2025-01-15: Temporary Hold on Storage and Compute Requests
As Wynton continues to expand, we are approaching the limits of space, cooling, and power capacity in the Byers Hall server room. In addition, as we work to move the administration of the cluster to the Academic Research Systems (ARS) Team, we are currently reprioritizing our workload with the support of Wynton faculty leadership to ensure that we continue to meet our most critical objectives. We understand that this may cause some delays and appreciate your patience and understanding during this period.
See also the Wynton-announcement email titled ‘Important Updates on Wynton Storage and Compute Requests’ sent to all users on 2024-11-25.
Example specifications and pricing as of August 2024
Please note: You cannot be charged for the hardware until it has been received and invoiced. Since lead times can be extensive, please plan as far ahead as you can.
Cluster compute nodes are purchased by Wynton in modules that have 4 “2U” compute nodes in a single chassis. If partial nodes are requested, contributions will be pooled with other requests towards a full system. Another option is to pay 180 USD/member.q slot.
This configuration comes with four compute modules, where each node includes:
CPU/node: 2 × Intel Xeon Gold 5520+ 2.2GHz 28-core processor
RAM/node: 256GiB DDR5-4800 ECC (16 × 16 GiB)
Storage/node: 1 × 1.92TB U.2 NVMe SSD, 1DWPD
Total price for a module with four (4) node: ~37,000 USD excluding taxes.
As of August 2024, lead time for compute nodes is approximate three months from order to delivery. Once the hardware is received, we will schedule time to configure hardware as time allows given priority projects. Unless you are purchasing a four-compute-node module, we will need to wait until we have enough requests to place an order for a whole unit. If we have capacity, we may be able to provision some shares in advance of the billing.
Wynton HPC has 61 GPU nodes with a total of 235 GPUs available to all users. Among these, 39 GPU nodes, with a total of 147 GPUs, were contributed by different research groups. GPU jobs are limited to 2 hours in length when run on GPUs not contributed by the running user’s lab. In contrast, contributors are not limited to 2-hour GPU jobs on nodes they contributed. There is also one GPU development node that is available to all users.
Single GPU cards cannot be purchased - only a full GPU node. However, just like partial compute nodes can be purchased, GPU purchase requests can also be pooled towards a full system.
There are three common configurations of GPU compute nodes available, which all share the same base setup, and differs only by type of GPU.
CPU: 1 × AMD EPYC 7543P 2.8GHz 32-core processor
RAM: 512GiB DDR4-3200 (8 × 64GiB)
Operating-system storage: 1 × 512GB SATA SSD
Storage: 1 × 960GB U.2 NVMe SSD
Then choose one of:
GPU: 4 × Nvidia A40 with 48GB GPU memory (~25,000 USD)
GPU: 4 × Nvidia L40 with 48GB GPU memory (~33,000 USD)
Note: A100s are no longer available
The prices listed are total prices in USD excluding taxes. Each price includes the base setup and the corresponding GPU configuration. Warranty is included. Taxes are calculated and added once hardware is received.
As of October 2023, lead time for GPUs is approximately 3 months from order to delivery. Once the hardware is received, we will schedule time to configure hardware as time allows given priority projects.
To purchase compute nodes or GPU nodes, please contact compute@wynton.ucsf.edu. Please include what you are considering contributing and your UCSF COA SpeedType or chartstring for billing, along with any questions.
If you are not ready to purchase but have technical questions while you explore options, please contact wynton-support@ucsf.edu.