DELTA is a dedicated, ACCESS-allocated resource designed by HPE and NCSA, delivering a highly capable GPU-focused compute environment for GPU and CPU workloads. Besides offering a mix of standard and reduced-precision GPU resources, DELTA offers GPU-dense nodes with both NVIDIA and AMD GPUs, with high-performance node-local SSD scratch filesystems and both standard Lustre and relaxed-POSIX parallel filesystems (docs pending) spanning the entire resource.
DELTA's standard CPU nodes are each powered by two 64-core AMD EPYC 7763 ("Milan") processors, with 256 GB of DDR4 memory. The DELTA GPU resource has four node types: one with 4 NVIDIA A100 GPUs (40 GB HBM2 RAM each) connected via NVLINK and 1 64-core AMD EPYC 7763 ("Milan") processor, the second with 4 NVIDIA A40 GPUs (48 GB GDDR6 RAM) connected via PCIe 4.0 and 1 64-core AMD EPYC 7763 ("Milan") processor, the third with 8 NVIDIA A100 GPUs in a dual socket AMD EPYC 7763 ("Milan") (128-cores per node) node with 2 TB of DDR4 RAM and NVLINK, and the fourth with 8 AMD MI100 GPUs (32GB HBM2 RAM each) in a dual socket AMD EPYC 7763 ("Milan") (128-cores per node) node with 2 TB of DDR4 RAM and PCIe 4.0.
Allocated CiDeR Resources
The Delta GPU resource comprises 4 different node configurations intended to support accelerated computation across a broad range of domains such as soft-matter physics, molecular dynamics, replica-exchange molecular dynamics, machine learning, deep learning, natural language processing, textual analysis, visualization, ray tracing, and accelerated analysis of very large in-memory datasets. Delta is designed to support the transition of applications from CPU-only to using the GPU or hybrid CPU-GPU models. Delta GPU resource capacity is predominately provided by 200 single-socket nodes, each configured with 1 AMD EPYC 7763 (“Milan”) processor with 64-cores/socket (64-cores/node) at 2.45GHz and 256GB of DDR4-3200 RAM. Half of these single-socket GPU nodes (100 nodes) are configured with 4 NVIDIA A100 GPUs with 40GB HBM2 RAM and NVLink (400 total A100 GPUs); the remaining half (100 nodes) are configured with 4 NVIDIA A40 GPUs with 48GB GDDR6 RAM and PCIe 4.0 (400 total A40 GPUs). Rounding out the GPU resource is 6 additional large-memory, “dense” GPU nodes, containing 8 GPUs each, in a dual-socket CPU configuration (128-cores per node) and 2TB of DDR4-3200 RAM but otherwise configured similarly to the single-socket GPU nodes. Within the “dense” GPU nodes, 5 nodes employ NVIDIA A100 GPUs (40 total A100 GPUs in “dense” configuration) and 1 node employs AMD MI100 GPUs (8 total MI100 GPUs) with 32GB HBM2 RAM. A 1.6TB, NVMe solid-state disk is available for use as local scratch space during job execution on each GPU node type. All Delta GPU compute nodes are interconnected to each other and to the Delta storage resource by a 100 Gb/sec HPE Slingshot network fabric.
A Delta GPU allocation grants access to all types of Delta GPU nodes (both GPUs and CPUs on those nodes). The Delta CPU allocation is needed to utilize the CPU-only nodes.
The Delta Storage resource provides storage allocations on the scratch file system for projects using the Delta CPU and Delta GPU resources. It delivers 4PB of capacity to projects on Delta and will be augmented by a later expansion of 3PB of flash capacity for high-speed, data-intensive workloads. Projects are provided with a 1TB allocation on scratch by default. Please do not request storage allocations smaller than 1TB.
The NCSA Delta CPU allocation type provides access to the Delta CPU-only nodes. The Delta CPU resource comprises 124 dual-socket compute nodes for general purpose computation across a broad range of domains able to benefit from the scalar and multi-core performance provided by the CPUs, such as appropriately scaled weather and climate, hydrodynamics, astrophysics, and engineering modeling and simulation, and other domains using algorithms not yet adapted for the GPU. Each Delta CPU node is configured with 2 AMD EPYC 7763 (“Milan”) processors with 64-cores/socket (128-cores/node) at 2.55GHz and 256GB of DDR4-3200 RAM. An 800GB, NVMe solid-state disk is available for use as local scratch space during job execution. All Delta CPU compute nodes are interconnected to each other and to the Delta storage resource by a 100 Gb/sec HPE Slingshot network fabric.