demalion said:
1) There is no reason I've heard so far to believe the nv30 effort has delayed the nv35, and I'd tend to think nVidia would have had to be rather foolish to change their plans to have that result when they hit the nv30 snags. I tend to believe that the nv35 delay would primarily be determined by the marketplace, given the original intended launch schedule...as we've guessed during the nv30 delay period, focus on preparing the nv35 was likely a priority.
In this light, killing the 5800 Ultra seems very sane if they've recently reached the conclusion that they can bring it to market soon enough. I think it lends validity to the May(?)/June launch rumors with the nv35 (as much before the launch of the R400 as possible), and indicates that the issues (I think) Mufu has hinted at are on-track for resolution.
Aren't you just about sick and tired of "rumors" when it comes to nVidia and its products? That's about all it's been for the past 8-9 months out of nVidia--rumors and speculation galore--and at the end the stark and naked truth:
nVidia has no chip with which to compete with the R300 from ATI.
There were plenty of good, solid reasons to cancel the GF FX Ultra which have absolutely nothing whatever to do with "nv35" (*chuckle* That's so funny--after hearing "nv30,nv30,nv30,nv30".......ad infinitum for the past six months--and how it would "wipe the floor" with the puny R300 from ATI.) How about:
(1) The chip has to be over volted and over clocked to even superficially compete with R300--not R350, not R400--R300, shipped last September.
(2) The dustbuster makes too much noise
(3) The dustbuster is too big
(4) The dustbuster is too expensive
(5) 12-layer PCBs
(6) 75W of heat displacement
(7) Too many heatsinks
(8 Warranty coverage too expensive
Those are just in-your-face, undeniable observations each of which by itself might be reason enough to cancel the product. Taken in total they are overwhelming.
Let's
speculate and add a few more:
(9) OEMs were unhappy at what the card would cost to make (or "What if you built a reference design that nobody wanted?")
(10) What if nv30 yields were so low they couldn't get enough chips to sell that would even overclock and overvolt to 500MHz without destroying themselves?
(11) DDRII ram is too expensive, and as the sole-supplier for the 1GHz spec stuff, SamSung thought to press its advantage and charge a premium?
That's enough. As you can see there are plenty of reasons to cancel this product, none of which are remotely connected to the mythical "nv35." (Ever heard the story about the boy who cried 'wolf'?)
2) 500/800 doesn't sound very sane to me...but 400/"1000" does.
(a) I'd expect heat issues with "DDR II" RAM to be less severe than those for the GPU, and yields at 400 MHz sound profitable.
(b) Bandwidth is the GF FX's primary problem.
(c) Not having read Kyle's comment yet, it seems like a plausible minor miscommunication.
400/800 is the real deal. Maybe nVidia will up the spec for the non-Ultras. But at a selling price of $300, don't count on it. nVidia already knows the non-Ultras won't compete with R300, and certainly not with R350. Why throw good money after bad? Let it compete with the $300 ATI cards....which *chuckle* will probably be the R300 anyway--just as soon as ATI ships the R350.....
BTW, I disagree about your bandwidth statement. Even with virtually *the same* raw physical bandwidth as a 9700P, the Ultra was much slower per clock, and that had nothing to do with bandwidth at all. I will say however that as the latencies in DDRII seem to be a fair amount higher than those in DDR I, and as nVidia's nv30 bus didn't seem optimal to me in the sense of even being very good at 128-bits wide, you might get by with a broad interpretation of "bandwidth" here--certainly not anything that putting slightly faster DDR II ram on the non-Ultra would solve. I think the GF FX non-Ultra might make good competition for the 9500Pro, maybe.
I think you'll have a wait on the nv35 though, I really do, as I think it will take significantly more than a 256-bit bus to make it competitive with R350/R400, specifically R400--and that will be a much greater challenge than R300 was, I believe.