GTC 2013 | Nvidia's Tegra Logan will be CUDA-capable
Nvidia has always been outspoken about its long-term ambition to scale ARM-based Tegra processors on everything from smartphones to super-computers. At the company’s annual GPU Technology Conference, CEO Jen-Hsun Huang revealed how Tegra’s next iteration (codenamed Logan) will take Nvidia one step closer to achieving its goal.
Many consumer technology reviewers and enthusiasts expected Tegra 4 to support CUDA, Nvidia’s parallel computing framework, which unfortunately was not the case. However Mr Huang officially announced yesterday that Logan will support CUDA 5 and Open GL 4.3. Hence Nvidia’s next generation mobile processor will finally acquire unified shader architecture in the form of a Kepler-based GPU. Logan is scheduled to be in production early next year.
After Logan comes Parker which will bring 64-bit capabilities needed for servers and super-computers. Parker will integrate both a Denver CPU and a Maxwell GPU and incorporate 3D FinFET transistors (or what Intel likes to call ‘Tri-Gate’). By 2015 Nvidia expects a 100x increase in Tegra performance over a period of just 5 years.
Mr. Huang also provided details on the next two Kepler GPU successors. First off is Maxwell with Unified Virtual Memory between CPU and GPU whereby, “all memory is visible to all the processors.” After Maxwell comes Volta focusing on energy efficiency and featuring ‘stacked memory’. In theory this should help Nvidia achieve one Terabyte per second of memory bandwidth. That is over 3x higher than the company’s current flagship “Titan”, capable of 282 GB/s.