At the GPU Technology Conference, Nvidia revealed its strategy of becoming a complete processing provider, releasing details of introducing CUDA and OpenCL support in its low-power parts. This is the first time the company officially released not just the codenames but key features of the new parts as well.
The 2013-2015 Nvidia Tegra roadmap: Meet Logan and Parker
As you can see on slide above, Nvidia is now stating that the company will release Logan SoC in 2014. The actual introduction is scheduled for the fourth quarter of this year, with mass production planned for 2014.
The key feature of Logan (James Howlett, e.g. Wolverine), besides the 20nm manufacturing process will be the inclusion of Kepler GPU. The company is pushing for full OpenGL 4.3 compliancy – albeit without any details of the actual performance of the parts. Given that we?ll probably see a single 192-core SMX cluster, we?re probably going to have a battle on our hands between AMD?s Beema, successor to Temash (with Jaguar x86 cores), Qualcomm Snapdragon 800 and Intel’s upcoming Bay Trail platform. Bear in mind this is high performance products, and getting them into the smartphones will be a tall order.
Nvidia Kayla – and no, as an american publication we won’t say what Kayla means in Croatian. This development does come from Europe, though – as it was originally backed by the EC grant
This ties nicely into the 2nd Generation ‘CUDA on ARM’. After the European Commission-backed "cARMa" project, Nvidia introduced Kayla, a reference platform that combines the Tegra 3 quinta-core processor with a yet unannounced low-power Tesla GPU based on the Kepler architecture (we believe this is either GK106 or GK107-based part). Kayla brings the whole philosophy together and gets things ready for the arrival of unifying silicon in 2014.
We were also introduced to a new name, Parker (Spiderman). This 2015 part is the culmination of almost 10 year development cycle for Nvidia, delivering the first processing part codenamed Project Denver. The chip should carry the internal number T60 (Tegra 6?), bringing 64-bit ARM instruction set based on a custom core, developed by Stexar, company which Nvidia acquired at the time when the company wanted to create a non-licensed x86 core which would run in a similar mode as now-dead-and-gone Transmeta, love child of Linus Torvalds.
Key element of Parker will be the GeForce GPU based on the Maxwell core, bringing things into sequence. Now we know that it will take approx. 12-18 months for Nvidia to adopt the new GPU architecture: Kepler debuted in May 2012 on desktop and notebook market, first quarter of 2014 will see Kepler arriving in a Tegra package. Maxwell GPU is set to debut in the first half of second quarter 2014, with ‘Tegra 6’ arriving in 2015. Nvidia also disclosed that Parker SoC will use FinFET transistor technology, leaving ample time for TSMC to catch up. Alternatively, we might see Parker SoC coming out of Samsung?s fab in Austin, TX.
But the most important part is not announced on this roadmap, which is Project Boulder – a server SoC packing Denver cores and Tesla-branded Maxwell GPU. This part needs to serve as the core feeder of big Tesla-branded GPUs based on Maxwell and Volta GPU architectures.
Nvidia’s short term Tegra roadmap shows Tegra 4i arriving only in 1st calendar quarter of 2014.
In terms of scheduling, here?s where the situation gets really interesting. In a call with the analysts, we learned that Tegra 4i will take up to a year to make a debut in mobile phones, as confirmed by the following slide:
As you can see, product formerly codenamed ‘Project Grey’ e.g. Tegra 4i won?t be on the market before the first quarter of next year, which prompted Jen-Hsun Huang to sound off that they?ve missed on a lot of design wins as ‘Tegra 4 was late’.
Where is the Tegra train heading? With 90% Year-on-Year growth, signs are good. But without design wins and losing the Google Nexus 7 deal – time will tell can SHIELD alone be enough to carry the company until Logan comes out.
The comforting part of the deal is that Tegra 4 raw performance outstrips the performance of Intel?s next gen part, but the competitor isn?t Intel, with practically zero market share in mobile ? their competitor is Qualcomm, and they?re not showing any sign of letting go their market share. However, Logan and Parker, as well as Project Boulder ? are the reasons why Qualcomm is panicking in their public statements, comparing themselves vs. Nvidia ? which is a completely wrong PR strategy.
But there is a pretty good reason why Intel, Apple and Qualcomm are showing nervousness, if not even fear when we talk about Nvidia, or AMD. If Nvidia executes its strategy until the end, market might shift in mysterious ways. AMD and Nvidia both have a chance to turn the industry upside down, and if you think that we?re completely off, just look at market shares of Apple, HTC and Nokia.
The big ‘if’ remains, though – can Nvidia execute without the executives coming out with all the shortsighted and arrogant (very offensive to Japanese culture) comments which caused them to lose the Nintendo 3DS, Sony PlayStation and Valve deals in the first place.