If you look at the dates in our outing of Nvidia’s the chart in our 28nm article, you can see first samples go to those in good standing with the company in early October, with mass production samples coming in late December. So far so good, but there is more to it than that. Short story part 1: We were wrong on the mass production dates. Short story part 2: There is more to this story than we heard the last time around.
For the slightly longer explanation, we screwed up on the mass production dates for the 64 and 128-bit parts, it is the end of January, not the end of December. This isn’t a slip, it is SemiAccurate getting some numbers wrong. That said, Nvidia won’t have any new product in laptops until mid-February best case, late February is more likely. It takes time to make machines and ship them. AMD/ATI is still set for December mass production shipments.
The schedule in full
More troubling is the dates that the chart above shows. Notice how everything slipped about 6 weeks? Not a big deal, there are problems in chips all the time, and schedules do slip. Even worse, the “in” thing for foundries to do is delay production, but thankfully, that isn’t the case here. In this case, everything just slipped by a magic number.
That magic number, 6 weeks, is the time it takes for a chip to have another spin, industry slang for a minor revision. Almost every new architecture has several spins, AMD’s Bulldozer/FX is on B2, indicating one major (Ax -> Bx) and three minor (B0 -> B2) spins. This is par for the course with a new design.
The big problem for Nvidia is that the first batch of 28nm parts isn’t a new design, it is a shrink of the GF11x parts currently on sale. There should be no logic changes to it this late in the game unless something is really screwed up, something we really doubt happened. That leaves the dreaded problem of indeterminate problems with a new process.
Nvidia has been claiming that they fixed the process related problems that plagued 40nm production on their 28nm designs. While we laughed when we read that rather curious spin, pun intended, on how things were going, insiders contacted by SemiAccurate said things aren’t much better this time around. 40nm may have been horrific for Nvidia, 28nm seems to be simply not good. I suppose this could be interpreted as progress, but it explains why we laughed when 28nm was spun as something akin to a Rainbow Brite cartoon by Dear Leader.
Back to the first 28nm parts and the spin, a dumb shrink like this, or even a mild reorganization of the older 40nm parts before the shrink should not have required a spin. If you look at the added ES2 sampling, that means that ES1 will be in very short supply, too short for OEM liking. That should tell you how tight Nvidia is cutting things with this spin to make those dates.
That too is indicative of a late in the game problem, something that is unlikely to be a logic bug considering this is the third generation of essentially the same chips. All signs point to 28nm problems, and considering Nvidia released a lot of early 28nm capacity a few weeks ago, it might not be the end of the line for the headaches. Is this 40nm/Fermi again? No. Is it smooth? No. Are 28nm parts later than hoped for? Yes. Are the process problems solved like Jen-Hsun said? Nope.S|A
Latest posts by Charlie Demerjian (see all)
- You can now buy bare Snapdragon SoCs with 410E and 600E - Sep 28, 2016
- Spin Transfer Technologies talks about their ST-MRAMs - Sep 27, 2016
- ARM adds CMN-600 interconnect and DMC-620 memory controllers - Sep 27, 2016
- Globalfoundries 7nm process isn’t even close to the name - Sep 26, 2016
- ARM upgrades realtime offerings to v8-R and adds Cortex-R52 - Sep 21, 2016