1. vitis_bd_measure

1.1 REPO TARGET

This repo is a part from Xilinx Vitis Tutorial. The goal of this repo is to show the bandwidth we can get with different transaction size.

1.2 Compile Step

Go to <ROOT>/vitis_bd_measure/au50 and execute build.sh. You can get results as below.

Open '''makefile_us_alveo.mk''', you can change the txSize and dsize. '''dsize''' represents the size of the data we will transmit between host and fabric.

Table 1: Throughput Measurement for vadd

krnl_loop	txSize	dsize (MB)	Throughput (GB/s)
1	64	16	1.24201
1	64	64	2.27867
1	64	256	7.7591
1	64	1024	11.2187
10	64	16	3.166
10	64	64	3.934
10	64	256	12.3211
10	64	1024	12.9559

2. ydma_bd_measure

2.1 Compile Step

Go to <ROOT>/ydma_bd_measure/au50 and execute build.sh.

2.2 Bandwidth Measurement

The throughput for ydma kernel. You can change the '''SIZE_MB''' macro to in host.cpp for different transmission size.

Table 2: Throughput Measurement for ydma

SIZE_MB (MB)	Throughput (GB/s)
1	3.350
10	5.708
16	5.915
64	6.303
256	10.093
512	10.173
1024	10.204

2.3 HiPR Bandwidth Measurement

The throughput for ydma kernel. You can change the '''SIZE_MB''' macro to in host.cpp for different transmission size.

Table 2: Throughput Measurement for ydma

SIZE_MB (MB)	Throughput (GB/s)
1	3.158
10	5.580
16	5.940
64	6.292
256	10.112
512	10.173
1024	10.180

3. increment_bd_measure

3.1 Compile Step

Go to <ROOT>/increment_bd_measure/au50 and execute build.sh.

3.2 Raw Bandwidth Measurement

The throughput for ydma kernel. You can change the '''SIZE_MB''' macro to in typedefs.h for different transmission size.

Table 2: Throughput Measurement for ydma

SIZE_MB (MB)	Throughput (GB/s)
1	0.913
10
16
64
256
512	1.417
1024

3.3 HiPR Bandwidth Measurement

The throughput for ydma kernel. You can change the '''SIZE_MB''' macro to in host.cpp for different transmission size. We should use host.cpp similar to ydma case to make sure the throughput can reach 10GB/s.

Table 2: Throughput Measurement for ydma

SIZE_MB (MB)	Throughput (GB/s)
1	0.919
10	1.199
16	1.241
64	1.278
256	1.396
512	1.409
1024	1.412
2048	1.401