Nvidia’s big news at this year’s SIGGRAPH in Anaheim, California is that the GPU you’ll find inside Project Logan (its next-gen Tegra) and perhaps in other SoCs, given a recent interest in licensing, is up and running, based on the company’s Kepler architecture.
Of course, this isn’t a completely unexpected revelation—Jen-Hsun spelled out the Tegra roadmap a few months ago at GTC, making it clear that Logan’s graphics component would be Kepler-based. Consistent with his announcement, company representatives maintain that Logan will ship in early 2014.
Mobile Kepler consists of one SMX
We do get some additional specifics, though. I asked Nvidia’s Matt Wuebbling about the composition of what it’s calling Mobile Kepler, and he replied that it’ll comprise a single SMX. That means we’ll be looking at 192 CUDA cores, 16 texture units, and, presumably 64 KB of L1.
According to Wuebbling, Mobile Kepler will use one-third the power of Imagination Technology's PowerVR SGX554MP4, at the heart of Apple’s A6X SoC, in an unspecified rendering workload. Nvidia’s using a 2 W figure to describe Mobile Kepler, comparing it to GeForce GTX Titan’s 250 W maximum TDP. Those are interesting claims, since Nvidia earlier specified that it was able to achieve better power efficiency from separate vertex and pixel shaders in Tegra 4 than a unified shader design allowed (for comparison, Tegra 4 sports 24 vertex and 48 pixel shaders).
Nvidia's team on the ground at SIGGRAPH gave us a demonstration of Mobile Kepler dialed back to the performance of Apple's fourth-gen iPad, illustrating the difference in power consumption through measuring the graphics rail of each system. You can see this in action below:
On the flip side, adopting the Mobile Kepler architecture enables support for APIs currently not available on Tegra 4's NV40-class architecture, such as OpenGL ES 3.0, OpenGL 4.4, DirectX 11, OpenCL, and CUDA 5.0. When Nvidia introduced Tegra 4, it shrugged off the lack of OpenGL ES 3.0 support by saying there wouldn’t be much content any time soon due to developers’ propensity for targeting the lowest common denominator of mobile devices. It’ll be interesting to see if that story changes in less than one year.
Some of the compute-oriented workloads enabled by CUDA/OpenCL
Should you hold off on Tegra 4-based devices like HP’s 21” Slate All-In-One due to today’s demonstration? Nvidia says no, naturally.