CF-204100_QuestionsAndAnswers-Set2_2020-02-28.pdf
PDF 107 KB Posted
- Attached to
- High Performance Supercomputer Cluster
- Solicitation number
- CF-204100
- Issued by
- Wisconsin
About this file
This is a Questions and Answers document (Set 2) for a High Performance Supercomputer Cluster procurement issued by the University of Wisconsin Eau Claire's Purchasing Services Department on behalf of the Chemistry and Materials Science and Engineering division. The document addresses technical specifications and clarifications regarding the proposed supercomputer system, including power distribution infrastructure, storage capabilities, system architecture, and acceptance testing procedures. Key clarifications indicate that existing Power Distribution Units (PDUs), specifically RACK PDU 2G 208V 8.6KW 0U 36C13 6C19 2520 METERED units, must be utilized and cannot be replaced by vendors. The procurement specifies that the head node and login node will be a single integrated system rather than separate components, and that data I/O patterns for the storage server will involve a mix of both large sequential and small random access requests.
Regarding acceptance testing and performance validation, the University indicated that measured peak performance results using LINPACK benchmarks (excluding GPU cards) and comparative performance data for VASP, Gaussian, and MedeA/VASPMedeA software applications will be required. However, the University clarified that detailed performance metrics, hardware specifications, runtime parameters, HPL.dat files, compiler specifications, and MPI distribution information for existing benchmark platforms are not currently available for vendor reference. This document represents the second set of responses to vendor inquiries and does not address pricing, contract terms, award timelines, or other commercial terms of the procurement.
View the file
Other files
On GovTribe
Work with this file
- Download the original file
- Contacts named in this file
- Similar government files
- Ask GovTribe AI about this file
Text version
High Performance Supercomputer Cluster
CF‐204100
Bid Questions
Q11: While an existing onsite cabinet will be used, can the vendor provide power distribution units
(PDUs), or are there existing PDUs already installed into the cabinet, which are to be used for the cluster?
A11: PDUs are already installed.
Q12: If there are existing units, can you tell us the manufacturer and model number, and whether using them is a requirement?
A12: RACK PDU 2G 208V 8.6KW 0U 36‐C13 6‐C19 2‐5‐20 METERED
Q13: If there are existing PDUs, and using them is not a requirement, can the vendor choose to use any? If the vendor can instead provide PDUs, can you indicate number and type of available power outlets available for the new cluster (voltage, phase, and amperage). If new power outlets can be installed, this will be useful to know, and will allow for some flexibility in specifying an optimal choice of PDU.
A13: Use of existing PDUs is a requirement.
Q14: For the storage server, can you tell us what range of data I/O patterns are expected? For example, will clients primarily be making large sequential read/write requests? Will there be significant of small random I/O? Or will there be some mix of both? We have found that in nearly all cases, there is a mix of I/O across this spectrum. However, if you can give us any further description of the data I/O access pattern, this would be useful information.
A14: There will be a mix of both.
Q15: Can you confirm that the system indicated as the head node and login node is in fact a single system, which will serve both functions as a head node and login node, and is not two separate systems, where one is the head node, and the other the login node?
A15: It is the same.
(CONTINUED…)
High Performance Supercomputer Cluster
CF‐204100
Bid Questions
Q16: For the acceptance testing, the RFQ reads:
"Acceptance testing will occur within ten (10) days after the item/service has been installed.
Acceptance testing will be performed by the University and will consist of verification of the specifications and performance requirements. The vendor shall include measured peak performance results (in FLOPS) from the latest LINPACK benchmarks. The LINPACK test shall exclude GPU cards. VASP shall run at speeds similar to the measured peak performance."
Can further information be provided on the details of existing performance benchmarks of
LINPACK, VASP, Gaussian, and MedeA/VASP‐MedeA?
For example, can performance metrics be released, along with the hardware specifications of the respective platforms on which they were run?
For LINPACK, it would be useful if runtime parameters could be provided, along with the HPL.dat file, compiler type used, along with compiler flags, such as optimization flags.
Along with this, it would also be useful to know the MPI distribution, and MPI version number with which LINPACK was run.
If separate single‐node and cluster‐wide LINPACK benchmark values can be provided for
LINPACK, this would be very helpful.
A16: This information is not currently available.
File details come from the government source that posted it. Updated .