GF100 Fermi specs speculation at B3D

[tviceman]

Are the mainstream fermi cards speculated to come out at the same time as high end with that supposed march release?

The only thing I've heard of is the two big dogs - the gtx 380 and 360. I think it's going to be a few months after before we see mid-range fermi stuff.

 

[T2k]

Are the mainstream fermi cards speculated to come out at the same time as high end with that supposed march release?

We don't know anything about any regular consumer cards but Teslas are already on pre-order, they even launched a "buy now, trade-in" program: http://www.nvidia.com/object/mad_science_promo.html

Essentially if you buy a Tesla C1060 (~$1,300) or s1070 ($10,000+) now then you can trade it in at the current full price when you order the news Tesla 2xxx-series when it arrives.

 

[Lonbjerg]

Ehh? 23% is what I'd consider the bare minimum 6 mmonths later to tell you the truth.

And prices won't fall because NV badly needs to recoup their enormous R&D costs - too bad, I would happily pick up a second 5870 for $200 or a 5970 for $400...

So if AMD's refresh falls shorts of this, you will be sadly dissapointed?
Or does your logic only apply to NVIDIA?

(Funny how the performance of cut down engineering samples is making a lot of people backpedal...and the tessalation power of Fermi is making AMD's "lead" look funny..and unimportant)

This reminds me of the G80 launch...the comming NDA-lift will make it even more funny....and the next NDA-lift will make it VERY funny.

In before the backpedal-mode of the doomer&gloomers

 

[Lonyo]

So if AMD's refresh falls shorts of this, you will be sadly dissapointed?
Or does your logic only apply to NVIDIA?

(Funny how the performance of cut down engineering samples is making a lot of people backpedal...and the tessalation power of Fermi is making AMD's "lead" look funny..and unimportant)

This reminds me of the G80 launch...the comming NDA-lift will make it even more funny....and the next NDA-lift will make it VERY funny.

In before the backpedal-mode of the doomer&gloomers

Can you repeat that in English please?

Also, ATI (and if you look back you will see I have posted this before), aren't really aiming to have the fastest single GPU, so NV should have a reasonably much faster product, otherwise they've failed.
The place that they might not be able to compete will be cost (exactly same as last gen), as their products will, by design, cost more to make.

 

[Lonbjerg]

Can you repeat that in English please?

Please quote the parts that confuses you...and drop fallacies.

Also, ATI (and if you look back you will see I have posted this before), aren't really aiming to have the fastest single GPU, so NV should have a reasonably much faster product, otherwise they've failed.
The place that they might not be able to compete will be cost (exactly same as last gen), as their products will, by design, cost more to make.

AMD is in dire need af profits, guess who a pricewar will hurt the most?
Those with figures in the red (pun intended)

 

[Lonyo]

Please quote the parts that confuses you...and drop fallacies.



AMD is in dire need af profits, guess who a pricewar will hurt the most?
Those with figures in the red (pun intended)

I'm wondering what fallacies need to be dropped.
And the financial situation of AMD will be revealed on Thursday, so commenting about them being in the red can hold off until then, since their graphics division made a profit last quarter.

Also most of your post confused me. I'm not sure whether you were trying to make coherent points against multiple people, or just try and imply that every aspect of Fermi will be superior to anything from ATI.

Also you seem to be confusing making a fast part with making a part which is good for NV.
Yes, GF100/whatever it's going to be called will be fast, should be fast.
But it's one part, or maybe two parts (GF104?). That's it.
ATI will have a top to bottom lineup of DX11 40nm parts before NV has a single DX11 part available to the public, which will be good for design wins, good from a cost point of view, and (one would assume) can only benefit them. NV will have a single high end 40nm product, and a smattering of other parts using various different process technologies and having various PCB and power requirements which pretty much necessitate them costing more than equivalent ATI parts to make.

The G80 managed to wipe the floor with ATI and start at under $300. Fermi isn't G80.

 

[evolucion8]

The G80 managed to wipe the floor with ATI and start at under $300. Fermi isn't G80.

And certainly, the HD 58x0 series isn't R600.

 

[T2k]

So if AMD's refresh falls shorts of this, you will be sadly dissapointed?

If ATI's next architecture will be 6+ months late and less than 23%-25% faster than Fermi (or whatever will be Nvidia's latest offering when R9xx arrives) then yes, I'll be pissed.

Or does your logic only apply to NVIDIA?

Nope. It applies to new architectures only but I have to admit it is impressive to see how ATI did it with its 58xx-line too, despite not being a new architecture but simply an updated 48xx-derivative.

(Funny how the performance of cut down engineering samples is making a lot of people backpedal...and the tessalation power of Fermi is making AMD's "lead" look funny..and unimportant)

Did I miss something? Leaked rumors rather point to lack of solid lead, most likely GTX360 will be a match for 5870 while GTX 380 will be slower than 5970... got linky?

This reminds me of the G80 launch...the comming NDA-lift will make it even more funny....and the next NDA-lift will make it VERY funny.

In before the backpedal-mode of the doomer&gloomers

I'm afraid I'm not following you...

 

[T2k]

And certainly, the HD 58x0 series isn't R600.

True. It's much better, let alone it's not a new architecture.

 

[Paratus]

I think the new GF midrange GTX 350 ++ GOOD will be based on the brand new G192B with 128 SP & 256 bit mem. It will support a new and streamlined DX 11 -1.

But seriously has any one heard a rumor about an NV DX 11 midrange?

 

[tviceman]

I think the new GF midrange GTX 350 ++ GOOD will be based on the brand new G192B with 128 SP & 256 bit mem. It will support a new and streamlined DX 11 -1.

But seriously has any one heard a rumor about an NV DX 11 midrange?

I haven't heard ANYTHING regarding this.

 

[T2k]

Correction: we haven't heard ANYTHING.

 

[Paratus]

So we may have 2 highend parts from NV & then nothing for several more months?

What do they think they're going to sell for $300 and inder market?

 

[Lonyo]

I think the new GF midrange GTX 350 ++ GOOD will be based on the brand new G192B with 128 SP & 256 bit mem. It will support a new and streamlined DX 11 -1.

But seriously has any one heard a rumor about an NV DX 11 midrange?

It's entirely possible that there won't be one.
Given that the low end 300 series has started off being 40nm DX10.1 parts, there's nothing stopping NV just redoing existing parts at 40nm and naming them 3xx parts, much like they did with the GTS250.

Just because AMD had a top to bottom DX11 roadmap laid out and almost launched, doesn't mean NV have to follow suit.

 

[T2k]

So we may have 2 highend parts from NV & then nothing for several more months?

What do they think they're going to sell for $300 and inder market?

Have you been around during HairForceFX times?
After similarly screwed-up development process & missed quarters they have announced it but technically they never shipped en masse, only a few thousand cards reached the shelves and never been any derivate made of it - they quickly
brought out its replacement.

I dont' think Fermi is such a mess, don't get me wrong, it will have decent performance unlike HairForceFX did but I can imagine they will postpone the cheaper parts until the can ship it en masse & to re-capture the lost high-end market first... dunno, I'm not a marketing nor a sales guy, I don't know which one will make more sense.

 

[Paratus]

Ya but during th FX days they released the 5200 & 5600 which were at least architecturally similar. The lower level cards NV has today are all DX10 or 10.1.

 

[tviceman]

It's entirely possible that there won't be one.
Given that the low end 300 series has started off being 40nm DX10.1 parts, there's nothing stopping NV just redoing existing parts at 40nm and naming them 3xx parts, much like they did with the GTS250.

Just because AMD had a top to bottom DX11 roadmap laid out and almost launched, doesn't mean NV have to follow suit.

I really don't know how complicated chip development/alteration is, and especially with Fermi since it's a huge chip, so the following statement I am about to make is based merely on a common sense guess.

Since Fermi has so many HPC specific features (lots of cache, ECC, etc.) I would imagine their midrange chips would be entirely stripped of some of those aspects (instead of just being switched off). And I would also imagine it would be easier to strip features away from a chip than to add features to it.

With that in mind, hopefully in a few months we'll see DX11 midrange stuff.