In my K1 vs. K3 comparison, the K3 surprisingly needed much longer than the K1 to run the shadow test suite (29 minutes 7 seconds vs. 9 minutes 24 seconds), despite its superior hardware. The most likely explanation was the network latency of the remote facility hosting that K3.

I have now repeated the tests on a K3 that sits on the same local network, removing that limitation. Everything else remained unchanged: Fedora 44 Omni, shadow-utils-4.19.0-7, and the same pytest-mh configuration.

Results

Metric	K1	K3 (remote)	K3 (local)
Network	Local	Remote facility	Local
Tests	129	147	154
Execution time	9m 24s	29m 7s	3m 44s

Analysis

With the network variable removed, the picture changes completely:

• The local K3 executed 154 tests in 3 minutes 44 seconds, roughly 8x faster per test than the remote run.
• It is also about 3x faster per test than the K1, which is what the hardware difference suggested all along.

Conclusion

These results confirm that network latency, not the SoC, was the bottleneck in the previous comparison. Under comparable network conditions, the K3 clearly outperforms the K1, and its test execution time is finally consistent with the architectural advantages described in the original post.

---