BRAND_NAME_SPECIFICATION_-_Redacted.pdf
PDF 11 KB Posted
- Attached to
- Quad-GPU and Single-GPU Workstations
- Solicitation number
- N66604-17-Q-0831
View the file
Other files
| File | Type | Posted |
|---|---|---|
| Minimum_Requirements_Table.pdf |
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
v1.2_JUNE2016
BRAND NAME SPECIFICATION
Microsoft Windows /Nvidia GPU / Intel CPU
PR 1300608061
BACKGROUND:
Naval Undersea Warfare Center Division Newport Code 15 requires the purchase of three (3) quad-GPU high-performance computers and three (3) single-GPU high-performance computers with capabilities to perform analysis of large datasets of programs and projects, as well as leverage current frameworks for machine learning. Market research was conducted and determined Brand Name requirements for the graphics card and processor. The Brand Names are available from multiple sources. There is no procurement history for this purchase.
REQUIREMENT:
1. Nvidia GPU
2. Intel CPU
3. Microsoft Windows 10 Operating System
Brand Name Specification:
Microsoft Windows 10 Operating System:
DON computers require a Microsoft operating system for required Microsoft software to operate effectively which includes OS, Office Professional Plus, and Core CALs. These components are commonly bundled and known as "Microsoft Enhanced Desktop Bundle" or "Desktop Pro (DTP)." The Core CALs provide access to the following standard server applications: Exchange, Windows, SharePoint, and system Center. This acquisition must be conducted on a brand name basis in order to eliminate a disruption in DON operations and to avoid a substantial duplication of costs that could not reasonably be recovered through brand competition at this time.
nVIDIA Graphics Card:
All modern frameworks for machine learning, including Theano and Tensorflow (which the workstations in this purchase will utilize) rely on the CUDA parallel computing platform. CUDA requires an Nvidia graphics processing unit to run. Without using Nvidia hardware, the only viable alternative to CUDA is an AMD graphics processing unit with AMD’s CUDA-equivalent, OpenCL. However, OpenCL is not an acceptable alternative at this time, as it is years behind CUDA performance-wise and not well supported by existing development frameworks. Use of these existing frameworks is critical, as it would not be possible to internally develop functional equivalents from scratch in a timely or cost-effective manner.
Intel Processor:
CPU choice is largely limited to chips made by AMD and Intel, and of the two manufacturers, the latest Intel processors are significantly more effective when paired with a powerful GPU, as will be the case in these workstations. Specifically, the analysis frameworks that will be used in these workstations requires a robust platform comprised at a minimum equal to that of the Intel Core i7 in order to ensure the accurate and reliable delivery of results for a wide variety of Navy v1.2_JUNE2016 products necessary to ensure their quality and safety for the fleet. The Intel chipset is currently used and therefore compatible with our current configuration and will minimize investigation into issues resulting from different processors. If such issues arise, it is estimated the cost to the government would be estimated at and several months of project delays. For the quad-GPU systems in this purchase, an i7 (4 cores) is sufficient, but for the single-GPU systems whole performance is CPU-bound, a more cores are required, necessitating either multiple i7s or newer Xeon units.
Requiring Department:
Based on the above, I recommend this acquisition be conducted on a Brand Name Specification basis. I certify the data which form the basis of this justification are complete and accurate.
File details come from the government source that posted it. Updated .