NV30 Update

[bdmosky]

and then there is a Geforce 3 Ti which was a die shrink and....oh yes enabled features the 1st Geforce 3 was marketed as having..

Geforce 3 Ti were not die shrunk, but "switched" to a new refined 0.15-micron process at TSMC, whatever that is supposed to mean. And 3d textures were fixed before the Ti series of cards were released.

Why I'm telling you something you already know? I have no idea. Perhaps so this conversation doesn't get any more confusing for those who can't access publicly known information on there own.

 

[Bigus Dickus]

Your definition of refresh is obviously quite different from what is normally used, and is thus wrong.

Here's a little bit of history:

1. TNT -> TNT2: die shrink, increased 32-bit performance
2. GeForce -> GeForce2: die shrink, performance optimizations, two bilinear textures per pixel.
3. GeForce3 -> GeForce4: second vertex shader unit, performance/visual quality optimizations, a few more pixel shader instructions.

Fine, you want to label those as the refreshes, so be it. Just keep in mind that the average there is much closer to one year, not six months, so if you stick to your guns there NVIDIA releases a new architecture every ~2 years and a refresh every year. You can't have it both ways... either the refreshes are clocked bumped parts, or they are one year apart, not 6 months.

My point is that somehow, based on NVIDIA's promises years ago I guess, people have convinced themselves that NVIDIA actually refreshes their core every six months. This is just not true.

 

[KimB]

nVidia's original strategy (pre-GeForce2 Ultra) was in releasing a new architecture in the fall, and a refresh part in the Spring.

The refresh part generally was a slight redesign of the fall-designed part, usually with performance optimizations.

Later on, it appears that nVidia aligned their new architecture cycle with the availability of die shrinks. This dropped nVidia back to an 18-month new architecture cycle, but nVidia still kept up with one "mild redesign" of their parts inbetween (Talking only about the high-end parts here, of course...), which lines up better with what was previously termed a refresh.

Basically, many people have been refusing to call the "clock bumped" parts refreshes as that's all they did. The previous refreshes from nVidia did more.

As a side note, while nVidia has fallen back to an 18-month new architecture cycle, ATI has stayed on a 12-month new architecture cycle.

In the coming year or two, will nVidia accelerate back to a 12-month new architecture cycle, regardless of a lack of availability of new process technology? Will ATI drop back to an 18-month new architecture cycle themselves now that they have cought up to nVidia? It could be interesting...but whatever happens, if ATI starts to consistently put out better parts (which I still consider unlikely, btw), I just hope they get their driver act together.

 

[KimB]

Well these..ummm...refresh parts were marketed as NEW cards including the Geforce 2 PRO

Yes, they were new cards.

 

[Bigus Dickus]

The following dates were the best I could come up with, and I'm not completely sure if they are launch dates or "on shelf" dates, but here goes...

NV2 (Riva128): Mid Fall 97

NV2-ZX (Riva128ZX): February 98
TNT: Mid Fall 98

~ 5 months to "refresh part," roughly one year to new architecture.

TNT2: March 99
NV10 (GF256): August 99

~ 6 months to "refresh part," rougly one year to new architecture.

GF256DDR: December 99
NV15 (GF2 GTS): April 00

~ 4 months to "refresh part," rougly 8 months to new architectures (debateable).

NV16 (GF2 Ultra): August 00
NV20 (GF3): March 01

~ 4 months to "refresh part," roughly 11 months to new architecture.

NV20-Ti (GF3-Ti): October 01
NV25 (GF4): February

~ 7 months to "refresh part," roughly 11 months to new architecture.

NV28 (??): September?
NV30 (??): January?

~ 7 months to "refresh part," roughly 11 months to new architecture.

As you can see, with the exception of the TNT to TNT2 cycle, all refresh parts were simply clock bumped parts (and the TNT2 did have a clock bumped refresh). The only release that didn't quite fit the consistent one year new core was the GF256 to GF2 GTS, which took ~8 months instead of 11/12. Also notice that many of the cores I am calling "new architecture" is what you are calling "refreshes." Your "new products" seem to only be new DX generations... NV10, NV20, NV30...

About that 18 month new product cycle... they've been averaging closer to two years ever since the Riva 128.

I could be off on some of these dates, so feel free to point out the errors.

 

[DemoCoder]

A refresh is a speed-bin usually. New architecture means silicon that is actually different, not "from scratch". GF1 and GF2 are functionally identical, but they are different silicon. Most of the new IMR renderers still use very similar pipelines to the older cards. Not many of them are radical departures like I'd expect if the engineers started from a clear slate and redesigned from the ground up.

 

[Bigus Dickus]

A refresh is a speed-bin usually. New architecture means silicon that is actually different, not "from scratch".

My thoughts exactly.

 

[RussSchultz]

I'm of the opinion that a tnt2 to GF is a 'new product', and GF2 to GF3 is a new product, but the rest are simply refreshes and/or variations on a theme.

Either die shrinks, process changes, or minor enhancements.

To me, as a consumer, I don't see much difference between the GF1DDR and the GF2GTS/ULTRA (excepting speed).

Same thing between the GF3 and GF3Ti500. The GF4 is a little trickier, but its enhanced video stuff and memory architecture don't affect my life at all (beyond it being faster).

Part of the problem is each of their parts in the Geforce line (particularly from GF3 on) has been so evolutionary that its tough to say 'between here and here' is a new product.

 

[Sage]

ok ok

ok i dont know if anyone has posted this yet cause im too lazy to read thru 8 pages of mostly crap. no offence to anyone intended, i just tire of the same thing being said over and over.

Um, yes, I'm aware they are "beta" drivers. However, they are officially released betas, not some "leaked" driver. They are also for an EXTREMELY mature architecture, the GeForce3/4.

ok, the 40.xx is NOT for GeForce3/4!!! the whole point of releasing the drivers was to get us to where we can work with the CineFX architecture without actually having the hardware out. The 40.xx drivers are for the NV30, but of course they are backwards compatible like all nVida drivers. The ONLY reason they have been released now is because programmers have access to ATi DX9 class hardware and nVidia needed to ensure CineFX architecture wasnt forgotten.