The official launch of the third generation of Cooper Lake comes as Intel unveils a roadmap for the future, including Ice Lake, which will pay Abu in the second half of the year, and Sapphire Rapids next year. Cooper Lake still uses the 14nm process and Skylake architecture, with a maximum of 28 cores, limited to four and eight markets, and the highlight is the addition of support for the BF16 instruction set to optimize deep learning acceleration.
Ice Lake is a single-way, two-way market, the first 10nm process, and the new Sunny Cove CPU architecture, has been used in the light field, but the specifications on the Toqiang product line is still a mystery, the core number is not expected to be too large, the frequency will not be too high, but there is a high probability of supporting PCIe 4.0.
Sapphire Rapids will once again unify the single-way to eight-way market, supporting the next generation of deep learning acceleration instruction AMX (Advanced Matrix Extension), which will significantly improve AI training and inference performance, with specific specifications to be released later this month.
No further details were disclosed, but a roadmap from Huawei showed that Sapphire Rapids would support DDR5 memory, increase from six channels to eight channels, and PCIe 5.0.
And next year, Intel Xe’s new high-performance computing discrete graphics card, Ponte Vecchio, is also expected to be unveiled, partnering Sapphire Rapids at the Aurora supercomputer at a national laboratory in Aragon, USA.
Intel also revealed that sapphire Rapids’ first silicon wafers have been successfully lit up and that many members of the development team are now working from home, which is a new experience for Intel.