Buy-In Period: Shared HPC Purchase Round Opportunity 2023

The Shared Research Computing Policy Advisory Committee (SRCPAC) is happy to offer the Columbia research community the option to buy into the next shared high performance computing cluster.  

TIMING

Please fill out our shared HPC order form between Monday, May 22, 2023 and Friday, June 16, 2023. All orders will require a chartstring account code. Please ensure that your chartstring is properly budgeted and will be valid until the end of the calendar year. We plan for the equipment to go live in late Fall 2023.

​CONTACT

For any questions or to schedule an information session please contact ​John Villa, Manager of CUIT High Performance Computing, at [email protected]​.

​INFORMATION 

The number of servers purchased by a research group defines a share of the total system (expressed as a percentage) that will ​determine job priority. System policies that determine job priority are approved by a faculty operations committee consisting of representatives of ​the ​participating research groups and are implemented by CUIT staff.

​PURCHASE INSTRUCTIONS

Please fill out our shared HPC order form. You may purchase one or more servers. A server will have a 5-year lifespan, and the purchase price will include racks, network switches, job scheduling software, user support, and other infrastructure-related costs. A research group can consist of one or more users.  

​PURCHASE OPTIONS 

These are current estimates. Final prices will be determined when we receive all orders. Note that this standard server has significantly more cores and memory than the last generation (cores 80 vs 32, memory 512 vs 192). We believe these prices will hold, but they may increase. 

We have already been informed to expect delivery delays on all GPU nodes.

• Standard Server ($14,105): Dual Xeon Platinum 8640Y+ processors (2 GHz, 40 cores each, 80 threads per server), 512 GB Memory, 480 GB SSD drive, 100 GB+ HSI
• High Memory Server ($16,892): Dual Xeon Platinum 8640Y+ processors (2 GHz, 40 cores each, 80 threads per server), 1 TB Memory, 480 GB SSD drive, 100 GB+ HSI
• GPU Server with 2 x H100 ($43,680): Dual Xeon Platinum 8640Y+ processors (2 GHz, 40 cores each, 80 threads per server), 512 GB Memory, 480 GB SSD drive, 100 GB+ HSI 2 x NVIDIA Hopper H100, PCIe, 350W, 80GB
• GPU Server with 2 x A100 ($29,774): Dual Xeon Platinum 8640Y+ processors (2 GHz, 40 cores each, 80 threads per server), 512 GB Memory, 480 GB SSD drive, 100 GB+ HSI,  2x  NVIDIA Ampere A100, PCIe, 300W, 80GB 
• GPU Server with 2 x A40 ($21,340): Dual Xeon Platinum 8640Y+ processors (2 GHz, 40 cores each, 80 threads per server), 512 GB Memory, 480 GB SSD drive, 100 GB+ HSI, 2 x NVIDIA Ampere A40, PCIe, 300W, 48GB
• Scratch Storage (~$350 per Terabyte): Dell’s PixStor storage system uses a GPFS-based, high-performance, scalable parallel file system.

FAQs

Q1:  Is the purchase of a server considered "major equipment"?

A1:  The Office of the Controller indicates that any purchase in the shared resource is considered "major equipment". It means that Facilities and Administrative costs (F&A) are not charged.

Q2: I’m not ready to commit to purchase. When will be my next chance?

A2: Historically, SRCPAC offers a chance to purchase each year. You also can purchase a lower-priority rental share at any time.

Q3: A new recruit is interested in shared HPC, but won’t join Columbia until after July 1. Is there anything we can do?

A3: Contact Victoria Hamilton (212-854-1343 or [email protected]) to discuss options.

Shared Research Computing Policy Advisory Committee

Columbia University in the City of New York

www.columbia.edu/srcpac