Intel sets out its plans for 2019: Ice Lake, Lakefield, and Project Athena

Enlarge / Complete Lakefield motherboard. (credit: Intel)
Intel had a big presentation at CES last year, which was strange, because the company didn't really have many things to announce. The delays to its 10nm manufacturing process meant that, instead of exciting new chips, the best we could really hope for was rehashed versions of its current chips, which duly arrived. This year's presentation was very different. The company's 10nm process is finally due to achieve volume production this year, and late last year the company told us that 10nm was bringing with it a new architecture named Sunny Cove, a new, much faster GPU, and new manufacturing techniques with 3D die stacking. All this meant that the presentation could actually introduce a range of new products that will ship in 2019.
10nm processors for the masses: Ice Lake-U
The most important product announced, as it will likely be the highest-volume part, is Ice Lake-U. This is a 10nm mobile processor with a Sunny Cove CPU and a Gen11 GPU. Since its 2015 introduction, Intel has produced a number of variations of its Skylake architecture. Performance improvements have come from increasing clock speeds and core counts, with the core design essentially unaltered. Sunny Cove, by contrast, is a meaningful update to and improvement of that architecture and will represent Intel's first improvement in instructions-per-cycle in four years. This means that it should offer across-the-board performance improvements, regardless of workload. The Ice Lake-U parts will retain the U-series 15W power rating and will offer a maximum of four cores and eight threads.
The Gen11 GPU is similarly slated to give a big performance boost. The standard configuration (named GT2 in Intel's parlance) will include 64 execution units as compared to the 24 execution units that are standard in Skylake/Kaby Lake/Coffee Lake processors. This will more than double the floating point performance of the GPU from about 420 gigaFLOPS to 1 teraFLOPS. This increased performance increases the memory bandwidth requirements of the GPU, and the chip will accordingly sport two channels of LPDDR4X memory.
Read 5 remaining paragraphs | Comments