Compute Specifications#
Quest comprises seven login nodes that users connect to directly and 1231 compute nodes with a total of 80,398 cores used for scheduled jobs. These nodes include 100 GPU nodes and 22 high-memory nodes. Both the login and compute nodes are running the Red Hat Enterprise Linux 8.10 operating system. See Quest Resources for how to request access to compute resources on Quest via ColdFront.
Tip
Login nodes are where most users land when connecting to Quest via applications like ssh or FastX. They support management tasks like launching jobs and transferring files. Login nodes should not be used for heavy computing tasks.
Regular Compute Nodes#
A portion of Quest compute nodes are retired and replaced regularly with new nodes using current technology. This allows Quest to grow in response to researcher needs, but it means that Quest nodes have different architectures. Quest currently consists of the following architectures:
Quest 13 - Intel Emerald Rapids
Number of Nodes: 140 nodes with 17920 cores total, 128 cores per node
Processor: Intel(R) Xeon(R) Platinum 8592+ CPU @ 1.9GHz
Memory: Per node (Per Core) 512 GB (4 GB), Type: DDR5 5600 MHz
Interconnect: Infiniband HDR
Expected retirement: Spring 2030
Quest 12 - Intel Ice Lake
Number of Nodes: 214 nodes with 13696 cores total, 64 cores per node
Processor: Intel(R) Xeon(R) Gold 6338 CPU @ 2.0GHz
Memory: Per node (Per Core) 256 GB (4 GB), Type: DDR4 3200 MHz
Interconnect: Infiniband HDR
Expected retirement: Fall 2028
Quest 11 - Intel Ice Lake
Number of Nodes: 209 nodes with 13376 cores total, 64 cores per node
Processor: Intel(R) Xeon(R) Gold 6338 CPU @ 2.0GHz
Memory: Per node (Per Core) 256 GB (4 GB), Type: DDR4 3200 MHz
Interconnect: Infiniband HDR Compatible
Expected retirement: Fall 2027
Quest 10 - Intel Cascade Lake
Number of Nodes: 555 nodes with 28860 cores total, 52 cores per node
Processor: Intel(R) Xeon(R) Gold 6230 CPU @ 2.10GHz
Memory: Per node (Per Core) 192 GB (3.7 GB), Type: DDR4 2933MHz
Interconnect: Infiniband EDR
Expected retirement: Fall 2026
GPU Nodes#
Quest has a total of 308 GPU cards on 100 GPU nodes across General Access and Priority Access. For more information on how to run on a GPU, see GPUs on Quest.
GPU resources available to General Access users:
Quest 13 - Intel Emerald Rapids with NVIDIA H100 GPUs
Number of Nodes: 24 nodes with 1536 cores total, 64 cores per node
Processor: Intel(R) Xeon(R) Platinum 8562Y+ CPU @ 2.8GHz (Emerald Rapids)
Memory: Per node (Per Core) 1TB (16 GB), Type: DDR5 5600 MHz
GPU cards per node: 4 x 80GB NVIDIA H100 (Connected with SXM5 and HBM3)
Interconnect: Infiniband NDR
Expected retirement: Fall 2029
Quest 12 - Intel Ice Lake with NVIDIA A100 GPUs
Number of Nodes: 18 nodes with 1152 cores total, 64 cores per node
Processor: Intel(R) Xeon(R) Gold 6338 CPU @ 2.0GHz (Ice Lake)
Memory: Per node (Per Core) 512 GB (8 GB), Type: DDR4 3200 MHz
GPU cards per node: 4 x 80GB NVIDIA A100 (Connected with SXM4 and HBM2)
Interconnect: Infiniband HDR
Expected retirement: Fall 2028
Quest 10 - Intel Cascade Lake with NVIDIA A100 GPUs
Number of Nodes: 16 nodes with 832 cores total, 52 cores per node
Processor: Intel(R) Xeon(R) Gold 6230 CPU @ 2.10GHz (Cascade Lake)
Memory: Per node (Per Core) 192 GB (3.7 GB), Type: DDR4 2933 MHz
GPU cards per node: 2 x 40GB NVIDIA A100 (Connected with PCIe)
Expected retirement: Fall 2026
High-Memory Nodes#
Quest has a total of 22 high-memory nodes with 500 GB to 2 TB of memory that can be used for batch jobs. This includes four nodes with 1.5 TB of memory that support General Access users; the remaining nodes support Priority Access users. For more information on how to run a job on a high-memory node, see Quest Partitions.
Batch Job Limits#
A significant amount of computing is available to everyone through the General Access proposal process. Currently, there are 535 regular nodes, 58 GPU nodes, and 4 high-memory nodes available exclusively for use in General Access. Furthermore, the General Access jobs can run on the majority of Priority Access Quest nodes for up to 4 hours. Researchers using a General Access compute resource can request appropriate partitions depending on their computational needs. For instance, short/normal/long partitions can be used to access the regular nodes. The short queue has access to the majority of Quest nodes and all regular node architectures.