性能基准

1. Workstation

1.1 设备信息

类别	说明	内存/缓存
CPU	Intel(R) Xeon(R) CPU E5-2678 v3 @ 2.50GHz	31.23 GB
	KERNEL 0, 1200.229MHZ	30720 KB
	KERNEL 1, 1200.154MHZ	30720 KB
	KERNEL 2, 1200.281MHZ	30720 KB
	KERNEL 3, 1199.975MHZ	30720 KB
	KERNEL 4, 1200.699MHZ	30720 KB
	KERNEL 5, 1200.039MHZ	30720 KB
	KERNEL 6, 1200.195MHZ	30720 KB
	KERNEL 7, 1201.029MHZ	30720 KB
	KERNEL 8, 1547.491MHZ	30720 KB
	KERNEL 9, 1200.374MHZ	30720 KB
	KERNEL 10, 1200.858MHZ	30720 KB
	KERNEL 11, 1201.412MHZ	30720 KB
	KERNEL 12, 1200.833MHZ	30720 KB
	KERNEL 13, 1204.285MHZ	30720 KB
	KERNEL 14, 1200.868MHZ	30720 KB
	KERNEL 15, 1205.363MHZ	30720 KB
	KERNEL 16, 1202.910MHZ	30720 KB
	KERNEL 17, 1204.690MHZ	30720 KB
	KERNEL 18, 1201.506MHZ	30720 KB
	KERNEL 19, 1203.445MHZ	30720 KB
	KERNEL 20, 1258.832MHZ	30720 KB
	KERNEL 21, 1201.113MHZ	30720 KB
	KERNEL 22, 1203.082MHZ	30720 KB
	KERNEL 23, 1200.645MHZ	30720 KB
GPU	GeForce RTX 2080	7.79 GB

1.2 2022-06-11

1.2.1 神经元规模性能测试

Benchmark for NeuronNumber, SynapseNumber / NeuronNumber = 1000
Number	steps	General(ms)		DelaySynapse(ms)		LIF(ms)
		CUDA	CPU	CUDA	CPU	CUDA	CPU
4000	1	1.15	9.11	1.66	10.58	1.25	1.78
8000	1	1.24	12.68	2.89	14.84	1.29	2.47
16000	1	1.66	23.41	5.74	28.26	1.77	5.91
20000	1	2.29	28.10	7.29	35.50	2.03	7.82
40000	1	4.65	52.95	15.19	71.61	4.70	15.52
80000	1	14.34	105.10	20.63	143.00	14.46	32.30
160000	1	26.51	204.03	42.83	274.95	25.52	63.58
200000	1	35.31	255.06	54.60	334.18	33.18	78.49
4000	5	4.60	44.24	6.69	51.96	4.38	8.10
8000	5	5.12	61.66	13.12	70.50	4.43	11.46
16000	5	7.67	115.87	27.66	140.17	7.41	28.89
20000	5	9.22	141.49	36.95	178.38	8.54	41.57
40000	5	21.45	267.94	74.18	345.23	22.63	76.84
80000	5	69.22	517.35	102.20	677.99	69.15	162.06
160000	5	134.75	966.40	204.18	1,379.89	135.07	311.59
200000	5	174.43	1,197.79	292.96	1,702.85	169.01	408.36
4000	10	8.39	83.30	13.60	99.64	8.51	16.16
8000	10	8.74	119.81	25.60	139.15	8.69	22.86
16000	10	14.32	232.75	56.09	278.62	14.41	59.99
20000	10	16.86	274.54	72.03	351.51	16.97	72.76
40000	10	43.40	525.67	151.69	700.34	43.62	156.70
80000	10	130.71	1,042.51	214.54	1,407.54	135.10	313.41
160000	10	264.54	1,993.39	432.59	2,736.78	267.78	639.89
200000	10	345.98	2,548.53	550.13	3,336.72	334.22	792.65
4000	20	16.11	165.19	27.64	195.08	15.42	32.31
8000	20	17.13	237.32	50.21	272.05	17.28	46.22
16000	20	27.35	458.58	107.61	551.11	32.12	116.88
20000	20	36.67	549.31	141.46	698.10	35.38	146.04
40000	20	86.85	1,064.62	289.89	1,387.56	91.54	307.53
80000	20	267.81	2,097.26	429.20	2,787.89	271.87	639.57
160000	20	510.71	3,841.73	839.92	5,425.19	536.73	1,259.22
200000	20	663.16	4,457.87	1,131.26	6,675.62	690.52	1,536.29

1.2.2 突触比例的性能测试

Benchmark for SynapseScaled, NeuronNumber = 8000
Scaled	steps	Backend(ms)
		CUDA	CPU 1T	CPU 2T	CPU 4T	CPU 8T
100	1	1.00	2.49	1.00	0.58	0.41
500	1	1.10	7.88	3.02	1.49	0.86
1000	1	1.26	12.82	4.97	2.48	1.88
100	5	3.28	11.40	4.34	2.43	1.47
500	5	3.82	36.41	14.19	6.67	3.80
1000	5	4.84	62.95	23.68	11.64	9.43
100	10	5.79	22.09	8.44	4.63	2.83
500	10	7.62	71.68	28.16	13.41	7.69
1000	10	9.14	121.27	47.34	23.02	15.74
100	20	11.44	44.49	16.68	9.05	5.38
500	20	14.58	141.26	55.91	26.90	15.10
1000	20	17.32	237.94	93.14	44.96	31.76
100	50	26.32	112.25	41.62	22.71	14.14
500	50	34.59	346.35	140.66	65.70	36.78
1000	50	39.81	594.24	231.40	111.50	71.70