Skip to content

Fujitsu FX1000 S1 M4 C48

Thomas Gruber edited this page Sep 28, 2021 · 5 revisions

System

  • Processor: Fujitsu A64FX FX1000
  • Base frequency: 2.2 GHz
  • Number of sockets: 1
  • Number of memory domains per socket: 4
  • Number of cores per socket: 48
  • Number of HWThreads per core: 1
  • MachineState output: NA

Tool chain

+----------+-----------------------------+
| Compiler | fcc (FCC)                   |
|----------|-----------------------------|
| Version  |  fcc (FCC) 4.4.0a 20210127  |
+----------+-----------------------------+

Optimizing flags: -Kfast -Kocl -Koptmsg=2 -Nlst=t -Kzfill -Kprefetch_line=6 -Kopenmp

Please note that runs were performed with huge pages (2MB).

Results

All results are in GB/s.

Summary results:

+--------------------------------------------------+
| Single core   |  	92.72 (SDaxpy)             |
| Memory domain |  	230.03 (Sum with 8 cores)  |
| Socket        |  	870.95 (Sum with 8 cores)  |
| Node          |  	870.95 (Sum with 8 cores)  |
+--------------------------------------------------+

Results for scaling within a memory domain:

#nt	Init	Sum	Copy	Update	Triad	Daxpy	STriad	SDaxpy
1	13.24	57.29	26.15	80.14	38.22	89.81	49.41	92.72
2	26.51	117.00	52.39	132.17	76.15	159.29	99.31	180.45
3	39.79	168.38	78.55	171.95	114.22	203.63	148.67	214.14
4	53.11	203.77	104.60	198.39	152.39	213.95	196.94	215.06
5	66.48	218.87	130.04	208.79	190.06	213.73	208.89	215.36
6	80.00	226.30	155.77	213.38	207.44	213.69	212.09	214.96
7	85.95	222.49	164.64	208.16	207.38	212.42	211.67	214.28
8	107.11	230.03	202.29	214.65	211.58	212.13	212.99	214.17
9	110.59	228.35	203.24	210.91	205.07	211.34	209.35	213.26
10	133.32	228.69	205.94	212.66	210.84	211.85	212.49	213.68
11	131.18	227.63	208.70	212.26	210.38	211.58	212.24	213.37
12	133.24	227.19	208.84	209.98	210.47	211.67	211.93	213.24

Results for scaling across memory domains. Shown are the results for the number of memory domains used (nm) with columns number of cores used per memory domain.

Init:

#nm	1	2	3	4
1	13.24	26.49	39.73	52.96
2	26.51	53.01	79.49	105.94
3	39.79	79.53	110.35	158.88
4	53.11	106.18	159.09	211.97
5	66.48	132.81	199.04	265.06
6	80.00	159.83	219.82	319.11
7	85.95	171.35	256.64	341.38
8	107.11	213.79	320.11	425.84
9	110.59	220.76	330.20	437.92
10	133.32	265.29	396.50	525.45
11	131.18	259.50	383.31	519.40
12	133.24	264.95	393.73	527.38

Sum:

#nm	1	2	3	4
1	57.29	114.50	170.48	226.70
2	117.00	231.74	344.43	453.07
3	168.38	333.02	418.70	646.18
4	203.77	400.74	592.29	779.26
5	218.87	430.32	633.73	828.78
6	226.30	443.33	627.61	865.72
7	222.49	435.47	645.35	847.42
8	230.03	451.53	665.44	870.95
9	228.35	449.76	665.32	864.43
10	228.69	448.83	661.70	864.71
11	227.63	446.43	659.02	859.17
12	227.19	445.79	655.94	864.56

Copy

#nm	1	2	3	4
1	26.15	52.40	78.60	104.78
2	52.39	104.74	157.08	209.39
3	78.55	157.00	215.78	313.31
4	104.60	209.10	312.30	416.26
5	130.04	259.82	389.30	518.41
6	155.77	311.05	423.25	619.02
7	164.64	329.27	492.46	655.47
8	202.29	402.55	599.54	797.99
9	203.24	403.31	601.76	796.74
10	205.94	411.32	616.34	817.48
11	208.70	415.97	622.32	827.97
12	208.84	416.56	615.35	817.13

Update

#nm	1	2	3	4
1	80.14	160.16	240.20	319.72
2	132.17	263.95	395.49	526.63
3	171.95	343.19	432.03	683.28
4	198.39	396.26	592.45	790.80
5	208.79	416.82	623.73	829.48
6	213.38	425.51	598.87	850.88
7	208.16	414.01	623.04	827.46
8	214.65	427.37	639.56	849.91
9	210.91	422.03	633.41	837.44
10	212.66	423.60	635.51	845.11
11	212.26	422.42	632.77	840.07
12	209.98	419.61	630.04	839.04

Triad

#nm	1	2	3	4
1	38.22	76.68	114.99	153.31
2	76.15	152.17	228.30	304.08
3	114.22	228.18	304.35	455.71
4	152.39	304.20	455.11	606.28
5	190.06	378.81	567.17	753.18
6	207.44	412.90	592.12	820.72
7	207.38	412.86	616.03	812.76
8	211.58	420.05	613.30	813.03
9	205.07	408.38	613.02	810.70
10	210.84	418.52	626.45	831.34
11	210.38	418.01	624.82	825.94
12	210.47	417.65	621.49	824.72

Scaling

Memory bandwidth scaling within one memory domain: Main memory bandwidth scaling plot

The following plots illustrate the the performance scaling over multiple memory domains using different number of cores per memory domain.

Memory bandwidth scaling across memory domains for init: Memory domain scaling plot

Memory bandwidth scaling across memory domains for sum Memory domain scaling plot

Memory bandwidth scaling across memory domains for copy Memory domain scaling plot

Memory bandwidth scaling across memory domains for Triad Memory domain scaling plot