One popular family of FPGAs, the Virtex-7 series, is built by Xilinx. A Virtex-7 XC7VX690T FPGA contains

Question:

One popular family of FPGAs, the Virtex-7 series, is built by Xilinx. A Virtex-7 XC7VX690T FPGA contains 3,600 25x18-bit integer multiply-add "DSP slices." Consider building a TPU-style design on such an FPGA.

a. Using one 25 x 18 integer multiplier per systolic array cell, what's the largest matrix multiplication unit one could construct? Assume that the matrix multiplication unit must be square.

b. Suppose that you could build a rectangular, nonsquare matrix multiplica- tion unit. What implications would such a design have for hardware and soft- ware? (Hint: think about the vector length that software must handle.)

c. Many FPGA designs are lucky to reach 500 MHz operation. At that speed, calculate the peak 8-bit operations per second that such a device might achieve. How does that compare to the 3 T FLOPS of a K80 GPU?

d. Assume that you can make up the difference between 3600 and 4096 DSP slices using LUTs, but that doing so will reduce your clock rate to 350 MHz. Is this a worthwhile trade-off to make?

Fantastic news! We've Found the answer you've been seeking!

Step by Step Answer:

Related Book For  book-img-for-question

Computer Architecture A Quantitative Approach

ISBN: 9780128119051

6th Edition

Authors: John L. Hennessy, David A. Patterson

Question Posted: