Intel and AMD Path to Zettaflop Supercomputers

In 2021, Intel talked reaching Zettascale in three phases. AMD CEO talked about getting to a Zettaflop in 2030-2035. AMD CEO indicated that an earlier Zettaflop supercomputer would need about 500 Megawatts of power. Exascale systems today consume 21MW of power.

AMD and Intel have managed to roughly double the performance of their CPUs and GPUs every 2.4 years. HPE, Atos, and Lenovo have achieved similar gains roughly every 1.2 years at the system level, Su says power efficiency is lagging behind. Citing the performance and efficiency figures gleaned from the top supercomputers, AMD says gigaflops per watt is doubling roughly every 2.2 years, about half the pace the systems are growing.

AMD aims to resolve this efficiency issue by innovating and utilizing creative packaging technologies. As per AMD, a 3D stacked approach is around 50x more efficient than an off-package copper solution.

AMD was more specific about using 3D stacked approaches and chiplets to achieve the performance gains. AMD CEO Su says the package is the new motherboard. Chiplets will allow chipmakers to address three of the low hanging fruits when it comes to compute efficiency: compute energy, communications energy, and memory energy. Modular chiplet or tile architectures have numerous advantages. Chipmakers using chiplets can use optimal process tech for each component.

Assuming this trend continues unchanged, AMD estimates that we’ll achieve a zettaflop-class supercomputer in about 10-years give or take

Intel’s general path to a Zettaflop supercomputer that is 1000 times more powerful than the best supercomputer today.
1. Optimizing Exascale with Next-Gen Xeon and Next-Gen GPU in 2022/2023;
2. In 2024/2025 with the integration of Xeon plus Xe called Falcon as well as Silicon Photonics or ‘LightBringer’;
3. Zettascale around 2027.

Intel has gone on the record saying that Aurora, the upcoming supercomputer for Argonne, will be in excess of two ExaFLOPs of 64-bit double-precision compute.

In February, 2023, Intel recently announced that the test system (Borealis) for the exascale Aurora deployment at Argonne National Laboratory in Illinois is finally live.

Aurora will feature around 10,000 server blades, each featuring 2x 4th Generation Intel Xeon Scalable Processors (which we know as Sapphire Rapids) and 6x Intel Data Center GPU Max Series (which we know as Ponte Vecchio) chips. The Intel Borealis test system (also based out of Argonne) will feature just 128 server blades although in an identical configuration and scalable setting as its larger variant. Aurora will be the size of 2 basketball courts and 600 tons and a rated peak performance of 2 EXAFLOPs, this will likely be one of the first exascale supercomputers in the US and one of the fastest in the entire world.

Intel talks about getting an architecture jump of 16x, power and thermals are 2x, data movement is 3x, and process is 5x. That is about 500x, on top of the two ExaFLOP Aurora system, to get to a ZettaFLOP.

The 16X architecture jump, Intel says the foundational element is the IPC per watt improvement. Intel thinks we know how to do the 16x performance improvement pretty easily, or relatively. The power efficiency is the challenge there in terms of both the architecture and microarchitectural opportunities that are ahead of us.