The City University of New York Campus of

The City University of New York Campus of The City College of New York of The City University of New York 9/16/2021 1

CUNY-Background • 19 th Century – City College of New York • Established in 1847 as the “Free Academy” to provide tuition-free college education to all, based solely on merit, including the economically disadvantaged and those precluded from attending the leading universities of the day because of ethnicity or gender. – Hunter College • • Established in 1870 as a tuition-free teacher-training school, based solely on merit, for young women. • Incorporated an elementary and high school for gifted children. Twentieth Century – Expansion – “City University of New York” • Established in 1961 and incorporates the pre-existing City public colleges. • 23 Institutions. – End of free tuition as a result of NYC’s 1975 financial crisis (currently $4, 000 per year at the senior colleges, $2, 800 at the community colleges). – Tuition increases to $4, 600 per year in 2009 9/16/2021 2

CUNY-Background • Largest urban university in the United States. • Third largest university system in the United States. – 231, 000 degree credit students. – 230, 000 adult, continuing and professional education students. – Baruch College - largest School of Business in the U. S. – College Now • Academic enrichment program for 32, 500 high school students. • Offered at CUNY campuses and more than 300 high schools. – University Teacher Academy • Free tuition for highly motivated mathematics and science majors who commit to teach in New York City. – Extensive Science, Technology, Engineering, and Mathematics and Louis Stokes Alliances for Minority Participation Programs. 9/16/2021 3

CUNY-Demographics • • • Bernard Baruch, Financier Senator Barbara Boxer Sidney 'Paddy' Chayefsky, Author Alan M. Dershowitz, Lawyer Felix Frankfurter, US Supreme Court Justice Andy Grove, co-founder and former CEO, Intel Corp. Irving Howe, Literature Robert Kahn, Co-developer of TCP/IP General Colin Powell, former Commander, Joint Chiefs of Staff and Secretary of State Dr. Jonas Salk, Polio Vaccine Bernard L. Schwartz, former CEO, Loran Upton Sinclair, Author 9/16/2021 • More CEOs and top executives in America graduated with first degrees from CUNY than any other degree-granting institution in the nation (Standard & Poors). • 12 Nobel prize winners • Many Pulitzer Prize Winners 4

Decade of Science • • • Initiated in 2005 Enhance CUNY’s position as a research university. Hire 1, 200 new faculty. Upgrade research facilities. Cyber Infrastructure/High Performance Computing Initiative. 9/16/2021 5

CUNY HPC Center • • • College of Staten Island - 204 acre park-like campus Free cruise on a 24 x 7 basis to Wall Street 4, 500 sq. ft computer room $3 million allocated for facility renovation – 2. 25 MW – Chilled water – Raised floor $6. 5 million allocated for design of a 90, 000 square foot Computational Science Facility – 2014 -15 planned completion – Expected to cost > $75 million Campus of the College of Staten island 9/16/2021 6

Why GPUs? • Processor clock speed/computational performance CFP 2006 • Woodcrest • Clovertown • Barcelona • Harpertown • Nehalem 9/16/2021 year 2006 2008 2009 clock 2. 83 GHz 2. 66 GHz 2. 30 Ghz 3. 16 GHz 2. 93 GHz 7 cfp 2006 base 13 14 17 24 36

Why GPUs-CUNY Perspective? • User Interest – “Older faculty” in physics and mathematics remember vectors • “I did that 20 years ago, I can do that” – Newer faculty have used CUDA and GP/GPUs on laptops and desktops • “Let me at it” – Faculty in the social sciences and biology need more computational power, but are not interested in complex programming exercises • Matlab (parallel Matlab and Parallel Matlab with CUDA extensions) • Mathematica (ditto) • R 9/16/2021 8

Why GPUs? • Comparison of X 86 -64 and NVidia S 1070 (preliminary) – X 86 -64 = Harpertown running at 3. 16 GHz – One Tesla processors (240 cores) out of the four on the Nvidia S 1070 1 U board – CFD calculation - full Navier-Stokes – Used one X 86 -64 core and one Tesla processor – 5. 8 times speed-up • Not fully optimized • Goal is a minimum of 10 X over a single core 9/16/2021 9

Why GPUs-User Perspective? • Scaling, for most applications, is not forever • CUDA is easier than MPI – Of course, we want to do both • Scaling with dense computation adds complexity, but for the right applications provides significant performance advantages 9/16/2021 10

Why GPUs-System Management Perspective? • Scaling out – Adds dramatically to system management and interconnect network complexity • GPUs – Does not add to cluster OS adminstration/management complexity – Does not add to interconnect complexity – Plug-in and play – But, scheduling cores and Teslas adds complexity 9/16/2021 11

HPC Infrastructure (Proposed Addition) • Configuration – 100 teraflops peak performance – 48 Intel X 86 -64 (Nehalem Core i 7 -960) nodes – 24 NVidia S 1070 general purpose GPUs – Heterogeneous nodes • 96 sockets of Nehalem • 96 Tesla boards • PCIex-16 interconnect • Efficient support for scalable dense computational algorithms – Monte Carlo • Supported 3 rd party applications – Parallel Matlab – Star-P – Parallel Mathematica – Others 9/16/2021 12

Economics • Proposed Configuration – One Tesla per X 86 -64 socket • CFD application was not fully optimized – Have not worked on using all the cores on the X 86 socket in this example • Given the above caveats, for this application, at its current state and extrapolating: – Adding one Tesla equates to adding 1. 5 to 2. 5 X 86 sockets – Adding 96 Teslas to a 96 socket cluster only adds 25% to the cost of the cluster • No interconnect impact • No OS impact 9/16/2021 13

Futures • Tighter integration between vectors/GPU and X 86 architectures – More 64 -bit floating point support in the GPU – Higher speed interconnects • Better, unified software support – Simplified computational capability 9/16/2021 14