04SAAHPCpresentationScalableSystolicArray

Finance molecular dynamics collaboration with boston

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: FPGA vs. fast CPU core 14 n Preliminary results (speed, energy) very promising q NeedlemanWunsch Performance Evaluation on NovoG NW 900 800 700 850 PEs/FPGA Baseline: 192225, length 850 Sequence Comparisons Software Runtime: 11,026 CPU hours on 2.4GHz Opteron # FPGAs Runtime (sec) 47,616 12,014 503 391 270 Speedup 833 3,304 78,914 101,518 147,013 Average Sequence Length (Nucleotides) 1 4 96 128 192 (est.) Speedup 600 500 400 300 200 100 450 0 1K 4K 16K 256K 1M 250 4M 16M 50 32M 850 650 Scaling performance with varying number of FPGAs under optimal input conditions Number of Sequence Comparisons Performance of single FPGA under varying input conditions 15 15 SW/ND Performance Evaluation on NovoG Database Length (Nucleotides) 800 700 600 500 400 300 200 100 0 50 100 150 200 3500 3000 2500 2000 450 400 350 300 250 500 0 1K 4K 16K 64K 256K 1M 4M 200 150 100 50 16M 64M 4M 256K 250 300 350 400 16K 450 500 550 600 1K 650 1500 1000 Average Sequence Length (Nucleotides) Number of Sequence Comparisons Baseline: Human X Chromosome v 19200, length 650 Seqs Software Runtime: 5,481 CPU hours on 2.4GHz Opteron # FPGAs 1 4 96 128 192 (est.) Runtime (sec) 23,846 5,966 250 188 127 Speedup 827 3,307 78,926 104,955 155,366 Baseline: 192224, length 450 Distance Calculations Software Runtime: 11,673 CPU hours on 2.4GHz Opteron # FPGAs 1 4 96 128 192 (est.) Runtime (sec) 13,522 3,429 144 118 77 Speedup 3,108 12,255 291,825 356,125 545,751 16 16 Avg. Sequence Length (Nucleotides) SW Speedup 650 PEs/FPGA ND 450 PEs/FPGA Speedup Conclusions n n n Developed in-stream control approach for systolic array to maximize hardware acceleration Capitalized on reconfigurable computing at scale to achieve sustainable supercomputing performance Demonstrated performance and sustainability results on Novo-G: Baseline: Human X Chromosome v 19200, length 650 Seqs Software Runtime: 5,481 CPU hours on 2.4GHz Opteron Baseline: 192224, length 450 Distance Calculations Software Runtime: 11,673 CPU hours on 2.4GHz Opteron Baseline: 192225, length 850 Sequence Comparisons Software Runtime: 11,026 CPU h...
View Full Document

This note was uploaded on 03/27/2012 for the course EEL 4930 taught by Professor Staff during the Spring '08 term at University of Florida.

Ask a homework question - tutors are online