Memory bandwith and latency measurements Tier-1#
Memory bandwidth and latencies for main memory, as well as latencies for L2
and LLC cache. Measurements have been performed using Intel’s Memory Latency
Checker (mlc and mlc_avx512).
skylake nodes Tier-1#
Intel(R) Memory Latency Checker - v3.5
Measuring idle latencies (in ns)...
Numa node
Numa node 0 1
0 79.8 131.4
1 130.6 77.4
Measuring Peak Injection Memory Bandwidths for the system
Bandwidths are in MB/sec (1 MB/sec = 1,000,000 Bytes/sec)
Using all the threads from each core if Hyper-threading is enabled
Using traffic with the following read-write ratios
ALL Reads : 221550.1
3:1 Reads-Writes : 191572.6
2:1 Reads-Writes : 188084.9
1:1 Reads-Writes : 181321.7
Stream-triad like: 168425.4
Measuring Memory Bandwidths between nodes within system
Bandwidths are in MB/sec (1 MB/sec = 1,000,000 Bytes/sec)
Using all the threads from each core if Hyper-threading is enabled
Using Read-only traffic type
Numa node
Numa node 0 1
0 111180.9 34369.0
1 34380.0 110908.4
Measuring Loaded Latencies for the system
Using all the threads from each core if Hyper-threading is enabled
Using Read-only traffic type
Inject Latency Bandwidth
Delay (ns) MB/sec
==========================
00000 147.22 221598.3
00002 150.03 221530.8
00008 146.98 221641.0
00015 146.20 221253.5
00050 134.65 214083.3
00100 109.19 148517.0
00200 95.20 95575.8
00300 91.28 68113.2
00400 89.16 53086.4
00500 87.85 42872.3
00700 85.61 31657.3
01000 83.95 22734.5
01300 83.71 17786.4
01700 83.31 13849.3
02500 84.12 9699.3
03500 82.46 7183.2
05000 82.13 5268.8
09000 81.33 3273.9
20000 79.21 1936.2
Measuring cache-to-cache transfer latency (in ns)...
Local Socket L2->L2 HIT latency 49.4
Local Socket L2->L2 HITM latency 49.7
Remote Socket L2->L2 HITM latency (data address homed in writer socket)
Reader Numa Node
Writer Numa Node 0 1
0 - 110.8
1 111.2 -
Remote Socket L2->L2 HITM latency (data address homed in reader socket)
Reader Numa Node
Writer Numa Node 0 1
0 - 177.1
1 178.1 -
broadwell nodes Tier-1#
Intel(R) Memory Latency Checker - v3.1a
Measuring idle latencies (in ns)...
Numa node
Numa node 0 1 2 3
0 73.8 151.1 185.7 196.0
1 141.1 76.5 184.6 194.2
2 185.3 195.0 73.7 150.9
3 184.8 194.6 141.1 76.4
Measuring Peak Memory Bandwidths for the system
Bandwidths are in MB/sec (1 MB/sec = 1,000,000 Bytes/sec)
Using all the threads from each core if Hyper-threading is enabled
Using traffic with the following read-write ratios
ALL Reads : 144906.6
3:1 Reads-Writes : 130347.1
2:1 Reads-Writes : 126626.7
1:1 Reads-Writes : 114846.3
Stream-triad like: 125606.5
Measuring Memory Bandwidths between nodes within system
Bandwidths are in MB/sec (1 MB/sec = 1,000,000 Bytes/sec)
Using all the threads from each core if Hyper-threading is enabled
Using Read-only traffic type
Numa node
Numa node 0 1 2 3
0 36481.7 18909.5 16039.3 15127.7
1 20194.2 36376.3 15812.5 14946.7
2 16090.9 15191.0 36507.0 18850.3
3 15775.5 14939.6 20151.5 36348.1
Measuring Loaded Latencies for the system
Using all the threads from each core if Hyper-threading is enabled
Using Read-only traffic type
Inject Latency Bandwidth
Delay (ns) MB/sec
==========================
00000 203.58 145200.4
00002 202.96 145260.1
00008 198.28 145314.3
00015 192.23 145305.6
00050 164.33 144868.6
00100 103.73 115424.2
00200 91.29 70055.4
00300 87.57 49582.2
00400 85.15 38364.5
00500 83.05 30718.6
00700 80.21 22785.3
01000 78.70 16523.2
01300 78.55 13031.8
01700 77.55 10250.7
02500 76.68 7307.4
03500 75.41 5504.1
05000 74.85 4127.0
09000 74.42 2685.3
20000 74.12 1687.0
Measuring cache-to-cache transfer latency (in ns)...
Local Socket L2->L2 HIT latency 34.4
Local Socket L2->L2 HITM latency 39.2
Remote Socket LLC->LLC HITM latency (data address homed in writer socket)
Reader Numa Node
Writer Numa Node 0 1 2 3
0 - 46.9 77.5 80.6
1 48.8 - 82.1 85.3
2 78.3 80.7 - 46.8
3 82.6 85.5 47.6 -
Remote Socket LLC->LLC HITM latency (data address homed in reader socket)
Reader Numa Node
Writer Numa Node 0 1 2 3
0 - 98.9 132.0 145.5
1 95.5 - 134.4 148.4
2 132.2 145.5 - 98.9
3 134.6 148.7 94.1 -