Semiconductor & Computer Engineering
| Line 18: | Line 18: | ||
== Overview == | == Overview == | ||
| − | Frontier is expected to be one of the fastest - if not the fastest - supercomputer when delivered in 2021. Comprising over 100 {{cray|Shasta}} cabinets, each node will feature a custom [[AMD]] {{amd|EPYC}} (likely a {{amd|Milan|l=core}} derivative) along with four custom Radeon GPUs. [[interconnect architecture|Interconnects]] will be comprised of AMD {{amd|infinity fabric}} for the node and {{cray|Slingshot}} for the system. All in all, Frontier is targeting | + | Frontier is expected to be one of the fastest - if not the fastest - supercomputer when delivered in 2021. Comprising over 100 {{cray|Shasta}} cabinets, each node will feature a custom [[AMD]] {{amd|EPYC}} (likely a {{amd|Milan|l=core}} derivative) along with four custom Radeon GPUs. [[interconnect architecture|Interconnects]] will be comprised of AMD {{amd|infinity fabric}} for the node and {{cray|Slingshot}} for the system. All in all, Frontier is targeting 2-4x the performance and capacity of {{\\|Summit}}'s I/O subsystem and 5-10x the real-application performance of {{\\|Summit}} with an expected peak performance of over 1.5 exaFLOPS. |
[[category:supercomputers]] | [[category:supercomputers]] | ||
Revision as of 10:21, 7 May 2019
| Edit Values | |
| Frontier | |
 | |
| General Info | |
| Introduction | 2021 |
| Peak FLOPS | 1.5 exaFLOPS |
| Succession | |
Frontier (OLCF-5) is Summit's successor, a planned exascale supercomputer that will be operated by the DoE Oak Ridge National Laboratory. Frontier is expected to go into operation in the 2021-2022 timeframe.
History
Frontier is a planned exascale supercomputer with a theoretical peak performance of over 1,500 petaFLOPS (1.5 EF). The design goal of Frontier is to achieve around 50-100x performance improvement in real science applications or alternatively around 5-10x application performance improvement over Summit.
Overview
Frontier is expected to be one of the fastest - if not the fastest - supercomputer when delivered in 2021. Comprising over 100 Shasta cabinets, each node will feature a custom AMD EPYC (likely a Milan derivative) along with four custom Radeon GPUs. Interconnects will be comprised of AMD infinity fabric for the node and Slingshot for the system. All in all, Frontier is targeting 2-4x the performance and capacity of Summit's I/O subsystem and 5-10x the real-application performance of Summit with an expected peak performance of over 1.5 exaFLOPS.
| introductory date | 2021 + |
| main image |  + |
| name | Frontier + |
| peak flops (double-precision) | 1.5e+18 FLOPS (1.5e+15 KFLOPS, 1,500,000,000,000 MFLOPS, 1,500,000,000 GFLOPS, 1,500,000 TFLOPS, 1,500 PFLOPS, 1.5 EFLOPS, 0.0015 ZFLOPS) + |
