Associate
- Joined
- 3 Jan 2010
- Posts
- 1,379
...That'd take a lot of cooling to stop destroying your rig
Plus an army of mayans would need a heck of a case 
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Nvidia works on second generation Fermi
- Thread starter Dark Sun
- Start date
Plus an army of mayans would need a heck of a case 
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
Yet again on the whole 40nm yields are up things, the storys now talking about it are mentioning the 9k wafers being comparable to the 9k wafers being made at 65nm. Which again proves my point they are simply talking about the fab yields and NOT the process yields.
Basically TSMC are saying yay us, we're back up to 9k wafers a month like we had on 65nm. The problem is yields per wafer are still WAY down on 65nm, and 9k wafers on 65nm isn't a hugely impressive number all told. You're talking about just over 100k wafers a year, which compared to other fabs doing 300nm wafers, is pretty damn weak. TSMC's main business is smaller cheaper chips on older cheaper processes.
So the whole "we've fixed 40nm, and our 28nm is fantastic" is really, "we're back up to production we had before our huge screw up in October, on our year late and crappy yield process. Though we've promised our last 4 processes on time and missed each and every one, our new process is on time and certainly not maintaining all the mistakes that made 40nm process suck on yields, as it still does over a year after we started it."
The major problem is there are fundamental design flaws with their 40nm, they took the "cheap and easy" route which played a huge issue in how useless it was and how long it took to get up to releasable yields. It also played a massive part in how difficult it was to make Fermi "work" on the process due to design limits when putting a design onto silicon. A large number of these problems are still present in the 28nm design, even if its got HKMG added, the lack of TSMC to get anything work right doesn't actually guarentee their HKMG is going to actually help, design wise it doesn't help.
The advantage for Nvidia will be huge drop in size and more space to better organise the transistors for less leakage, but on the same track, they can't double transistors again as they'd end up with the same behemoth design that was hard to produce. If Nvidia go way less than going for a 1k shader design, while AMD can double their power and have no space issues, we'll have to see where they come in. If Nvidia do go stupid and go yet again for another high power massive core design, they'll likely run into the same issues again and not be competitive on price.
Basically TSMC are saying yay us, we're back up to 9k wafers a month like we had on 65nm. The problem is yields per wafer are still WAY down on 65nm, and 9k wafers on 65nm isn't a hugely impressive number all told. You're talking about just over 100k wafers a year, which compared to other fabs doing 300nm wafers, is pretty damn weak. TSMC's main business is smaller cheaper chips on older cheaper processes.
So the whole "we've fixed 40nm, and our 28nm is fantastic" is really, "we're back up to production we had before our huge screw up in October, on our year late and crappy yield process. Though we've promised our last 4 processes on time and missed each and every one, our new process is on time and certainly not maintaining all the mistakes that made 40nm process suck on yields, as it still does over a year after we started it."
The major problem is there are fundamental design flaws with their 40nm, they took the "cheap and easy" route which played a huge issue in how useless it was and how long it took to get up to releasable yields. It also played a massive part in how difficult it was to make Fermi "work" on the process due to design limits when putting a design onto silicon. A large number of these problems are still present in the 28nm design, even if its got HKMG added, the lack of TSMC to get anything work right doesn't actually guarentee their HKMG is going to actually help, design wise it doesn't help.
The advantage for Nvidia will be huge drop in size and more space to better organise the transistors for less leakage, but on the same track, they can't double transistors again as they'd end up with the same behemoth design that was hard to produce. If Nvidia go way less than going for a 1k shader design, while AMD can double their power and have no space issues, we'll have to see where they come in. If Nvidia do go stupid and go yet again for another high power massive core design, they'll likely run into the same issues again and not be competitive on price.