Second generation Fermi to be launched at the end of the year

[CAT-THE-FIFTH]

If everything works out fine Nvidia will be releasing the second generation Fermi chips in late 2010:

http://www.fudzilla.com/content/view/17387/34/

[shaithis]

Absolutely stinks of a knee-jerk press release to try and steel some of the limelight from AMDs 28nm in Q3/4 announcement.

Perhaps nVidia should get Fermi out before talking about release dates for the next generation? lol

I know that this game is played via press releases....hell, how many "paper launches" have we seen? but come on.....get this generation out first!!

[Zak33]

/\ what he said.. with bells on tbh

[mikemikemi]

Absolutely stinks of a knee-jerk press release to try and steel some of the limelight from AMDs 28nm in Q3/4 announcement.

Perhaps nVidia should get Fermi out before talking about release dates for the next generation? lol

I know that this game is played via press releases....hell, how many "paper launches" have we seen? but come on.....get this generation out first!!

Couldn't agree more. Plus it's fud, so will take with a tonne of salt.

[scaryjim]

... it's fud, so will take with a tonne of salt.

You'll be lucky to find that much salt - the entire country's stock was put on the roads, wasn't it?

But back on topic, this is double-plus-ungood from nvidia: seriously, bigging up the next generation when the previous generation isn't even launched? Is it just me, or does this sound like "the next round of launches are gonna suck ass, but we'll make them better a few months later" to anyone else? Talk about getting your excuses in early...

We also learned that the mainstream Fermi won’t be significantly delayed from the original plan

The original plan was to release Fermi almost two years after their previous new generation consumer cards? 6 months after they "detailed" the technology at GTC? To give ATI almost an entire generation's advantage?

And let's not forget that their stated roadmap for Fermi was to put it into Tesla and workstations before releasing "consumer" GeForce parts. If they stick with that strategy we'll see Tesla 2000s in March, but still don't know when that will derive down to a retail GTX380...

[Funkstar]

I've not been following this Fermi business very closely, so excuse my ignorance.

When is Gen 1 supposed to be out? I realise it was supposed to be out a while ago now, but when are we actually going to see parts? any ideas?

And they have pre-announced the replacement already to the end of the year. What's the chances of anyone actually buying the first gen product at all now? I know I wouldn't bother.

[dangel]

History repeating: big hot product out first, refined smaller/cooler product out later.

It's to be expected, but as to whether I get it - just down to whether it's needed or actually better than ATI's offering at the time.

[scaryjim]

When is Gen 1 supposed to be out?

The last estimate was by this March at the latest, I believe; from an "actually making money" point of view Gen 1 was due before Christmas (IMNSHO, obviously). Ati are about to complete their top-to-bottom DX11 line up before NVidia have even one card out for their next generation, and I'm still very sceptical about their ability to derive mid-range cards from the Fermi architecture: they have completely failed to do this with the GT200 core, to the extent that a 96core GT240 with 128bit DDR5 RAM is unable to significantly outmuscle a HD4670. Seriously, it's got twice the memory bandwidth of a 4670, and 50% more cores than a 9600GT. What *exactly* is holding it back? Not very clever, NVidia...

[=assassin=]

So basically with this news, why would anybody want to buy the '1st gen' product when the next one is coming quickly after.

[mikemikemi]

So basically with this news, why would anybody want to buy the '1st gen' product when the next one is coming quickly after.

Very true if the rumour is accurate, by the time Fermi arrives in reasonable quantities gen 2 will be round the corner.

Gen2 will likely be on a smaller process though and Nvidia would probably prefer to sell these for better margins.

Maybe Gen1, is just to get the product out the door as something to show and Gen2 is the real deal?

[TooNice]

So basically with this news, why would anybody want to buy the '1st gen' product when the next one is coming quickly after.

That's exactly what I was going to say as I scrolled down to type my answer. Assuming that the Fermi is not going to be a repeat of, say the HD2900XT amongst other late yet underwhelming releases.

[GheeTsar]

I can't see how this will work. If yields are poor on the current process, then a die shrink is likely to make things worse and I can't see them redesigning the architecture just yet.

At best I think it will be a tweaked version of the existing card on the same process with a view to increasing clock speeds and yields.

[scaryjim]

I can't see how this will work. If yields are poor on the current process, then a die shrink is likely to make things worse and I can't see them redesigning the architecture just yet.

Any thoughts on whether they could produce a Fermi-sized die on a larger process, e.g. 55nm? Just a thought, but if they're struggling to yield @ 40nm could they launch @ 55nm then shrink to 40nm in 6 months? And if they tried that, how much of a power hog would this beast be?

I'm mad aren't I - they can't actually do that. Can they?!?

[GheeTsar]

Any thoughts on whether they could produce a Fermi-sized die on a larger process, e.g. 55nm? Just a thought, but if they're struggling to yield @ 40nm could they launch @ 55nm then shrink to 40nm in 6 months? And if they tried that, how much of a power hog would this beast be?

I'm mad aren't I - they can't actually do that. Can they?!?

If rumours are true, the current die size is already almost at the 300W limit for PCIe, so it would be hard to imagine going back up without a lot of the chip being disabled. It would also be an major embarassment.

[scaryjim]

Anyone know how big a deal it is to transfer between processes with a design? Does the design specifically target a given node / half-node and have to be completely reworked for a transition, or can you basically take an existing design and just make it smaller?

[MadduckUK]

its not an option for many reasons but to condense them down - because that would be bat****