Going over the GTC presentation about Tegra, it is clear that the roadmap has massively slipped just as SemiAccurate stated. The only thing unusual about Tegra in this presentation is how long it took the company to admit what it was really doing.
Nvidia announced Project Denver during CES 2011 but officially gave no details about architecture or timing. If you were an analyst in good favor with the company they would tell you that Denver was going to be on sale in Q3 or Q4 of 2012, Q1/2013 at the latest if something went badly wrong. SemiAccurate heard the same story directly from three analysts who claimed it was told to them individually after the press conference. Other than being an ARM ISA CPU with an attached GPU, nothing relevant was revealed about the architecture.
Instead, SemiAccurate provided great detail about Project Denver down to the piepline and internal organization of the part. The core itself is called T50, Tegra 5 at the time, and Denver is the core with an uncore and Maxwell GPU. Essentially it is a CPU/SoC for the desktop or server and we detailed it here.
Given the state of the project at the time of writing, almost two years ago, we stated that the dates given to analysts were absolutely farcical. More directly we said that Nvidia knew at the time that there was no way those dates were possible either and we stand by that statement.
One of the main reasons for this was that Denver used Nvidia’s Maxwell GPU cores for graphics, quite necessary once you realize that Maxwell is the first to have coherent memory capabilities. Maxwell isn’t due until 2014, so how can a SoC that uses those shaders come out 18 months earlier? The only question SemiAccurate has about the timing is how anyone with a shred of technical knowledge could have believed the story in the first place.
The state of Tegra as of GTC 2013
Going back to the GTC presentation, we have Tegra 4 for 2013, Logan for 2014, and Parker for 2015. Parker is the first Project Denver based part, and there is one more generation between the current Tegra 4 and Denver. If the merry marketeers on the San Tomas Freeway don’t change the numbering scheme, that would mean that Parker/Project Denver is Tegra 6 just like we said.
Luckily according to Nvidia their Tegra roadmap did not slip at all, any analyst who asked about it after SemiAcurate exclusively revealed the year+ slippage 18 months ago was told we were making things up. Now there is the slide above confirming it, and for some reason none of the press is calling it a slip. Logic says that if there was no slip, the company lied to analysts at CES 2011. If there was a slip, they lied to analysts in October of 2011, pick which one you feel is, err, true.S|A
Latest posts by Charlie Demerjian (see all)
- More on Intel’s 10nm process problems - Sep 17, 2018
- Intel puts out another 14nm 2020 server platform - Sep 11, 2018
- Why Can’t Intel Supply Enough 14nm Xeons? - Sep 10, 2018
- Intel can’t supply 14nm Xeons, HPE directly recommends AMD Epyc - Sep 7, 2018
- AMD reintroduces the Athlon name with two CPUs - Sep 6, 2018