NVIDIA GF100 & Friends speculation

[Kaotik]

So why would Nvidia move to the next "generation" of model numbers when they have only just barely released the 4xx series? Wouldn't it make much more sense to go with something like 490, especially given the rumors it's just a GF100b? Does this mean they're planning on renaming the entire 4xx line as well?

Well, they're clearly running out of numbers in 4xx-series, there's little room for full GF104 for example.

So GTX580 = GF100b/GF110, GTX560 = full GF104, lowend... well, renaming wouldn't be the first time, who knows what'll happen there

 

[Silent_Buddha]

So why would Nvidia move to the next "generation" of model numbers when they have only just barely released the 4xx series? Wouldn't it make much more sense to go with something like 490, especially given the rumors it's just a GF100b? Does this mean they're planning on renaming the entire 4xx line as well?

That didn't stop them from almost skipping 3xx series entirely, except for mobile parts in order to make Fermi appear more attractive.

If it has enough changes to qualify as a new chip then it would make sense. At the price point the Asus site seems to be implying, I'm not even sure GTX 580 is just a fixed or full spec GF100. Hell, going from the price, it could just as well be a dual GF104, but then that would in theory be a GTX x90 or x95. But hell, who the F knows. Still I'd lean towards a reworked GF100 with enough changes to qualify as a new "refresh" chip.

And with AMD now also playing with the naming schenanigans, there's no way for a consumer to have any product naming confidence with either company.

Regards,
SB

 

[caveman-jim]

So why would Nvidia move to the next "generation" of model numbers when they have only just barely released the 4xx series?

Because then they can launch the their second generation of dx11 products at the financial analyst day?

 

[Xenus]

It's probably mainly cause they filled up all their space on the lower end already with the 460 465 and 470. What would you call the the 560? All I could see left there is 475 and even then there are just too many parts there.

 

[FenderBender]

Since the specs of GF110 seem to be matching up with being a GF100 respin.. that implies we won't see functional improvements. I'm thinking especially of GF104's improvements over GF100, like 2x execution units for superscalar processing, full speed FP16, and faster L1. (Hits on GF100's L1 cache give significantly lower throughput than shared memory, even though it using the same physical memory, probably due to cache lookup overhead. GF104 does not seem to have this overhead when reading from an L1 cache hit.)

This makes me speculate in early 2011 we could see a double-kicker, a GF120, which uses the same GF104 improvements scaled up to perhaps 576 SPs. GF110 could effectively be a process-only improvement/respin to get a better handle on TSMC's 40nm behavior after a year of experience. GF120 would be a combination of the GF104 arch updates with the GF110 process updates. Pure speculation of course.

A double kicker like this hypothetical GF120 would buy more time for Kepler, which has a long wait for TSMC's 28nm process, which isn't going to be ready for large scale production for a while yet.

 

[Alexko]

It is official Asus China page, and says MSRP.
http://www.asus.com.cn/ContentPage.aspx?Content_Type=staticwebpage&Content_Id=3985

Thanks! That makes it quite expensive indeed… Perhaps NVIDIA intends to get a few orders before Cayman is released.

 

[DarthShader]

full speed FP16

Carsten suggested, that the big performance increase in Vantage could come from this, so it might be implemented already.

This makes me speculate in early 2011 we could see a double-kicker, a GF120, which uses the same GF104 improvements scaled up to perhaps 576 SPs. GF110 could effectively be a process-only improvement/respin to get a better handle on TSMC's 40nm behavior after a year of experience. GF120 would be a combination of the GF104 arch updates with the GF110 process updates. Pure speculation of course.

Believe it or not, but I heard one person, who seemes to have some insider knowledge, saying this would indeed happen... in march already. He also said availabilty for GF110 will be January. Apply a train of salt where neded.

 

[Xenus]

Id they were planning on going that route though you'd think they'd call the new chip the 485 and then Call the one in March the 580.

 

[Bouncing Zabaglione Bros.]

Thanks! That makes it quite expensive indeed… Perhaps NVIDIA intends to get a few orders before Cayman is released.

A few people will pay lots to be one of the owners of a limited edition 580.

 

[FenderBender]

Id they were planning on going that route though you'd think they'd call the new chip the 485 and then Call the one in March the 580.

As frequently proved by history, NVidia marketing is not shy in using any excuse for a new model number. Look at G92, which was released as 8800 GT, 9800 GTX, GTS 150, GTS 250, and GT 330!

The marketing guys will happily grab an opportunity to release a hypothetical GF120 as GTX680.

 

[Xenus]

Oh I'm sure they would but I was more considering the timescale. It doesn't really make sense to release the 580 in december/jan and then the 680 in March unless the 580 is an absolute disaster you wouldn't want to kill it off that quickly.

 

[jaredpace]

 

[FenderBender]

Can we infer anything by the size of the heat spreader about the size of the GF110 chip?
Probably not, I think we're all expecting something very similar in chip size to GF100, perhaps identical.

 

[AnarchX]

GT200A 580mm² and GT200B 490mm² had the same package-size.
But its interesting to see that they can ship A1 of a >400mm² chip.

This is one of the changes over GF100:

Support DTS-HD Audio Bitstreaming

 

[dkanter]

Can we infer anything by the size of the heat spreader about the size of the GF110 chip?
Probably not, I think we're all expecting something very similar in chip size to GF100, perhaps identical.

Probably not. NV is very likely to re-use the whole socket infrastructure, since the pin-out is probably similar. Even if it runs 100W cooler, you'd probably still use the same heatspreader just to avoid unnecessary re-engineering. Of course, you'd use a smaller HSF though...

If you really want to figure out power consumption try counting VRMs, capacitors and other parts of the power delivery system.

DK

 

[neliz]

If you really want to figure out power consumption try counting VRMs, capacitors and other parts of the power delivery system.

DK

It's still 6+2 like the 480.

 

[NathansFortune]

Oh I'm sure they would but I was more considering the timescale. It doesn't really make sense to release the 580 in december/jan and then the 680 in March unless the 580 is an absolute disaster you wouldn't want to kill it off that quickly.

Indeed, GTX 590 will probably be the model number. March is too soon to switch to a new 'generation'. This is assuming that Nvidia can deliver the chip on time. If it gets pushed back then I see no reason not to use the GTX 680 moniker.

 

[Jawed]

So, this is using the 375 binning code. I think it's the binning code.

Which is the same as GF100 had. That binning code indicated a chip that was "severely" cut-down from a fully-capable chip. Perhaps GF110 is also cut-down.

Not necessarily in shaders/ROPs, but in clocks.

Maybe NVidia is going to indulge in some numbering inflation, like the HD6870 fiasco. So GTX580 is the down-binned chip and GTX590 is the response to HD6970. The indications appear to be that GF110 has thermal headroom.

Though we should also expect GTX570, with various bits disabled, shouldn't we?

 

[neliz]

Though we should also expect GTX570, with various bits disabled, shouldn't we?

IF the 580 = 512CC, wouldn't ik make sense the 570 = 480? effectively replacing the 480 with lower temps/consumption.

 

[I.S.T.]

As frequently proved by history, NVidia marketing is not shy in using any excuse for a new model number. Look at G92, which was released as 8800 GT, 9800 GTX, GTS 150, GTS 250, and GT 330!

The marketing guys will happily grab an opportunity to release a hypothetical GF120 as GTX680.

You forget the 8800 GTS and the 9800 GTX+.

Plus the mobile version being at one time the GTX 280!