Kepler Taped Out

[GeorgiD]

No, they were like just over half a year apart, 480 and 580. But, the guy most probably meant that it had taken 15 months them to fix it...

[fatty2nd]

Originally Posted by GeorgiD

No, they were like just over half a year apart, 480 and 580. But, the guy most probably meant that it had taken 15 months them to fix it...

That's what I thought.

[265586888]

GF100-375-A3 = GeForce GTX 480
GF110-375-A1 = GeForce GTX 580 = GF100-375-B1...

By the way, the GT200-GF110 transition took way more than 15 months.

[distinctively]

Originally Posted by kac

He probably meant "unfixable" from a market perspective. Whenever you've got die 30% bigger than your competition yet you've got performance that's only -5 to 10% higher you've got a problem.

Unfixable was unfortunately part of the title of the article. Charlie explained his reasoning in that article, stating that 28nm chips would obsolete Fermi before it could be fixed. The incredibly long delay in 28nm production allowed a window of saving face for nVidia there.

Originally Posted by fatty2nd

Were they really 15 months apart?

The first Fermi taped out around mid July, 2009 IIRC. The GTX 580 launched in November 2010. The point I was trying to make was simply that it took 15 months to fix the chip in a way that it would work as intended. The GTX 480 did not work as intended.

[fatty2nd]

Originally Posted by distinctively

snip

The first Fermi taped out around mid July, 2009 IIRC. The GTX 580 launched in November 2010. The point I was trying to make was simply that it took 15 months to fix the chip in a way that it would work as intended. The GTX 480 did not work as intended.

I see, I thought you meant 15 months between the 480 and 580. I was starting to think I had a massive amount of time disappear from me.

[jaydeejohn]

Lucks been on nVidias side, like was mentioned, the slow drawn out 40nm and the 28nm was on their side.
As stated by durr leader, they had no fallback plan, like the 3870 was, so to me they were extremely lucky
If 40nm had been ready, marketshare may also have been way different as well

I guess its better to be lucky than good heheh

[Regenweald]

Big die 'all or nothing' strategy not exactly working out for them twice in a row ?
@ Wandrey, those people you mentioned in other forums.........are really stupid, is all I can think of. They treat the 500 series like a new generation when all it is, is a fix. That took what, over a year and a half to kind of get right?
But never mind, I need and industrial pump, 15 feet of tubing, 4 rads and 3 reservoirs to keep my rig cool, Nvidia HAS to be the fastest.....

think i went off topic somewhere in here.....I should get beck to my project

[Ailuros]

Was GF110 really a complete "fix" or rather a quick and dirty job with whatever was really possible withing the rather short timeframe between GF100 and 110? Albeit a bad parallel example wasn't RV670 = "R600 done better" and RV770 = "R600 done right"? Think about it, no IHV has a magic wand and time is money either way.

I'd love to be surprised but don't expect any major architectural changes in Kepler to be honest. More like "Fermi done right" with some efficiency changes left and right. With the resources NVIDIA poured into Fermi, it sounds mighty stupid to me to waste that kind of sum into something that'll live only that shortly.

[Poisoner]

Originally Posted by Ailuros

Was GF110 really a complete "fix" or rather a quick and dirty job with whatever was really possible withing the rather short timeframe between GF100 and 110? Albeit a bad parallel example wasn't RV670 = "R600 done better" and RV770 = "R600 done right"? Think about it, no IHV has a magic wand and time is money either way.

I'd love to be surprised but don't expect any major architectural changes in Kepler to be honest. More like "Fermi done right" with some efficiency changes left and right. With the resources NVIDIA poured into Fermi, it sounds mighty stupid to me to waste that kind of sum into something that'll live only that shortly.

Rv770 had over twice the stream processors of rv670. Rv770 was an evolutionary step, not a fix. Rv670 was the shrink and fix.

[distinctively]

Originally Posted by Poisoner

Rv770 had over twice the stream processors of rv670. Rv770 was an evolutionary step, not a fix. Rv670 was the shrink and fix.

+1 to the analogy. I believe the Rv670 also bumped up to DX10.1 from DX10.

Originally Posted by Ailuros

Was GF110 really a complete "fix" or rather a quick and dirty job with whatever was really possible withing the rather short timeframe between GF100 and 110? Albeit a bad parallel example wasn't RV670 = "R600 done better" and RV770 = "R600 done right"? Think about it, no IHV has a magic wand and time is money either way.

The nVidia internal code name for the GF110 was actually GF100b. The GF110 is really nothing more than a base layer respin of the GF100. Nothing changed except fixing the leaking areas.

Originally Posted by Ailuros

I'd love to be surprised but don't expect any major architectural changes in Kepler to be honest. More like "Fermi done right" with some efficiency changes left and right. With the resources NVIDIA poured into Fermi, it sounds mighty stupid to me to waste that kind of sum into something that'll live only that shortly.

I agree that it would be a waste of money if they just completely dumped the Fermi architecture. I don't think they can maintain that arch for long unless the professional side can continue support the gaming side. Without the professional sales, nVidia is in the red for the year AFAIK.