8.3. Runtime Instrumentation Testing - N1SDP

For this test we used the N1 System Development Platform (N1SDP), which contains an SoC consisting of two dual-core Arm N1 clusters. Testing was done after the rc0 tag.

Please see the Runtime Instrumentation Testing Methodology page for more details. The tests were ran using the tf-psci-lava-instr/n1sdp-runtime-instrumentation,n1sdp-runtime-instrumentation:n1sdp-fip.tftf-firmware configuration in CI.

8.3.1. Results

8.3.1.1. CPU_SUSPEND to deepest power level

CPU_SUSPEND latencies (ns) to deepest power level in parallel (current release)

Cluster	Core	Powerdown	Wakeup	Cache Flush
0	0	1980 (-41.42%)	9420 (-24.52%)	400 (+17.65%)
1	0	2420 (-36.98%)	10460 (-41.69%)	180 (-18.18%)
0	0	3240 (+26.56%)	15060 (-12.14%)	140 (-12.50%)
1	0	2120 (-30.26%)	17020 (+70.88%)	140

CPU_SUSPEND latencies (ns) to deepest power level in parallel (previous release)

Cluster	Core	Powerdown	Wakeup	Cache Flush
0	0	3380 (+1.81%)	12480 (-7.00%)	340 (+21.43%)
0	1	3840 (-4.48%)	17940 (-1.10%)	220 (-21.43%)
1	0	2560 (-5.19%)	17140 (-1.38%)	160 (-38.46%)
1	1	3040 (+2.70%)	9960 (-8.29%)	140 (-46.15%)

CPU_SUSPEND latencies (ns) to deepest power level in serial (current release)

Cluster	Core	Powerdown	Wakeup	Cache Flush
0	0	1620	9340 (-7.89%)	280 (-6.67%)
1	0	1580 (-20.20%)	9060 (-13.22%)	140 (-50.00%)
0	0	1880 (+14.63%)	9580 (-7.35%)	160
1	0	1740 (-23.68%)	9640 (-8.71%)	140 (-76.67%)

CPU_SUSPEND latencies (ns) to deepest power level in serial (previous release)

Cluster	Core	Powerdown	Wakeup	Cache Flush
0	0	1640 (+1.23%)	10140	300
0	1	1980 (+6.45%)	10440 (-1.69%)	280
1	0	1640 (-8.89%)	10340 (+1.77%)	160 (-50.00%)
1	1	2280 (+5.56%)	10560	600 (+20.00%)

8.3.1.2. CPU_SUSPEND to power level 0

CPU_SUSPEND latencies (ns) to power level 0 in parallel (current release)

Cluster	Core	Powerdown	Wakeup	Cache Flush
0	0	1440 (+4.35%)	10460	240
1	0	1780 (+7.23%)	13900 (-12.47%)	140
0	0	1740 (+6.10%)	14520 (-4.47%)	140
1	0	1700 (-4.49%)	9280 (-6.07%)	140

CPU_SUSPEND latencies (ns) to power level 0 in parallel (previous release)

Cluster	Core	Powerdown	Wakeup	Cache Flush
0	0	1380 (+72.50%)	10600 (-11.52%)	240 (-7.69%)
0	1	1660 (+7.79%)	15880 (-3.41%)	140 (-46.15%)
1	0	1640 (+1.23%)	15200 (-5.59%)	140 (-53.33%)
1	1	1780 (-1.11%)	9880 (-6.26%)	140 (-30.00%)

CPU_SUSPEND latencies (ns) to power level 0 in serial (current release)

Cluster	Core	Powerdown	Wakeup	Cache Flush
0	0	1440 (-7.69%)	9220 (-4.75%)	280 (-6.67%)
1	0	1580 (-16.84%)	9040 (-12.91%)	180 (-35.71%)
0	0	1680 (-4.55%)	9480 (-7.24%)	140 (-58.82%)
1	0	1820 (+3.41%)	9620 (-8.38%)	260 (+85.71%)

CPU_SUSPEND latencies (ns) to power level 0 in serial (previous release)

Cluster	Core	Powerdown	Wakeup	Cache Flush
0	0	1560 (+8.33%)	9680 (-2.22%)	300
0	1	1900 (+9.20%)	10380	280
1	0	1760 (-2.22%)	10220 (+1.79%)	340
1	1	1760 (-10.20%)	10500	140 (-68.18%)

8.3.1.3. CPU_OFF on all non-lead CPUs

CPU_OFF on all non-lead CPUs in sequence then, CPU_SUSPEND on the lead core to the deepest power level.

CPU_OFF latencies (ns) on all non-lead CPUs (current release)

Cluster	Core	Powerdown	Wakeup	Cache Flush
0	0	1660 (-2.35%)	9320 (-6.99%)	300
1	0	13660 (-5.40%)	15980 (-6.77%)	180 (-50.00%)
0	0	13940	16720	140 (-46.15%)
1	0	14220	17080 (-2.95%)	140 (-53.33%)

CPU_OFF latencies (ns) on all non-lead CPUs (previous release)

Cluster	Core	Powerdown	Wakeup	Cache Flush
0	0	1700 (+3.66%)	10020 (-2.15%)	300
0	1	14440 (+3.74%)	17140 (-3.16%)	360 (+20.00%)
1	0	13680 (+1.03%)	16680	260 (-13.33%)
1	1	14260	17600 (-3.72%)	300

8.3.1.4. CPU_VERSION in parallel

CPU_VERSION latency (ns) in parallel on all cores (current release)

Cluster	Core	Latency
0	0	120 (-14.29%)
1	0	240
0	0	340 (+54.55%)
1	0	280 (+7.69%)

CPU_VERSION latency (ns) in parallel on all cores (previous release)

Cluster	Core	Latency
0	0	140 (+16.67%)
0	1	240 (+20.00%)
1	0	220 (+10.00%)
1	1	260 (+8.33%)

---

Copyright (c) 2023-2026, Arm Limited. All rights reserved.