Nvidia's mainstream Fermi parts arrive in late Q1

Quote:

Originally Posted by saaya

right, but oems wanted that last christmas or maybe in early 2009, and its not like they only told them then, they probably told nvidia in spring 2008 that they want 10.1 in winter 2008 or spring 2009... but it took nvidia until NOW to finally get it done. so again, how is that flexible?

how can you say a product that took more than a year from concept to retail is a good example of how flexible the architecture its based on is? 2 or less quarters is already not exactly flexible... 1 quarter or less is kinda flexible... 2 months or less is VERY flexible...

or what, are you claiming nvidia only started working on gt2xx 6 months ago? :rolleyes: sure...

you cant really compare gt200 and gt2xx with rv870 and rv840 though, cause rv840 is a cut down rv870, gt2xx is a reworked gt200... they added a new imc, dx10.1 and probably tweaked something else... thats obviously going to take a lot more work and time to do... the problem is that nvidia hasnt done any cut down chips in ages, they always did reworks... they always changed the architecture and didnt just chop some parts of, which takes a lot more time... a chopped down gt200 might be possible within 2 months like ati did it, yes, gt200 MIGHT very well be very flexible... but how could we know? even nvidia cant know cause they never even tried it afaik... so how can you say gt200 is flexible if they never created a cut down part of it and brought it to market quickly... they did a very slow rework... drawing a conclusion from that to how fast they are at cutting down a chip by chopping some bits off is impossible... but if anything youd think itll take them a while as their rework wasnt exactly a fast one...

this might be the main issue they are facing actually, as that dragged their entire roadmap along and slowed everything down...
instead of cutting a chip down and THEN reworking/refreshing it, they always did both at the same time... they basically tried to do a tick and tock at the same time... and thats really stupid... both intel and amd learned a while ago that you really dont want tick and tock at the same time...

The main point here is that they spent a lot of time with figuring out how to scale down the GT200 and have created Fermi in such a manor that it will be very easy to scale down, and Nvidia has confirmed this in either some press report or an interview, I forgot where.

But point being is they have learned much from GT200 and applied what was learned to Fermi.

Quote:

Originally Posted by 003

But point being is they have learned much from GT200

Yes.

Quote:

Originally Posted by 003

and applied what was learned to Fermi.

No.

I think fermi is too soon to apply 'what you learnt'... it was already either designed as scalable, or it wasnt. It'll be at least GT400, possibly 500, before the lesson learnt shows up in the architecture.

People seem to consider dev times so selectively around here.

Quote:

Originally Posted by 003

The main point here is that they spent a lot of time with figuring out how to scale down the GT200 and have created Fermi in such a manor that it will be very easy to scale down, and Nvidia has confirmed this in either some press report or an interview, I forgot where.

We'll know how easy Fermi will actually scale down by how long it will take Nvidia to have value parts based on Fermi.

And I declare that Fermi should be now known as Hermi, until it's actually available.