Profiling and Benchmarks - 2024.1 English

The Dense Similarity Kernel is validated on AMD Alveo™ U50 board at 260MHz frequency. The hardware resource utilization and benchmark results are shown in the following tables.

Table 100 Table 1 Hardware resources
Name	LUT	Register	BRAM	URAM	DSP
denseSimilarityKernel (uint + float)	221256	329187	402	16	1273
denseSimilarityKernel (int)	134446	160671	402	16	807
denseSimilarityKernel (int + 2CU)	272521	333259	618	48	2364

Table 101 Table 2 Performance comparison of dense graph between TigerGraph on CPU and FPGA
Datasets	Vertex	Edges	Similarity Type	FPGA Time / ms	TigerGraph (32 core 512 GB)
					Time / ms	Speed up
Patients(1 GB)	1250000	200	Cosine	11.2	585.7	52.3

Note

Tigergraph running on platform with Intel(R) Xeon(R) CPU E5-2640 v3 @2.600GHz, 32 Threads (16 Core(s)).
The uint + float version and integer version have relatively similar performance.
Time unit: ms.