I do not disagree with you at all, in an ideal benchmark everything should be the exact same, but that would be different from the reality of what the bulk of people will experience using their boards with their customised “official” distributions.
The origin benchmark in the first post was using “qemu” compile times, so that would be a constant.
As for the kernel, for the VF2 most people will be on the heavily patched official StarFive LTS (LongTerm Support) 5.15.0-starfive kernel (released 2021-10-31, Projected EOL 2026-10), for the Lichee Pi 4A it is currently limited to 5.10.113 kernel (released 2020-12-13, Projected EOL 2026-12) - as far as I know. I got the kernel versions from looking at the results from people running sbc-bench, strangely no one has run sbc-bench on a HiFive Unmatched and submitted the result. And the Raspberry Pi 4B has multiple entries for 1500MHz and 1800MHz and different kernel, but mostly 5.10 and 5.15. But if you go to download a new image file, the current version for the RPi uses the latest LTS kernel 6.1 (released 2022-12-11, Projected EOL 2026-12). Again it is odd that no one has yet submitted a benchmark for the RPi 4B to sbc-bench running the 6.1 kernel.
As for the kernel configs all being the same, that would move the comparison away from what most people would experience. It would be a better benchmark and far more valid. And it would be the exact same for the compilers, moving from the default for each to the same version would provide a better synthetic benchmark of the hardware, but it would be different than the user experience with each SBC using the configured defaults.
And yes, this is what people will experience using their boards though most people who build the kernel would understand that not building the same version and/or same config can lead to different build time. I’m not saying compiling the kernel is not relevant, but especially on the RISC-V architecture, just the compiler version can have huge impact, things are still quite fresh and each new release of a compiler do add more on the RV than it would on other more mature arch.
Anyway, I was just trying to find what could cause that difference in time
I tried to answer your questions as much as I know.
My LPi4A has 16GB RAM and 128 GB eMMC.
In my tests eMMC on LPi4A has excellent performance, actually comparable to NVME performance on VF2. The read performance using DD on large blocks is ~300 MB/sec. But I also tested compile entirely from RAM to eliminate that factor and it produced the same result as from eMMC.
No network activity.
I compiled exact same kernel code with exact same configuration, which was a kernel from VF2 git repository. I compiled that code on both boards. gcc version is the same on both boards: 13.1.0.
I’m not sure what is the base idle level or what is the System Controller.
About thermals, the LPi4A is running quite a bit hotter but I don’t think it’s throttling. I use the sink and fan that came with the board and as soon as you turn it on it’s running at 40C, after running for a while it runs at 50C at idle and during compile at ~60C. For VF2 I have ice tower type of cooler, which is probably overkill. It runs ~35C at idle and ~45C during compile.