NVIDIA's 28nm "Wayne" Tegra to Come in Two Versions?
6/30/2011 by: Theo Valich
Even though Kal-El hasn't even launched yet, we have received some preliminary information on NVIDIA Wayne. You're free to take this information with a grain of salt, we're just the messengers here.
NVIDIA's Public Tegra Roadmap Shows 28nm Wayne Arriving in 2012, ahead of Logan and Stark
T40 is 4th Generation Tegra, said the people, who asked not to be identified because details are confidential. The prototype Wayne silicon will be released to manufacturers and software developers in December 2011, in order to get ready for the Windows 8 release in October 2012. With Wayne, NVIDIA plans to target smartphone, tablet, netbook and notebook market. In order to do just that, the company decided on Batman/Robin strategy, i.e. developing two silicons.
The initial silicon is a quad-core design based on a yet undisclosed ARM architecture. There are two choices: NVIDIA licensed Cortex-A15 and own in-house 64-bit architecture. Given that Project Denver is scheduled for tape-out by the end of 4th quarter - we doubt it. It is safe to assume that Cortex-A15 is being used.
Initial silicon also features "at least 24-core" CUDA-enabled GPU part which resembles contemporary GPU architectures, bringing a lot of computing performance. The people said that the "10x" performance increase (from Tegra 2) includes a lot of GPU computing power, which is being used for numerous segments. One of such uses was motion-picture sharpening a'la CSI i.e. technology similar to what MotionDSP developed. Lightfield focusing technology was also mentioned to us as "the next big thing" in mobile - and processing side will handle Lightfield on-the-fly using GPU cores.
This first silicon is targeting the 1.5GHz clock for the ARM cores and packs triple-digit GFLOPS number. Given that the competition already announced 2GHz and 2.5GHz parts based on Cortex-A15 CPU core, we asked wouldn't 1.5GHz clock prove uncompetitive. The answer was that 1.5GHz is targeted for core-per-core comparison with the A9-equipped "Kal-El" T30 and that final clock speeds depend on the power consumption as the "Robin silicon" needs to keep the same power envelope as Kal-El.
Second Wayne silicon can be described as "Holy Smokes Batman". Unlike 4/24+ "Robin", NVIDIA's is planning an 8CPU/32-64GPU part for tablets, netbook and ultra-low-power notebooks. Think up to 13.3" in screen size. We were confused by the "up to 64" claim as NVIDIA's contemporary Fermi architecture is designed in 32 and 48-core blocks. However, in order to reach the power targets, engineers are working to be as flexible as possible. Whatever the final design ends up being it is a DX11+ compliant part, with full support for OpenGL 4.x and OpenCL 1.x. Also, an interesting note is that PhysX was explicitly mentioned in the "we have full support for PhysX now" sense. Draw your own conclusions on what that means.
The 8-core ARM design is being designed to go head to head against not just ARM competitors, but against the X86 processors as well. Remember, for Windows 8 Intel should cover top-to-bottom market with the next-gen Atom architecture (22nm Silvermont) on the low end and 22nm Ivy Bridge. At the same time, AMD plans 28nm 4-core Krishna and 2-core Wichita, both based on Enhanced Bobcat CPU core and a 32nm Trinity APU with dual and quad-core designs based on Enhanced Bulldozer core.
In a nutshell, Wayne is actually a split personality - quad-core Robin and octa-core Batman. No doubt, competitors will probably make a reference to Harvey Two-Face Dent.
NVIDIA, Wayne, Tegra 4, T40, NVIDIA T40, T45, Kal-El, Clark Kent, Superman, NVIDIA Wayne, Tegra 3, Tegra 3D, Batman, Robin, The Dark Night, Logan, Wolverine, Stark, Iron Man, 28nm, TSMC
© 2009 - 2011 Bright Side Of News*, All rights reserved.