Packaging architecture for a highly parallel multiprocessor system

• US 5,251,097 A
• Filed: 06/11/1990
• Issued: 10/05/1993
• Est. Priority Date: 06/11/1990
• Status: Expired due to Term

• Current Assignee: Silicon Graphics International Corporation (Hewlett-Packard Enterprise Company)
• Original Assignee: Supercomputer Systems Limited Partnership
• Inventors: Girling, Dennis F., Paffel, Douglas C., Massopust, Dan L., Heid, Lisa, Priest, Edward C., Lesmerises, Felix R., Sperry, Christopher J., Simmons, Frederick J., Chen, Steve S., Pautsch, Greg W., Rabska, Michael H.
• Primary Examiner(s): Reynolds, Bruce A.
• Assistant Examiner(s): SWITZER, MICHAEL
• Application Number:
  US07/536,395
• Time in Patent Office: 1,212 Days
• Field of Search: 361/384, 361/385, 361/386, 361/388, 361/393, 361/407, 361/413, 364/200, 395/325, 395/425, 307/475
• US Class Current: 361/679.53
• CPC Class Codes: G06F 13/409 Mechanical coupling back panels H05K7/1438

Abstract

The present invention includes methods and apparatus for creating a packaging architecture for a highly parallel multiprocessor system. The packaging architecture of the present invention can provide for distribution of power, cooling and interconnections at all levels of components in a highly parallel multiprocessor system, while maximizing the number of circuits per unit time within certain operational constraints of such a multiprocessor system.

Claims

1. An apparatus for packaging circuitry means for a computer processing system, said apparatus comprising:

a first substrate and a second substrate spaced apart from said first substrate, said first and second substrates each comprising a generally planar structure including;

a first substrate area having a plurality of circuitry elements and a plurality of backplane external interconnections operably connected to said first substrate area and means for providing a plurality of signal paths to and from said circuitry elements and to and from said backplane external interconnections; and

a second substrate area having a plurality of circuitry elements operably connected to said second substrate area and means for providing a plurality of signal paths to and from said circuitry elements; and

transformation board means intersecting said first and second substrates for providing a plurality of signal paths for connecting said signal paths between said first and second substrate areas and between said first and second substrates,

such that the intersection of said transformation board means with said first and second substrates defines a planar boundary between said first and second substrate area on each of said first and second substrates and the only connections to and from said circuitry elements on said second substrate areas are via said signal paths on said transformation board means to said signal paths and said backplane external interconnections on said first substrate areas.

2. A modular brick for packaging a portion of a plurality of circuitry means for a computer processing system, said brick comprising:

switching circuit means for controlling the routing of signals to and from said brick operably connected to a plurality of spaced apart first substrates;

functional circuit means for performing functional operations operably connected to a plurality of spaced apart second substrates, each of said second substrates being associated with one of said first substrates; and

transformation board means intersecting each of said plurality of first and second substrates for providing a plurality of signal paths to interconnect said switching circuit means and said functional circuit means.

3. The modular brick of claim 2 wherein said switching circuit means comprises an arbitration node and said functional circuit means comprises two or more scalar/vector processors.

4. The modular brick of claim 2 wherein said switching circuit means comprises an arbitration network and said functional circuit means comprises a memory section.

5. A packaging architecture for a multiprocessor cluster for a highly parallel computer processing system, said multiprocessor cluster comprising shared resource means for storing and retrieving data and control information, a plurality of processors capable of parallel processing, and one or more arbitration node means operably connected to said processors and said shared resource means for symmetrically interconnecting said processors with said shared resource means, said packaging architecture comprising:

a plurality of modular brick means for packaging a portion of electronic circuitry for said computer processing system, each brick means comprising;

switching circuit means for controlling the routing of signals to and from said brick means operably connected to a plurality of spaced apart first substrates;

functional circuit means for performing functional operations operably connected to a plurality of spaced apart second substrates, each of said second substrates being associated with one of said first substrates; and

transformation board means intersecting each of said plurality of first and second substrates for providing a plurality of signal paths to interconnect said switching circuit means and said functional circuit means; and

a plurality of columns, each column comprised of two or more of said brick means operably connected together.

6. The packaging architecture of claim 5 wherein said brick means in one or more specified columns contain said switching circuit means and said functional circuit means associated with said processors and said arbitration node means and said brick means in one or more different columns contain said switching circuit means and said functional circuit means associated with said shared resource means.

7. The packaging architecture of claim 6 wherein there are two processor columns associated with said processors and said arbitration node means and four memory columns associated with said shared resource means, each of said processor columns comprised of four brick means each having two arbitration node means and four processors and each of said memory columns comprised of six brick means.

8. The packaging architecture of claim 7 wherein said processor columns and said memory columns are vertically oriented and arranged together in a generally rectangular horizontal cross-section with one of said memory columns positioned at each comer of said rectangular cross-section and said processor columns positioned between said memory columns along parallel sides of said rectangular cross-section.

9. The packaging architecture of claim 7 wherein two or more of said multiprocessor clusters may be operably connected together, each multiprocessor cluster further comprising remote cluster adapter means operably connected to remote cluster access means in all other clusters for allowing said arbitration node means to access said shared resource means of all other clusters and for allowing all of said other remote cluster access means of all other clusters to access said shared resource means of this cluster, said packaging architecture further comprising at least one modular brick means in one of said processor columns containing said switching circuit means and said functional circuit means associated with said remote cluster access means.

10. The packaging architecture of claim 7 wherein at least one of said processor columns further comprises a modular brick means containing said switching circuit means and said functional circuit means associated with a scan path unit for collecting and processing diagnostic information provided by all of said other modular brick means in said multiprocessor cluster.

11. The packaging architecture of claim 7 further comprising means for providing power to said multiprocessor cluster located above and below said processor columns and said memory columns and having a horizontal crosssection similar to said rectangular cross-section.

12. The packaging architecture of claim 7 further comprising means for providing a liquid coolant to said processor columns and said memory columns.

13. The packaging architecture of claim 12 wherein said means for providing a liquid coolant includes a plurality of impingement plates interspaced within said modular brick means for impinging said liquid coolant on said switching circuit means and said functional circuit means.