yeild and I can't image they do a X2 with this chip, the cooling, power consumption, cost of construction would be prohibitively high.

 

[akugami]

I was replying to his statement, I said nothing about him. Stop trying to make something out of nothing. However, your "reading level" comment, is clearly a personal attack.

You may have not said anything directly to him but one of your classic tactics is to make implications about the person if you have no valid arguments. You've done this towards me and others with out of context quotes as well as ignoring any facts that we bring up. In this case, you took Nemesis's quote out of context to make it seem like he's speaking nonsense. This is not an isolated case but one that has been going on for a long time. That is wrong. It is not nothing. It is an issue about your attitudes towards other members of this forum when you don't like what they say.

Sigh. He claims to know the future. You keep side stepping the fact that he has already picked at least 3 different months in which Fermi will be released. How you don't see that as just pulling dates out of...somewhere.... Is beyond me. Just as I don't see why you feel compelled to argue with me and try and make things up about me.

What part of "Charlie is in the rumor business" do you not understand? You yourself have pointed out in the past and in this thread that Charlie passes on rumors and speculation and now you're surprised that we can't take what he says at face value?

What Charlie has done is taken the time line and given us a more detailed look at what needs to happen for Fermi to get released. Only you have said Charlie knows the future. Charlie has never made such a claim (at least not seriously I hope). What Charlie has done is take information he has been given, analyzed it, and then gave his thoughts on it. In this particular article, what Charlie has done is no different than what someone at a large research and analysis company like Gartner, Forrester, and IDC would do. Everyone and their grandmother except you acknowledges what Charlie has done and thought the article was interesting.

Lastly, I don't need to make things up about you. Your past history is littered with lies and deception. I'm not here to dig back through every single thread where it has been shown, with proof, how you have lied and falsified information as well as your history of attacking another person when you have no valid counter argument to what the other person is saying. Your claim that I'm making stuff up about you is just an example of that. Your comment towards Nemesis is another example of it. While I am in no way perfect nor am I free from making mistakes, I don't give out falsified information or lies on purpose and say it is the truth.

Likely done with this thread because, as usual, you refuse to see the truth due to your bias and I don't have the rest of the week to refute your false claims and biased opinions nor do I want to get dragged into personal attacks as you are attempting to do. So unless someone else has an interesting comment or provide further information I'm keeping out.

 

[Grooveriding]

NV has lost this round by coming so late to table with their new offering. I have no preference company wise, before my current setup, I was using a GTX 285, before that a 260-216, before that a 4870, before that an 8800GTX etc.

This is moreso true whereby we have reached a sort of dead zone in terms of what games demand vs the hardware available to us. I game at the max res available to the pc gamer, my current setup plays everything fully maxed at this res with the exception of Crysis which I have to use 2xaa in to get about a 40fps avg.

NV coming out with their new cards in March will offer me nothing beyond at best, 10 more fps in Crysis or the ability to run with 4xAA by buying two of the top end Fermi cards.

What is the point in this ? There are no new game engines on the horizon to rival CryEngine 2, CryEngine 3 is said to be less demanding than it's predecessor.

Compound this with ATI already having their cards out for a good six months before NV brings us anything and then further compound this with ATI's promise of the release of their 6XXX series in Sept 2010, which will with a 99% chance turn out to be faster than Fermi.

Really at this point to the educated video card consumer, there is no point in upgrading anymore. If you game at 1920x1200 or less and have a 5870, why upgrade ? NV's new cards are only going to sell to people with brand preference.

We need an innovative new game engine at this point to make -any- new video cards on the horizon worth buying, as it stands now, we have more power than we need. Interesting situation.

 

[Idontcare]

What I wonder about is clock speed. How much clock speed improvement can be expected in a stepping? If A2 wasn't able to clock high enough to beat Cypress, than what can be reasonalbly expected in A3? Another 100MHz? Another 300MHz? I have absolutely nothing to base this on, but unless A2 -> A3 can improve clockspeeds by a very decent amount, I have a feeling that Fermi will end up not too exciting (from a gaming perspective).

If there is anything critically disabling about the speedpaths in A2 that precipitated the respin to A3 in the first place then sure we could see a considerable speed-bump in A3 vs. A2.

I seem to recall some of the A-steppings of Phenom had speedpath issues that limited clockspeeds to something around 800Mhz, by the time that B-stepping came around they were up to what? 2.6GHz or so at time of release?

(funny, didn't intend for it but a phenom vs. Kentsfield comparison is apropos on so many levels including the post below)

Now the question is how NV and ATI is going to split the available TSMC 40nm capacity considering ATI who is using 40nm now is having trouble to fill their existing orders on the DX11 parts.

$, and committments for future contracts. (loyal customers get rewarded)

Another thing is that considering how much more time it takes NV to design this monster chip, I am surprised that they are not going the route ATi is taking with simpler but scalable design instead of one monolithic chip design. Also the size of it will definitely hurt yeild and I can't image they do a X2 with this chip, the cooling, power consumption, cost of construction would be prohibitively high.

Not sure if we are seeing the effects of designing a monster chip or the effects of implementing a new architecture. They way these things scale up and down with core counts I imagine a lot of the work is done by synthesizers and automated layout iterated with some hand layout done to remove speedpath bottlenecks.

But yeah why they didn't go for a 1.5B chip (1/2 Fermi) first and much earlier followed up by a 3B chip once the yields supported it is a good question. I can only assume they had great confidence in TSMC meeting their internal commit dates and Nvidia was counting on them to do just that.

 

[PingviN]

It's a damn shame TSCM screwed AMD over on their best chance at a comeback.

That Fermi is out sooner than later is good news, that will hopefully give the pricing of Cypress a good kick in the head.

 

[Nemesis 1]

as anyone printed or shown how many chips AMD/ATI are actually recieving???

If so link it. Not heresay . Real facts. Of course there will be shortages in retail when OEMs are sucking up all the parts.

If I was seller I would sell to OEMS first and formost. Retail supply is great for us. But its not were the $$$ are . Thats in OEMs

PingIN your post is a little obscure. So what is it you want . A faster NV part so you can buy a cheaper ATI part. Or so you can buy an even higher priced NV part?????

By the way I love your handle . I just love to pingin it really raises hell with the server and pisses the guy off who gets pingout.

If its the later of the 2 which makes zero sense and invalidates your thinking ATI parts are overpriced right now. There is no refutting this its one or the other no inbetween no fence walking here. I have seen were many want it out so they can get a better deal on ati part. Which makes sense. All this ATI hasn't parts forsales is getting really old . Apple /Intel did the same thing . and the crying was just as amusing as this is.

 

[Lonyo]

List of shipping NV 40nm parts:
Geforce 310
GeForce G210M
GeForce G220M
GeForce G230M
GeForce G240M
GeForce G250M
GeForce G260M

I know it might seem crazy to you guys, but there's more to 40nm than the HD5xxx series, from both NV and ATI.
TSMC is going to split 40nm production probably along lines similar to the way they are already doing, I would guess.

 

[Nemesis 1]

OH! No . Is the 310 part in retail ? I am in trouble if it is. Sleeping on the job .

I want out of this consumer section anyway . There bigger fish to fry in the Home consumer sector . But do you have a link to were I can buy a 310. or even a review of the 310 . link please i been busy in the home consumer section and let this slip by.