In February, Intel announced a special fusion processor "falcon shores" (Falcon Shores), officially called "xpu", which can be roughly understood as the integration of strong x86 CPU and Xe GPU. It is designed based on tiles and has very high scalability and flexibility. It can better meet the needs of HPC and AI applications.
During the ISC supercomputing conference, Intel announced more details.
In particular, falcon shores can be configured with different block modules as required, especially x86 CPU cores and Xe GPU cores. The number and proportion are very flexible, so it can be used as what it is
At the same time, it will adopt EMI level manufacturing processes (Intel 18a/14a, etc.) and the following generation of advanced packaging technology. It has extremely high shared memory bandwidth, industry-leading IOS input and output, and can greatly simplify the programming model.
According to the figures given by Intel, compared with today's level, Falcon shores' energy consumption ratio has increased by more than 5 times, x86 computing density has increased by more than 5 times, and memory capacity and density have increased by more than 5 times**
From the road map, Falcon shores XPU products will uniformly replace sapphire rapids HBM, Ponte Vecchio, Rialto Bridge and other accelerators in the future, while the traditional Xeon CPU processors will continue to develop. The next two generations of code names are emerald rapids and granite rapids.
