Just after winning BullSequana last month, AMD has become the CPU provider for another supercomputer, the Archer2, which is slated to launch in May of 2020 with a total of 11,696 Rome CPUs. Archer2 also features 16 AMD GPUs that are tied to just a few of the 64 core Rome CPUs that are inside the supercomputer. This new supercomputer is theorized to be one of the fastest in the world when it launches, though not quite the fastest in the world.
Archer2’s predecessor, Archer, is an Ivy Bridge based supercomputer with 9,840 12 core Xeon CPUs that support 307.5 TB of DDR3 memory. By comparison, the new Archer2 will have 1.57 PB of DDR4, which is 5 times larger and also much faster with a peak bandwidth of 410 GB per second, provided all the RAM slots are filled.
The Next Platform gives the following performance improvements Archer2 has over its predecessor:
- 7X for CP2K, a quantum chemistry and solid state physics package
- 5X for OpenSBLI, a Navier-Stokes solver for shock-boundary layer interactions (SBLI)
- 3X for CASTEP, a materials modeling code
- 9X for GROMACS, a molecular dynamics package aimed at biological chemistry
- 18X for HadGEM3, the Hadley Centre Global Environmental Model
Unsurprisingly, Archer2 is another supercomputer built by Cray (which also worked on the original Archer); Cray is providing many other AMD based supercomputers such as the Frontier supercomputer for the US Department of Energy. This time, Cray’s client is UK Research and Innovation, an organization sponsored by the British government. Though Archer2 is a much smaller project based on Rome (unlike Frontier which is based on a post-Milan architecture), the contract to build it still commands a value of $102 million.
Though Archer2 is more expensive than the original Archer, it provides much better value thanks to being about 11 times faster. Whereas in 2012 each TFLOP costed $27,450, in 2020 it will only cost $3,642 thanks to the use of Rome CPUs which AMD has positioned at a much lower price than Intel’s Xeons, even in spite of Rome’s better power efficiency and its often times outright greater performance.
Archer2 is just another in a string of victories for AMD in high performance computing, and for Intel it is yet another supercomputer lost. As we covered earlier this week, Intel is focusing on mobility these upcoming years when it comes to 10nm volume (which is already low as it is), so it’s highly unlikely Intel will ever be able to provide enough 10nm Xeons to prevent AMD from claiming more supercomputers. Intel is waiting until its 7nm node is ready to go before challenging AMD head on in this arena.