Profiling - 2023.2 English

The kernel is built by Vivado tools and benchmard in U250 FPGA card at 275MHz. The hardware resource utilization and benchmark performance are listed in the table below.

Table 110 Table 1 Hardware resources
Name	LUT	BRAM	URAM	DSP
Platform	104112	165	0	4
scc_kernel	164311	523.5	110	6
Total	268423 (16%)	688.5 (26%)	110 (9%)	10 (0%)

Table 111 Table 2 Comparison between spark on CPU and FPGA
Datasets	Vertex	Edges	Number of SCC	Iteration number in Spark	FPGA Time (u250)	Spark (4 threads)		Spark (8 threads)		Spark (16 threads)		Spark (32 threads)
Datasets	Vertex	Edges	Number of SCC	Iteration number in Spark	FPGA Time (u250)	Spark Time	Speed up	Spark Time	Speed up	Spark Time	Speed up	Spark Time	Speed up
cit-Patents	3774768	16518948	3774768	6	20711	52137	2.52	60517	2.92	51390	2.48	39939	1.93
hollywood	1139905	57515616	1139905	6	9780	75681	7.74	45935	4.70	39595	4.05	29665	3.03
soc-LiveJournal1	4847571	68993773	971232	6	39952	424444	10.62	304755	7.63	244916	6.13	231465	5.79
ljournal-2008	5363260	79023142	1119171	16	34840	540199	15.51	458633	13.16	378304	10.86	402120	11.54
GEOMEAN					23043	173431	7.53X	140397	6.09X	117178	5.09X	102476	4.45X

Note

1. Spark running on platform with Intel(R) Xeon(R) CPU E5-2690 v4 @2.600GHz, 56 Threads (2 Sockets, 14 Core(s) per socket, 2 Thread(s) per core)

2. Time unit: ms.