2. Reference Systems

2.1. El Capitan

El Capitan is the CORAL2 flagship supercomputer at Lawrence Livermore National Laboratory (LLNL), commissioned in 2024. It is an HPE Cray system designed for large-scale simulation and modeling, featuring AMD EPYC CPUs and AMD MI300A APUs (GPU+CPU). El Capitan is deployed in the Secure Computing Facility (SCF) and is a primary platform for the Advanced Simulation and Computing (ASC) program.

OS

TOSS 4

Scheduler

Flux

Interconnect

HPE Slingshot 11

Peak Performance

2,792.9 PFLOPS (CPU+GPU)

Login Nodes

16

Debug Nodes

56

Batch Nodes

11,424

Total Nodes

11,520

Total CPU Cores

1,069,056

Total GPUs (APUs)

44,544

Total System Memory

5,701,632 GiB

2.1.1. Node Architecture

CPU

4th Gen AMD EPYC

CPU Cores/Node

96

GPU (APU)

4x AMD MI300A (CDNA 3) per node

GPU Memory/Node

512 GiB (shared across 4 APUs)

Node Memory

512 GiB

El Capitan Hardware Overview

2.2. CTS-2

Dane is a CTS-2 class supercomputer at Lawrence Livermore National Laboratory (LLNL), commissioned in 2023. It is based on Intel Sapphire Rapids processors and is designed for high-performance computing workloads supporting the ASC and M&IC programs.

OS

TOSS 4

Scheduler

Slurm

Interconnect

Cornelis Networks

Peak Performance

10.7 PFLOPS (CPU)

Peak Performance

10.723 PFLOPS (CPU+GPU)

Login Nodes

8

Batch Nodes

1,496

Total Nodes

1,544

Total CPU Cores

167,552

Total System Memory

382,976 GiB

2.2.1. Node Architecture

CPU Architecture

Intel Sapphire Rapids

CPUs/Node

2

CPU Cores/Node

112 (56 per socket)

Threads/Node

224

Memory/Node

256 GiB DDR5-4800

Memory Channels/Node

16 (8 per CPU)

Cache/Node

210 MB (105 MB per CPU)

Max Turbo Frequency

3.8 GHz

TDP/CPU

350 W

Dane Hardware Overview

2.3. H100 System

Matrix is a CTS-2 class GPU-accelerated supercomputer at Lawrence Livermore National Laboratory (LLNL), commissioned in 2025. It is based on Intel Sapphire Rapids CPUs and NVIDIA H100 GPUs, designed for advanced high-performance computing and AI/ML workloads.

Vendor

Dell

Location

LLNL CZ Zone

Year Commissioned

2025

Class

CTS-2

OS

TOSS 4

Scheduler

Slurm (GPU-scheduled)

Interconnect

InfiniBand

Peak Performance (CPU)

0.198 PFLOPS

Peak Performance (GPU)

3.800 PFLOPS

Peak Performance (Total)

4.000 PFLOPS

Login Nodes

2

Batch Nodes

26

Debug Nodes

2

Total Nodes

30

Total CPU Cores

1,792

Total GPUs

112 (NVIDIA H100)

Total System Memory

8,064 GiB

2.3.1. Node Architecture

CPU Architecture

Intel Xeon Platinum 8480+ (Sapphire Rapids)

CPUs/Node

2

CPU Cores/Node

112 (56 per socket)

Memory/Node

504 GiB DDR5

GPUs/Node

4x NVIDIA H100

GPU Memory/Node

320 GiB (80 GiB per GPU)

Matrix Hardware Overview