Daily change is hard to observe, but change is certain - and obvious - at annual conferences. SC23 is about to get underway in Denver, and fundamentally, SC23 will feel different: there will be a lot to talk about.
Amongst the most noticeable changes for many HPC observers and practitioners will be the availability of more Arm architecture servers - from Tier 1 OEMs and all major ODMs, the increasing Cloud options, and the major deployments of Arm that have been announced recently. Our booth will be highlighting some of these at the show, as will a significant number of other system integrator and partner booths!
In the Cloud, AWS’s hpc7g instances set a new bar for on-demand compute – the cloud’s first HPC-specific Arm instances – achieving performance and cost that wins outright across both compute and memory bound classic HPC simulations such as in molecular dynamics, engineering, weather or biosciences.
On premise, Arm Neoverse systems are lining up. NVIDIA’s Grace and Grace-Hopper systems are winning major installations. The UK’s largest Supercomputer – Isambard AI – at 200PF DP was announced this month, and SiPearl Rhea is set to power the compute partition of the EU HPC Exascale system, Jupiter at Juelich Supercomputing Centre. These join earlier announced Arm systems such as CSCS’s Alps, Los Alamos’s Venado. NCHC Taiwania and beyond.
As open source across every part of HPC and AI, and commercial software such as ANSYS Fluent and Dyna, or Siemens Star-CCM+ and Altair Radioss, demonstrate significant benefits on Arm – there is a reason for every attendee to come to our booth and learn more.
A robust wide software ecosystem combined with world-leading products is enabling a reach across every part of HPC and AI. HPC sites and users will be talking about impact at events like the “Arm in HPC: experiences and lessons learnt in operating Arm-based HPC systems” Birds of a Feather session for example.
We’re thrilled by what Arm’s partners and the HPC/AI community is achieving.
For Arm, what we think is the most major change is one that will accelerate the pace and reduce the cost and complexity of innovation in HPC and AI.
Earlier this year we announced two major advances to accelerate time to market for our partners: Arm Neoverse Compute Subsystems (CSS) and Arm Total Design.
Neoverse CSS delivers pre-validated RTL for the compute subsystem of modern SoC designs, combining Arm Neoverse CPUs, coherent mesh fabric, network on chip fabric, system memory management units, and interrupt controllers with integration of IO and memory controllers. CSS can be configured to target different market segments and has been proven to reduce the engineering effort of building silicon.
Our Total Design partners are expert in design services and can create a full product, such as developing or integrating third party accelerators, and can help partners to build truly differentiating products faster.
These two developments, along with the widening availability of 2.5D and 3D design and the composability of silicon via chiplets, mean that new products will be designed for lower cost from design through to production, in a shorter time, and with lower risk.
Please do stop by at Booth #1969 to learn more.
Arm at SC'23