The hardware resource utilizations are listed in the following table. Different tool versions may result slightly different resource.
Kernel | BRAM | URAM | DSP | FF | LUT | Frequency(MHz) |
kernel_louvain_0 | 150 | 208 | 118 | 169109 | 147344 | 214.1 |
kernel_louvain_1 | 150 | 208 | 118 | 169109 | 147344 | 214.1 |
Kernel | BRAM | URAM | DSP | FF | LUT | Frequency(MHz) |
kernel_louvain | 461 | 208 | 115 | 163521 | 130668 | 188.3 |
Table 3 Louvain FPGA acceleration benchmark by L2
Note
- 3 FPGA versions of kernel are used: * Normal kernel : 18.1 X * Prune kernel : 33.1 X without modularity loss VS. parallel Louvain on CPU * Opt kernel in u55c : 65.6 X without modularity loss by just 1cu VS. parallel Louvain on CPU * the Latency time = CPU time + FPGA time
- TigerGraph running on platform with Intel(R) Xeon(R) CPU E5-2640 v3 @2.60GHz, cache(20480 KB), cores(8).
- Parallel Louvain running on Intel(R) Xeon(R) CPU E5-2690 v4 @ 2.60GHz, cache(35840 KB), cores(14).
- time unit: second.
- FPGA platorm is Alveo u50 for Prune kernel, Alveo u55c for 2cu opt kernel