We want to migrate our program from 4 Orin units to 2 Thor units and need to evaluate CPU performance. Using the UnixBench tool, we found that in the same max power mode, Thor’s single-core CPU performance increased by 40%, and when testing with 12 cores, multi-core performance increased by 50%. This seems inconsistent with the official performance indicators. How are the official CPU performance benchmarks measured? What CPU performance testing tools are available on Thor that can better simulate the performance in real autonomous driving scenarios?
Using UnixBench to test Dhrystone 2, Double-Precision Whetstone, Pipe Throughput, Pipe-based Context Switching, and System Call Overhead.
Result is :
IN ORIN:
Benchmark Run: Fri Oct 31 2025 21:16:32 - 21:27:48
12 CPUs in system; running 1 parallel copy of tests
Dhrystone 2 using register variables 36509954.3 lps (10.0 s, 7 samples)
Double-Precision Whetstone 6291.2 MWIPS (9.9 s, 7 samples)
Pipe Throughput 1043127.6 lps (10.0 s, 7 samples)
Pipe-based Context Switching 169691.1 lps (10.0 s, 7 samples)
System Call Overhead 720476.5 lps (10.0 s, 7 samples)
System Benchmarks Partial Index BASELINE RESULT INDEX
Dhrystone 2 using register variables 116700.0 36509954.3 3128.5
Double-Precision Whetstone 55.0 6291.2 1143.8
Pipe Throughput 12440.0 1043127.6 838.5
Pipe-based Context Switching 4000.0 169691.1 424.2
System Call Overhead 15000.0 720476.5 480.3
========
System Benchmarks Index Score (Partial Only) 906.3
Benchmark Run: Fri Oct 31 2025 21:12:44 - 21:23:53
14 CPUs in system; running 1 parallel copy of tests
Dhrystone 2 using register variables 52616679.1 lps (10.0 s, 7 samples)
Double-Precision Whetstone 7658.9 MWIPS (9.9 s, 7 samples)
Pipe Throughput 2513828.5 lps (10.0 s, 7 samples)
Pipe-based Context Switching 88276.2 lps (10.0 s, 7 samples)
System Call Overhead 1786797.9 lps (10.0 s, 7 samples)
System Benchmarks Partial Index BASELINE RESULT INDEX
Dhrystone 2 using register variables 116700.0 52616679.1 4508.7
Double-Precision Whetstone 55.0 7658.9 1392.5
Pipe Throughput 12440.0 2513828.5 2020.8
Pipe-based Context Switching 4000.0 88276.2 220.7
System Call Overhead 15000.0 1786797.9 1191.2
========
System Benchmarks Index Score (Partial Only) 1272.4
IN THOR:
Benchmark Run: Fri Oct 31 2025 21:12:44 - 21:23:53
14 CPUs in system; running 1 parallel copy of tests
Dhrystone 2 using register variables 52616679.1 lps (10.0 s, 7 samples)
Double-Precision Whetstone 7658.9 MWIPS (9.9 s, 7 samples)
Pipe Throughput 2513828.5 lps (10.0 s, 7 samples)
Pipe-based Context Switching 88276.2 lps (10.0 s, 7 samples)
System Call Overhead 1786797.9 lps (10.0 s, 7 samples)
System Benchmarks Partial Index BASELINE RESULT INDEX
Dhrystone 2 using register variables 116700.0 52616679.1 4508.7
Double-Precision Whetstone 55.0 7658.9 1392.5
Pipe Throughput 12440.0 2513828.5 2020.8
Pipe-based Context Switching 4000.0 88276.2 220.7
System Call Overhead 15000.0 1786797.9 1191.2
========
System Benchmarks Index Score (Partial Only) 1272.4
Benchmark Run: Fri Oct 31 2025 21:23:53 - 21:35:03
14 CPUs in system; running 12 parallel copies of tests
Dhrystone 2 using register variables 625406403.7 lps (10.0 s, 7 samples)
Double-Precision Whetstone 91814.3 MWIPS (9.9 s, 7 samples)
Pipe Throughput 29932892.2 lps (10.0 s, 7 samples)
Pipe-based Context Switching 1298925.5 lps (10.0 s, 7 samples)
System Call Overhead 21495572.2 lps (10.0 s, 7 samples)
System Benchmarks Partial Index BASELINE RESULT INDEX
Dhrystone 2 using register variables 116700.0 625406403.7 53591.0
Double-Precision Whetstone 55.0 91814.3 16693.5
Pipe Throughput 12440.0 29932892.2 24061.8
Pipe-based Context Switching 4000.0 1298925.5 3247.3
System Call Overhead 15000.0 21495572.2 14330.4
========
System Benchmarks Index Score (Partial Only) 15854.4